Soil heavy metal pollution risk zoning and grading method based on machine learning

Abstract

The invention provides a soil heavy metal pollution risk zoning and grading method based on machine learning. The method comprises the following steps: respectively acquiring heavy metal sample data and environmental covariant data in a research area as a corresponding initial sample data set of target heavy metal; and preprocessing each initial sample data set to obtain a corresponding research sample data set. And obtaining pollution risk indexes of the target heavy metals. Mixing the research sample data sets, and randomly dividing the research sample data sets into a training set and a test set; and establishing a heavy metal risk zoning and grading model based on a random forest model according to the classification labels, the training set and the test set. And according to the effective environmental influence variables corresponding to the unknown areas in the research area, identifying and partitioning the soil heavy metal pollution risk of the research area. According to the method, the soil heavy metal pollution risk is quantitatively, qualitatively and from multiple angles, and the soil heavy metal potential pollution risk area is scientifically divided.

Description

Technical Field

[0001] The present invention relates to the field of environmental pollution prediction, and in particular to a soil heavy metal pollution risk zoning and grading method based on machine learning. Background Art

[0002] The process of urban industrialization has greatly increased the input of heavy metal concentrations in the soil, resulting in serious heavy metal pollution in the regional soil. Identifying and scientifically dividing potential pollution risk areas is of great research significance in the context of the complex characteristics of soil pollution in my country. Based on the characteristics of the contaminated sites and combined with regional planning needs, a risk assessment of regional soil heavy metal contaminated sites is carried out, and the risk levels and regions of the plots are divided to achieve classified control and zoning management of each contaminated plot.

[0003] Although the quantitative analysis method takes into account various indicators and factors, it rarely considers the spatial connection and interaction between adjacent units, and the zoning results do not meet actual needs. The qualitative zoning method is not very accurate and has a small amount of calculation. It is suitable for areas that are difficult to quantify. It mainly relies on the subjective judgment of researchers and cannot support the entire comprehensive zoning process. It can only be used for preliminary zoning. There is currently a lack of systematic analysis methods for regional soil heavy metal pollution risk identification and potential pollution risk area identification. Summary of the invention

[0004] The main purpose of the present invention is to provide a soil heavy metal pollution risk zoning and grading method based on machine learning to solve the technical problem of lack of systematic analysis methods for regional soil heavy metal pollution risk identification and potential pollution risk area identification.

[0005] To achieve the above object, the present invention provides a soil heavy metal pollution risk zoning and grading method based on machine learning, comprising the following steps:

[0006] Obtain the heavy metal sample data and environmental covariance data in the study area respectively as the corresponding initial sample data set of the target heavy metal. The heavy metal sample data at least includes the target heavy metal content data in the soil and the sampling point location information. The environmental covariance data includes soil physical and chemical properties, environmental impact factors, and socio-economic impact factors. The target heavy metals include two or more of Cd, Pb, Hg and As.

[0007] Each initial sample data set is preprocessed to obtain a corresponding research sample data set, wherein the preprocessing includes screening to obtain effective environmental impact variables.

[0008] The pollution risk index of each target heavy metal is obtained as the classification label of the heavy metal risk zoning and grading model.

[0009] After mixing the sample data sets of each study, they are randomly divided into training set and test set.

[0010] A heavy metal risk zoning and grading model is established according to the classification labels, the training set, the test set and based on a random forest model.

[0011] Based on the heavy metal risk zoning and grading model, the soil heavy metal pollution risks in the study area are identified and zoned according to the effective environmental impact variables corresponding to the unknown areas in the study area.

[0012] According to an embodiment of the present application, the soil physical and chemical properties include soil pH value and soil organic matter.

[0013] The environmental impact factors include soil type, land use type, vegetation cover, elevation, slope, slope aspect, atmospheric deposition, average wind speed, average precipitation, transportation network, and water system distribution.

[0014] The socio-economic influencing factors include GDP per capita, population density distribution and geographical location of heavy metal enterprises.

[0015] The environmental covariance data also includes the distances between the transportation network, water system distribution and the geographical location of heavy metal enterprises and the sampling points.

[0016] According to the implementation mode of the present application, the step of determining the distances between the transportation network, water system distribution and the geographical location of the heavy metal enterprise and the sampling point includes:

[0017] According to the fuzzy matching of heavy metal enterprise names based on administrative divisions, the geographical location information of heavy metal enterprises is obtained.

[0018] Convert the geographical location information of heavy metal enterprises into vector data based on administrative divisions.

[0019] Based on the vector data corresponding to the traffic network, water system distribution and heavy metal enterprises, the distances between the sampling points and the three are determined.

[0020] According to an embodiment of the present application, the step of preprocessing the initial sample data set to obtain the research sample data set includes:

[0021] The missing values ​​of the initial sample data set are interpolated to obtain a primary sample data set.

[0022] The features of different variables in the first sample data set are scaled to obtain the second sample data set.

[0023] The secondary sample data set is subjected to dimensionality reduction and independence test, and the sample data set corresponding to the effective environmental impact variables is determined as the research sample data set.

[0024] According to an embodiment of the present application, each sample data in the research sample data set further includes a first ionization energy and a regional background value.

[0025] According to the implementation mode of the present application, effective environmental impact variables include pH, organic carbon, population, temperature, land use type, soil parent material, elevation, slope, aspect, parent rock, normalized vegetation index, PM2.5, carbon emissions, GDP per capita, rainfall, average annual wind speed, the distance between the sampling point and the water system distribution, the distance between the sampling point and the transportation network, and the distance between the sampling point and the heavy metal enterprise.

[0026] According to an embodiment of the present application, the method for obtaining the pollution risk index of each target heavy metal includes:

[0027] The pollution risk index is determined based on the following calculation formula:

[0028]

[0029] In the formula, As potential ecological risk factors, is the pollution index of target heavy metal i, T ir is the toxicity response parameter of target heavy metal i, C i is the target heavy metal content data of target heavy metal i, C b is the soil environmental background value of target heavy metal i, and RI is the overall potential ecological risk index.

[0030] According to an embodiment of the present application, the steps of establishing a heavy metal risk zoning and grading model based on the classification label, the training set, and the test set and based on a random forest model include:

[0031] The hyperparameters in the random forest model are determined based on the training set using a grid search method and 5-fold cross validation to obtain a heavy metal risk zoning and grading model. The maximum depth of the random forest model is 10 to 50. The number of trees in the random forest model is 1 to 150, and the minimum number of samples required for node splitting in the random forest model is 2 to 15.

[0032] The test set is used to evaluate the heavy metal risk zoning and grading model.

[0033] According to an embodiment of the present application, the step of evaluating the heavy metal risk zoning and grading model based on the training set includes:

[0034] The heavy metal risk zoning and grading model was evaluated based on the training set according to accuracy, precision, recall, F1 value, ROC curve and AUC.

[0035] According to the implementation mode of the present application, it also includes:

[0036] The results of identifying and zoning the soil heavy metal pollution risks in the study area were spatially visualized.

[0037] In the above-mentioned soil heavy metal pollution risk zoning and grading method based on machine learning, the mapping relationship between effective environmental impact variables and soil heavy metal content and different levels of heavy metal pollution risk data are used as binary labels for the risk classification combination model to accurately identify the potential heavy metal pollution risk areas. At the same time, the model interpretability is used to quantify the key factors of different heavy metal pollution and the pollution differences of different degrees of pollution risks. This method identifies soil heavy metal pollution risks quantitatively, qualitatively and from multiple perspectives, and realizes the scientific division of soil heavy metal pollution potential risk areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] 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 or the description of the prior art 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 the structures shown in these drawings without paying creative work.

[0039] Figure 1 It is a soil heavy metal pollution risk zoning and grading map of Hg and Cd within a province in one implementation mode of the present application;

[0040] Figure 2 It is an environmental factor identification diagram of high, medium, low and very low risks in one implementation mode of the present application.

[0041] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with the implementation methods and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0042] 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.

[0043] It should be noted that all directional indications (such as up, down, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0044] In addition, in the present invention, the descriptions such as "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features.

[0045] Furthermore, the technical solutions between the various embodiments of the present invention may be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in the field. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0046] The present invention provides a soil heavy metal pollution risk zoning and grading method based on machine learning, comprising the following steps:

[0047] S100: Obtain the heavy metal sample data and environmental covariance data in the study area respectively as the corresponding initial sample data set of the target heavy metal. The heavy metal sample data at least includes the target heavy metal content data in the soil and the sampling point location information. The environmental covariance data includes soil physical and chemical properties, environmental impact factors, and socio-economic impact factors. The target heavy metals include two or more of Cd, Pb, Hg and As.

[0048] The research area is not limited, such as counties, cities, provinces and other levels. The target heavy metals include two or more of Cd, Pb, Hg and As. The heavy metal sample data of each target heavy metal includes the target heavy metal content data (i.e. the metal content, such as Pb content) and sampling point location information (such as sampling time, sampling depth and sample detection method, etc.).

[0049] Environmental covariates include a variety of physical and chemical properties, such as soil pH, soil organic matter, etc.; as well as a variety of environmental and socio-economic influencing factors, such as: soil type, land use type, vegetation cover, elevation, slope, slope aspect, atmospheric deposition, average wind speed, average precipitation, transportation network, water system distribution, per capita GDP, population density distribution, and full-caliber data on key heavy metal-related industries and enterprises.

[0050] In some embodiments, the soil physical and chemical properties include soil pH and soil organic matter.

[0051] The environmental impact factors include soil type, land use type, vegetation cover, elevation, slope, slope aspect, atmospheric deposition, average wind speed, average precipitation, transportation network, and water system distribution.

[0052] The socio-economic influencing factors include GDP per capita, population density distribution and geographical location of heavy metal enterprises.

[0053] There are 46 relevant features of heavy metal sample data and environmental covariate data for each target heavy metal.

[0054] S200: Preprocessing each initial sample data set to obtain a corresponding research sample data set. The preprocessing includes screening to obtain effective environmental impact variables.

[0055] In some embodiments, the step of preprocessing the initial sample data set to obtain the research sample data set includes:

[0056] S21. Interpolate missing values ​​of the initial sample data set to obtain a primary sample data set. Missing value interpolation methods include linear interpolation, polynomial interpolation, spline interpolation, Kriging interpolation, and machine learning interpolation.

[0057] S22. Perform feature scaling on different variables in the initial sample data set to obtain a secondary sample data set.

[0058] Since different variables have different types and value ranges, in order to avoid such situations, the data set can be feature scaled according to actual needs to ensure that all variables are in a similar range, which is conducive to faster convergence of the algorithm and improves the accuracy of the model.

[0059] The feature scaling methods include normalization and standardization. Data standardization methods include Min-Max standardization, Z-Score standardization, decimal calibration standardization, mean normalization, vector normalization, and exponential conversion. Among them, common function methods of exponential conversion include lg function, Softmax function, and Sigmoid function.

[0060] S23. Perform dimensionality reduction and independence test on the secondary sample data set, and determine the sample data set corresponding to the effective environmental impact variables as the research sample data set.

[0061] The dimensionality reduction process is divided into linear dimensionality reduction and nonlinear dimensionality reduction. Linear dimensionality reduction includes principal component analysis (PCA) and linear judgment analysis (LDA); nonlinear dimensionality reduction includes KPCA and KICA based on kernel functions and ISOMAP and MVU based on eigenvalues. At the same time, the spearman correlation coefficient independence test is performed to simplify the attributes of the features, remove redundancy, obtain more representative attributes, and further improve the modeling accuracy.

[0062] In some embodiments, dimensionality reduction and the spearman correlation coefficient independence test are used to exclude features with high correlation, delete redundant features, and reduce the number of effective environmental impact variables from 46 to 19 to obtain an analysis sample data set. The effective environmental impact variables include pH, organic carbon, population, temperature, land use type, soil parent material, elevation, slope, slope aspect, parent rock, vegetation normalized index, PM2.5, carbon emissions, GDP per capita, rainfall, average annual wind speed, the distance between the sampling point and the water system distribution, the distance between the sampling point and the transportation road network, and the distance between the sampling point and the heavy metal enterprise.

[0063] S300: Obtain the pollution risk index of each target heavy metal as a classification label for the heavy metal risk zoning and grading model.

[0064] In some embodiments, the method for obtaining the pollution risk index of each target heavy metal includes:

[0065] The pollution risk index is determined based on the following calculation formula:

[0066]

[0067] In the formula As potential ecological risk factors, is the pollution index of target heavy metal i, T ir is the toxicity response parameter of target heavy metal i, C i is the target heavy metal content data of target heavy metal i, C b is the soil environmental background value of target heavy metal i, and RI is the overall potential ecological risk index.

[0068] Among them, the concentrations of each target heavy metal T ir , C b The values ​​can be obtained by looking up reference books or manuals. i is the target heavy metal content data of the target heavy metal i, that is, the target heavy metal content data in the heavy metal sample data, which is the actual measured value.

[0069] In this example, the toxicity response parameters of As, Cd, Pb, and Hg are 10, 30, 5, and 40, respectively.

[0070] S400: After mixing the research sample data sets, randomly divide them into training set and test set.

[0071] For example, the research sample data set is randomly divided into a training set and a test set at a ratio of 9:1. Each training set includes the research sample data set of each target heavy metal. Similarly, each test set includes the research sample data set of each target heavy metal.

[0072] S500: Establishing a heavy metal risk zoning and grading model according to the classification labels, the training set, the test set, and based on a random forest model.

[0073] Machine learning classification task models include KNN, neural network, random forest (RF), support vector machine (SVR), k-means clustering, self-organizing map (SOM), etc.

[0074] Among them, common methods for evaluating models include indicators such as accuracy (Ac), precision (Precision), recall (Recall), F1 value, ROC curve and AUC.

[0075] In the classification task, samples can be divided into true positive (TP), true negative (TN), false positive (FP), and false negative (FN) according to the combination of their true results and the model's predicted results. Based on the evaluation results of each model, the best performing model is selected as the final prediction model. Each parameter can be adjusted based on the original model, and the optimal parameters can be observed by visualizing the graphs.

[0076] Among them, the RF algorithm has the best prediction performance. Overall, the accuracy of the soil heavy metal risk classification model can reach 0.8.

[0077] S600: Based on the heavy metal risk zoning and grading model, the soil heavy metal pollution risk of the study area is identified and zoned according to the effective environmental impact variables corresponding to the unknown areas in the study area.

[0078] On the basis of the heavy metal risk zoning and grading model, the effective environmental impact variables of the unknown area are substituted into the model to obtain the zoning of the soil heavy metal pollution risk in the unknown area. Combined with the effective environmental impact variables of the known area, such as the sampling point, the zoning of the soil heavy metal pollution risk in the known area is obtained. In this way, the entire study area can be identified and zoned.

[0079] In the above-mentioned soil heavy metal pollution risk zoning and grading method based on machine learning, the mapping relationship between effective environmental impact variables and soil heavy metal content and different levels of heavy metal pollution risk data are used as binary labels for the risk classification combination model to accurately identify the potential heavy metal pollution risk areas. At the same time, the model interpretability is used to quantify the key factors of different heavy metal pollution and the pollution differences of different degrees of pollution risks. This method identifies soil heavy metal pollution risks quantitatively, qualitatively and from multiple perspectives, and realizes the scientific division of soil heavy metal pollution potential risk areas.

[0080] In some embodiments, the environmental covariance data also includes determining the distances between the transportation network, water system distribution, and geographical locations of heavy metal enterprises and the sampling points.

[0081] According to the implementation mode of the present application, the step of determining the distances between the transportation network, water system distribution and the geographical location of the heavy metal enterprise and the sampling point includes:

[0082] According to the fuzzy matching of heavy metal enterprise names based on administrative divisions, the geographical location information of heavy metal enterprises is obtained.

[0083] Convert the geographical location information of heavy metal enterprises into vector data based on administrative divisions.

[0084] Based on the vector data corresponding to the traffic network, water system distribution and heavy metal enterprises, the distances between the sampling points and the three are determined.

[0085] Specifically, the GDP per capita and the full-caliber enterprise data information of key industries involving heavy metals are converted into structured data. The specific method is: according to the fuzzy matching of the GDP per capita and the full-caliber enterprise data information of key industries involving heavy metals based on administrative divisions, each information is matched with the corresponding geographical location information to obtain an associated data set containing basic information of longitude and latitude. At the same time, the GDP per capita and the full-caliber enterprise data information element files are imported into ArcGIS10.8, converted into vector data based on administrative divisions, and the coordinate system is converted to the WGS-84 coordinate system. It should be noted that GDP per capita has nothing to do with the calculation of the following proximity distance. The purpose of this step is to facilitate subsequent visualization operations.

[0086] Furthermore, based on the vector data of traffic network, water system distribution and all-caliber key heavy metal-related industries, the "Analysis-Near-NEAR_DIST" tool in ArcGIS 10.8 was used to calculate the proximity distances from the sampling points to the above three elements. Machine learning technology has the ability to process high-dimensional data and categorical variables, and comprehensively considers the factors affecting heavy metal content from different environmental covariates, greatly improving the spatial prediction accuracy of machine learning technology.

[0087] In some embodiments, each sample data in the research sample data set further includes a first ionization energy and a regional background value.

[0088] A hybrid risk classification model is established using the training and test sets of all elements. The final hybrid risk classification model can classify the risks of one or more elements. However, the hybrid model may not perform well in classifying the risks of some metals. In order to distinguish different elements and facilitate the prediction of a single element using the hybrid model, two heavy metal descriptors, namely the first ionization energy (IE, kJ / mol) and the regional background value (Bv), were selected from the feature input data set. They are one of the important characteristics of heavy metals and have no correlation with each other.

[0089] In some embodiments, the step of establishing a heavy metal risk zoning and grading model based on the classification label, the training set, and the test set and based on a random forest model includes:

[0090] The hyperparameters in the random forest model are determined based on the training set using a grid search method and 5-fold cross validation to obtain a heavy metal risk zoning and grading model. The maximum depth of the random forest model is 10 to 50. The number of trees in the random forest model is 1 to 150, and the minimum number of samples required for node splitting in the random forest model is 2 to 15.

[0091] The test set is used to evaluate the heavy metal risk zoning and grading model.

[0092] In some embodiments, the step of evaluating the heavy metal risk zoning and grading model based on the training set includes:

[0093] The heavy metal risk zoning and grading model was evaluated based on the training set according to accuracy, precision, recall, F1 value, ROC curve and AUC.

[0094] In this step, common methods for evaluating models include accuracy (Ac), precision (Precision), recall (Recall), F1 value, ROC curve and AUC. The relevant calculation indicators are as follows:

[0095]

[0096] In the classification task, samples can be divided into true positive (TP), true negative (TN), false positive (FP), and false negative (FN) according to the combination of their true results and the model's prediction results. Based on the evaluation results of each model, the best performing model is selected as the final prediction model, and each parameter can be adjusted separately based on the original model.

[0097] In some embodiments, it also includes:

[0098] The results of identifying and zoning the soil heavy metal pollution risk in the study area are spatially visualized. For example, spatial visualization is performed on the map of the study area to intuitively observe the results of identification and zoning.

[0099] The technical solution of the present application is described below in conjunction with specific embodiments.

[0100] Example 1

[0101] In this example:

[0102] Continue the research with the provincial scale as the research object.

[0103] S1. Obtain soil heavy metal concentrations at multiple sampling points in the area to be tested, extract raster data of relevant factors from the public database of the Food and Agriculture Organization of the United Nations, global weather and climate data, and import them into geographic information software (Arc GIS) for spatial geographic coordinate matching. There are a total of 46 relevant features, such as pH, cation exchange capacity (CEC), population, rainfall, soil moisture, etc.

[0104] S2. Preprocess the initial sample data set to obtain the research sample data set

[0105] Specifically, random forest is used to fill missing values ​​(missing values ​​less than 5%) for sample points where influencing variables are not extracted. Random forest fills in the least missing values ​​first. When filling in a feature, the missing values ​​of other features are replaced with 0 first, and then the regression prediction is cyclically performed.

[0106] The 46 input features were standardized and normalized in turn, and the prediction targets were logarithmized. Dimensionality reduction and the spearman correlation coefficient independence test were used to exclude features with high correlation and delete redundant features, so that the effective environmental impact variables were reduced from 46 to 19, and the analysis sample data set was obtained, including: pH, organic carbon, population, temperature, land use type, soil parent material, elevation, slope, slope aspect, parent rock, vegetation normalization index, PM2.5, carbon emissions, GDP, rainfall, annual average wind speed, sample point to water distribution, traffic network and distance from heavy-related enterprises. At the same time, the soil heavy metal content Y was logarithmized, and the formula is as follows:

[0107] Y=lny

[0108] S3. Obtain the pollution risk index of each target heavy metal as the classification label of the heavy metal risk zoning and grading model.

[0109] S4. The research sample data set was randomly divided into a training set and a test set at a ratio of 9:1, and the random forest (RF) algorithm, support vector machine (SVM) algorithm, K nearest neighbor regression (KNR) algorithm and other algorithms were used to perform machine learning on the training set.

[0110] S5. Obtain heavy metal pollution risk indicators as labels for the risk classification combination model, build a machine learning risk classification task model based on the compiled heavy metal pollution risk data sets of different levels and the optimal influencing variable characteristics, train the machine learning risk classification task model based on the partitioned data sets, and divide the study area into potential risk areas of different levels;

[0111] A hybrid risk classification model is established using the training set and test set of all elements. The final hybrid risk classification model can classify the risks of one or more elements. However, the hybrid model may not perform well in classifying the risks of some metals. In order to distinguish different elements and facilitate the prediction of a single element using a hybrid model, the optimal influencing variable feature is to add two heavy metal descriptors, namely the first ionization energy (IE, kJ / mol) and the regional background value (Bv), on the basis of the effective environmental impact variable.

[0112] Comprehensive performance evaluation indicators of the three integrated algorithms used. Among them, the RF algorithm has the best prediction performance. In general, the accuracy of the soil heavy metal risk classification model can reach 0.8 (Table 1).

[0113] Table 1 Comprehensive performance evaluation indicators of three integrated algorithms

[0114]

[0115] Based on the soil heavy metal pollution risk classification combination model, the pollution risk of different heavy metals can be spatially visualized to identify potential risk areas of different levels. At the same time, combined with the interpretable analysis of the model, the source differences of different levels of risk can be clarified. Taking Hg and Cd as an example ( Figure 1 ), high values ​​of Hg are scattered in the study area, while the pollution risk of Cd is the most serious, and a large area of ​​the study area is at a high pollution risk level. Figure 2 , where a) to d) are environmental factor identification diagrams for high, medium, low, and very low risks, respectively. The high-risk area is closest to polluting enterprises, and the land use type is mainly construction land. This zone requires strict pollution monitoring; the medium-risk area has dense road and water networks, is located in the middle and lower reaches of the river, has dense upstream industrial distribution, and serious air pollution. The low-risk area has weaker air pollution than the medium-risk area, and the traffic road network is sparse; the very low-risk area is far from the heavily polluting enterprises and is mainly affected by the land use type and vegetation coverage index.

[0116] The above technical solutions of the present invention are only preferred embodiments of the present invention, and the patent scope of the present invention is not limited thereto. All equivalent structural changes made by using the contents of the present invention specification and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A soil heavy metal pollution risk zoning and grading method based on machine learning, characterized in that: The following steps are involved: Obtain the heavy metal sample data and environmental covariance data in the study area respectively as the corresponding initial sample data set of the target heavy metal; wherein the heavy metal sample data at least includes the target heavy metal content data in the soil and the sampling point location information; the environmental covariance data includes the soil physical and chemical properties, environmental impact factors, and socio-economic impact factors; the target heavy metals include two or more of Cd, Pb, Hg and As; Preprocessing each initial sample data set to obtain a corresponding research sample data set; the preprocessing includes screening to obtain effective environmental impact variables; Obtain the pollution risk index of each target heavy metal as the classification label of the heavy metal risk zoning and grading model; After mixing the data sets of each research sample, they are randomly divided into training set and test set; Establishing a heavy metal risk zoning and grading model based on the classification label, the training set, and the test set and on a random forest model; Based on the heavy metal risk zoning and grading model, the soil heavy metal pollution risks in the study area are identified and zoned according to the effective environmental impact variables corresponding to the unknown areas in the study area.

2. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: The soil physical and chemical properties include soil pH value and soil organic matter; The environmental impact factors include soil type, land use type, vegetation cover, elevation, slope, slope aspect, atmospheric deposition, average wind speed, average precipitation, transportation network, and water system distribution; The socio-economic influencing factors include GDP per capita, population density distribution, and the geographical location of heavy metal enterprises; The environmental covariance data also includes the distances between the transportation network, water system distribution and the geographical location of heavy metal enterprises and the sampling points.

3. The soil heavy metal pollution risk zoning and grading method based on machine learning according to claim 2 is characterized in that: The step of determining the distances between the transportation network, the water system distribution and the geographical location of the heavy metal enterprise and the sampling point comprises: According to the fuzzy matching of the heavy metal enterprise names based on administrative divisions, the geographical location information of the heavy metal enterprises is obtained; Convert the geographical location information of heavy metal enterprises into vector data based on administrative divisions; Based on the vector data corresponding to the traffic network, water system distribution and heavy metal enterprises, the distances between the sampling points and the three are determined.

4. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: The step of preprocessing the initial sample data set to obtain the research sample data set includes: Interpolating missing values ​​of the initial sample data set to obtain a primary sample data set; Perform feature scaling on different variables in the primary sample data set to obtain a secondary sample data set; The secondary sample data set is subjected to dimensionality reduction and independence test, and the sample data set corresponding to the effective environmental impact variables is determined as the research sample data set.

5. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 4 is characterized in that: Each sample data in the research sample data set also includes a first ionization energy and a regional background value.

6. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: Effective environmental influencing variables include pH, organic carbon, population, temperature, land use type, soil parent material, elevation, slope, aspect, parent rock, normalized vegetation index, PM2.5, carbon emissions, GDP per capita, rainfall, average annual wind speed, the distance between the sampling point and the water system distribution, the distance between the sampling point and the transportation network, and the distance between the sampling point and heavy metal enterprises.

7. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: The method for obtaining the pollution risk index of each target heavy metal includes: The pollution risk index is determined based on the following calculation formula: Toxicity response parameter of metal i, C i is the target heavy metal content data of target heavy metal i, C b is the soil environmental background value of target heavy metal i, and RI is the overall potential ecological risk index.

8. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: The steps of establishing a heavy metal risk zoning and grading model based on the classification label, the training set, and the test set and on the basis of a random forest model include: According to the training set, the hyperparameters in the random forest model are determined by using a grid search method and a 5-fold cross validation to obtain a heavy metal risk zoning and grading model; the maximum depth of the random forest model is 1 to 50; the number of trees in the random forest model is 1 to 150, and the minimum number of samples required for node splitting in the random forest model is 2 to 15; The test set is used to evaluate the heavy metal risk zoning and grading model.

9. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to claim 1 is characterized in that: The step of evaluating the heavy metal risk zoning and grading model based on the training set includes: The heavy metal risk zoning and grading model was evaluated based on the training set according to accuracy, precision, recall, F1 value, ROC curve and AUC.

10. The method for zoning and grading soil heavy metal pollution risk based on machine learning according to any one of claims 1 to 9, characterized in that: Also includes: The results of identifying and zoning the soil heavy metal pollution risks in the study area were spatially visualized.

Images