Regional soil pollution cause mode analysis method

Abstract

The invention discloses a regional soil pollution cause mode analysis method, belongs to the technical field of soil pollution analysis, and can solve the problem that an existing analysis method can only quantify or determine the current situation of soil pollutants singly and cannot quantify or determine the intermediate pollution process of soil pollution. The method comprises the following steps: S1, carrying out grid division on a research area, obtaining soil pollution data and potential pollution sources of each grid area, and determining a pollution mode of each grid area according to the soil pollution data; s2, training the machine learning model by using the soil pollution data, the potential pollution sources and the pollution modes of all the grid regions to obtain a pollution source prediction model; and S3, according to the soil pollution data of the to-be-detected area, predicting a pollution source of the to-be-detected area by using the pollution source prediction model to obtain a pollution source prediction result of the to-be-detected area. The method is used for analyzing the soil pollution cause mode.

Description

Technical Field

[0001] The invention relates to a method for analyzing a regional soil pollution cause pattern, and belongs to the technical field of soil pollution analysis. Background Art

[0002] Regional soil pollution has become a major national problem in the current environmental field. This is mainly because my country has experienced rapid industrialization and urbanization, which has largely led to the release of a large number of pollutants into the soil. The analysis of the causes of regional soil pollution is an important part of building a precise control system for the entire chain of regional soil pollution. The existing soil pollution cause analysis technology mainly relies on homogeneous single factor association analysis or traditional multivariate statistical methods, which can roughly analyze the pollution sources of regional soil pollutants, but has not yet risen to the level of source analysis for different soil pollution patterns. In addition, the soil pollution process is complex, and the soil pollution source analysis technology is difficult and uncertain. The existing soil pollution cause analysis methods can only quantify or characterize the current status of soil pollutants in a single way, and cannot quantify or characterize the intermediate pollution process of soil pollution. The defects in these methods seriously restrict the upgrading of pollution cause analysis from phenomenon description to process analysis. Summary of the invention

[0003] The present invention provides a method for analyzing the cause pattern of regional soil pollution, which can solve the problem that the existing analysis method can only quantify or characterize the current status of soil pollutants but cannot quantify or characterize the intermediate pollution process of soil pollution.

[0004] The present invention provides a method for analyzing a regional soil pollution cause pattern, the method comprising:

[0005] S1. Divide the study area into grids, obtain soil pollution data and potential pollution sources in each grid area, and determine the pollution pattern of each grid area based on the soil pollution data;

[0006] S2. Use soil pollution data, potential pollution sources and pollution patterns of all grid areas to train the machine learning model to obtain a pollution source prediction model;

[0007] S3. According to the soil pollution data of the area to be tested, the pollution source of the area to be tested is predicted using the pollution source prediction model to obtain a pollution source prediction result of the area to be tested.

[0008] Optionally, determining the pollution mode of each grid area according to the soil pollution data in S1 specifically includes:

[0009] Determine the pollution code sequence of each grid area according to the soil pollution data;

[0010] The pollution mode of each grid area is determined according to the pollution code sequence.

[0011] Optionally, determining the pollution code sequence of each grid area according to the soil pollution data specifically includes:

[0012] Obtain the pollutant concentration in the soil pollution data of each grid area collected at preset time intervals to obtain a time series of the pollutant concentration in each grid area;

[0013] The pollutant concentration greater than the threshold standard in the time series is marked as a pollution code, and the pollutant concentration less than or equal to the threshold standard is marked as a clean code, so as to obtain a pollution code sequence for each grid area.

[0014] Optionally, determining the pollution code sequence of each grid area according to the soil pollution data specifically includes:

[0015] Perform secondary grid division on each grid area to obtain multiple grid sub-areas, and determine the pollutant concentration of each grid sub-area corresponding to it according to the soil pollution data of each grid area;

[0016] The pollutant concentrations of multiple grid sub-areas corresponding to each grid area are sorted according to the spatial positions to obtain the spatial sequence of the pollutant concentrations of each grid area;

[0017] The pollutant concentration greater than the threshold standard in the spatial sequence is marked as a pollution code, and the pollutant concentration less than or equal to the threshold standard is marked as a clean code, so as to obtain a pollution code sequence for each grid area.

[0018] Optionally, determining the pollution mode of each grid area according to the pollution code sequence specifically includes:

[0019] Traversing the contaminated coding sequence;

[0020] When all codes in the pollution code sequence are pollution codes, determining that the pollution mode of the corresponding grid area is a persistent mode;

[0021] When all codes in the contaminated code sequence are clean codes, the contamination mode of the corresponding grid area is determined to be an uncontaminated mode;

[0022] When the first code in the pollution code sequence is a pollution code, and the remaining codes after the first appearance of the clean code are all clean codes, it is determined that the pollution mode of the corresponding grid area is a control mode;

[0023] When the first code in the contamination code sequence is a clean code, and the remaining codes after the first appearance of the contamination code are all contamination codes, it is determined that the contamination mode of the corresponding grid area is an outbreak mode;

[0024] Otherwise, it is determined that the pollution mode of the corresponding grid area is a fluctuation mode.

[0025] Optionally, the potential pollution sources include industrial sources, transportation sources, natural sources, domestic sources and agricultural sources.

[0026] Optionally, the S2 specifically includes:

[0027] Determine the training data set based on soil pollution data, potential pollution sources and pollution patterns of all grid areas;

[0028] The training data set is used to train the machine learning model to obtain a pollution source prediction model.

[0029] Optionally, determining the training data set based on the soil pollution data, potential pollution sources and pollution patterns of all grid areas specifically includes:

[0030] The soil pollution data, potential pollution sources and pollution patterns of all grid areas are combined into an original data set, and the original data set is subjected to maximum and minimum scaling and standard scaling respectively to obtain a maximum and minimum scaling data set and a standard scaling data set, and the original data set, the maximum and minimum scaling data set and the standard scaling data set are used as training data sets.

[0031] Optionally, the machine learning model is any one of log-probability regression, decision tree, gradient boosting decision tree, random forest, multi-layer perceptron and support vector machine.

[0032] Optionally, the S3 specifically includes:

[0033] Divide the area to be tested into grids, obtain soil pollution data for each grid area, and determine the pollution mode of each grid area according to the soil pollution data;

[0034] All soil pollution data and pollution patterns of the area to be tested are input into the pollution source prediction model to obtain the pollution source prediction result of the area to be tested.

[0035] The beneficial effects that the present invention can produce include:

[0036] The regional soil pollution cause pattern analysis method provided by the present invention constructs a pollution pattern for each grid area in the study area to characterize the development trend of soil pollution in the study area, so that the trained pollution source prediction model can better predict potential pollution sources. This solves the problem that the existing methods can only quantify or qualify the current status of soil pollutants but cannot quantify or qualify the intermediate pollution process of soil pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flow chart of a method for analyzing regional soil pollution cause patterns provided by an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of grid division provided by an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of the correspondence between the pollution coding sequence and the pollution mode provided by an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of a flow chart of determining a pollution mode according to a pollution coding sequence provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention is described in detail below in conjunction with embodiments, but the present invention is not limited to these embodiments.

[0042] The embodiment of the present invention provides a method for analyzing the cause pattern of regional soil pollution. Figure 1 As shown, the method includes:

[0043] S1. Divide the study area into grids, obtain soil pollution data and potential pollution sources in each grid area, and determine the pollution pattern of each grid area based on the soil pollution data.

[0044] refer to Figure 1 As shown in the figure, the spatial location of the sample points is arranged based on the grid point method. In practical applications, the study area is first divided into grids. For example, the grid interval can be 3×3m in length and width or 3×3×3m in length, width and height; Figure 2 The points are arranged in the center of the grid, and the sample collection and analysis are strictly carried out in accordance with the requirements of the Technical Specifications for Soil Environmental Monitoring (HJ / T166-2004).

[0045] Soil pollutant data mainly consists of pollutants (such as arsenic, beryllium, cadmium, chromium, copper, lead, mercury, nickel, zinc, tin, cyanide, fluoride, polycyclic aromatic hydrocarbons, etc.), point coordinates (such as X, Y or X, Y, Z), concentration values, ID numbers, auxiliary variables (soil physical properties, spatial layout of the study area, etc.); potential pollution sources mainly include industrial sources, transportation sources, natural sources, domestic sources and agricultural sources.

[0046] In S1, the pollution pattern of each grid area is determined based on the soil pollution data, including:

[0047] (1) Determine the pollution coding sequence for each grid area based on soil pollution data.

[0048] Constructing a pollution coding sequence based on the time process specifically includes: firstly obtaining the pollutant concentration in the soil pollution data of each grid area collected at preset time intervals, and obtaining the time series of the pollutant concentration of each grid area; then marking the pollutant concentration greater than the threshold standard in the time series as a pollution code, and marking the pollutant concentration less than or equal to the threshold standard as a clean code, and obtaining the pollution coding sequence of each grid area.

[0049] The above-mentioned pollution code can be represented by character 1, and the above-mentioned clean code can be represented by character 0, so the pollution code sequence of each grid area is a binary code sequence.

[0050] Samples are collected at preset intervals (such as one week, one month or one year) at the spatial locations of the sample points, and soil pollution data at the sample points in each grid area are obtained. If the sample point location is damaged and data cannot be collected after a period of time, the sample point is re-deployed and data is collected at the location that meets the deployment conditions. Assuming that 10 months of sample data have been collected, a time series of pollutant concentrations in each grid area is formed. Then, according to the national soil exceeding threshold standard (such as: "Site Soil Environmental Risk Assessment Screening Value (DB11 / T 811-2011)"), it is judged whether the pollutant sample concentration value exceeds the standard and marked. If it exceeds the standard, it is indicated by "1", otherwise it is indicated by "0". (For example: the sample point concentration of arsenic is 50mg / kg, and the exceeding threshold for residential land is 20mg / kg, that is, the sample point is judged to exceed the standard and marked as "1"; otherwise, it is judged to be not exceeded and marked as "0"). Finally, according to the chronological order of sample collection, a binary coding sequence, namely the pollution coding sequence, is constructed for the sample points at the same location (e.g., the binary coding sequence of the first sample of arsenic contamination in month 10 is “0001111111”).

[0051] A pollution coding sequence is constructed based on the spatial process, specifically including: first, each grid area is divided into a second grid to obtain multiple grid sub-areas, and the pollutant concentration of each grid sub-area corresponding to it is determined according to the soil pollution data of each grid area; then the pollutant concentrations of the multiple grid sub-areas corresponding to each grid area are sorted according to the spatial position to obtain the spatial sequence of the pollutant concentration of each grid area; finally, the pollutant concentration greater than the threshold standard in the spatial sequence is marked as a pollution code, and the pollutant concentration less than or equal to the threshold standard is marked as a clean code to obtain the pollution coding sequence of each grid area.

[0052] Assuming that sample data can only be collected once according to research requirements, a second grid division is performed on each grid area. Figure 2 As shown in the example, the second division grid interval can be 1×1m in length and width or 1×1×1m in length, width and height, such as Figure 1Series A division in China. Obtain the soil pollution data at the sample point positions in each grid sub-region. Then, following the principle from left to right and top to bottom, sequentially take out the pollutant concentration values of 9 grid sub-regions in each grid area to obtain the spatial sequence of pollutant concentrations in each grid area. Then, according to the national soil over-standard threshold standard, determine whether the sample point exceeds the standard and mark it. Finally, construct a binary coding sequence, that is, the pollution coding sequence.

[0053] (2)Determine the pollution pattern of each grid area according to the pollution coding sequence.

[0054] Specifically include: traversing the pollution coding sequence;

[0055] When all the codes in the pollution coding sequence are pollution codes, determine that the pollution pattern of the corresponding grid area is the persistent pattern;

[0056] When all the codes in the pollution coding sequence are clean codes, determine that the pollution pattern of the corresponding grid area is the unpolluted pattern;

[0057] When the first code in the pollution coding sequence is a pollution code, and the remaining codes after the first occurrence of a clean code are all clean codes, determine that the pollution pattern of the corresponding grid area is the control pattern;

[0058] When the first code in the pollution coding sequence is a clean code, and the remaining codes after the first occurrence of a pollution code are all pollution codes, determine that the pollution pattern of the corresponding grid area is the outbreak pattern;

[0059] Otherwise, determine that the pollution pattern of the corresponding grid area is the fluctuation pattern.

[0060] The above pollution code can be represented by the character 1, and the above clean code can be represented by the character 0. Therefore, the pollution coding sequence of each grid area is a binary coding sequence.

[0061] Reference Figure 3 and Figure 4 As shown, determine the pollution pattern according to the Hamming distance. Assume that the length of the pollution coding sequence (B) of the soil pollution point is L. The value of the first position in B is B1. Then traverse B from left to right. If B1 = "1", then find the position (K, 1 < K ≤ L) where the character "0" first appears in B; similarly, when B1 = "0", K represents the position where the character "1" first appears in B.

[0062] When B1 = "1" and K does not exist, B is the "persistent pattern".

[0063] When B1 = "0" and K does not exist, B is the "unpolluted pattern".

[0064] When B1 = "1", K < L and S = Bk +B k+1 +…+B L =0, then B is "control mode".

[0065] When B1=“0”, S=LK, then B is “burst mode”.

[0066] When K exists and S∈(0,LK), then B is a "wave mode".

[0067] S2. Use the soil pollution data, potential pollution sources and pollution patterns of all grid areas to train the machine learning model to obtain a pollution source prediction model.

[0068] Specifically include:

[0069] (1) Determine the training dataset based on the soil pollution data, potential pollution sources, and pollution patterns of all grid areas.

[0070] Specifically, the soil pollution data, potential pollution sources and pollution patterns of all grid areas are combined into an original data set, and the original data set is subjected to maximum and minimum scaling and standard scaling respectively to obtain a maximum and minimum scaling data set and a standard scaling data set, and the original data set, the maximum and minimum scaling data set and the standard scaling data set are used as training data sets.

[0071] The embodiment of the present invention uses three data sets (original data set, maximum and minimum scaling data set, and standard scaling data set) for model training. Among them, the original data set contains soil pollution data, potential pollution sources and pollution patterns in all grid areas; the maximum and minimum scaling data set scales all features in the original data set to a specified range, usually [0, 1]. This method is implemented by the following formula: ,in, and are the minimum and maximum values ​​of each feature in the original data set, X is the sampled value, Represents the maximum and minimum scaled values. The standard scaled data set is to standardize the original data set. It transforms the data by subtracting the mean and dividing by the standard deviation so that the result has zero mean and unit variance. The conversion formula is: , where μ is the mean of each feature in the original dataset, σ is the standard deviation of each feature in the original dataset, and X is the sampled value. Represents the standard scaled value.

[0072] (2) Use the training data set to train the machine learning model to obtain a pollution source prediction model.

[0073] The machine learning model may be any one of log-odds regression, decision tree, gradient boosting decision tree, random forest, multi-layer perceptron, and support vector machine.

[0074] Based on the preprocessed training data set, we built a variety of different machine learning models (Logistic Regression (LR), Decision Tree (DT), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), Multilayer Perceptron (MLP) and Support Vector Machine (SVM), etc.), and used the Scikit-learn toolkit in Python to perform machine learning modeling.

[0075] First, determine the key hyperparameters of each machine learning model and the search range of GridSearchCV. Call the GridSearchCV module in Scikit-learn to determine the best hyperparameters for different algorithms and optimize the performance of the model.

[0076] GridSearchCV trains each machine learning model using the tuned hyperparameters to find the most accurate parameters in the validation set from all the parameters.

[0077] The Bayesian optimization (BO) hybrid model is introduced to adjust the machine learning model and compare it with the GridSearchCV results, including model performance and time consumption. BO allows to master prior knowledge, that is, to incorporate historical information in the iterative process, and avoid redundant and unnecessary evaluation of the objective function. In the present invention, the execution of BO is implemented by the Bayes_opt toolkit in Python.

[0078] In the present invention, the prior knowledge of the characteristics of different pollution sources includes but is not limited to: natural sources such as geological background are generally stable, usually in a "persistent mode". Living sources are related to population density and garbage disposal cycles. For example, more garbage may be generated during holidays, resulting in increased pollution concentrations, which are usually in a "fluctuation mode". Typical periodic characteristics of traffic sources, such as daily cycles (peak hours in the morning and evening), weekly cycles (differences between weekdays and weekends), and possible seasonal changes (such as an increase in diesel vehicles during the winter heating period), are usually in a "fluctuation mode"; industrial sources are related to production cycles and accidental emissions, such as regular emissions or sudden leaks, which are usually in a "fluctuation mode" or "outburst mode". Agricultural sources may be related to fertilization and irrigation seasons. For example, more fertilizers are used during the planting season, resulting in increased pollutant concentrations, which are usually in a "fluctuation mode".

[0079] The best model was determined based on the best results predicted by the model, and the SHAP method was used to analyze the soil pollutant data, auxiliary variables and potential pollution sources under different modes to screen out the key variables under different modes.

[0080] In the embodiment of the present invention, descriptive analysis can also be performed on the collected soil pollutant data to obtain the basic characteristics of the data (such as maximum value, minimum value, average value, coefficient of variation and geo-cumulative index, etc.), and preliminarily analyze the pollution degree of soil pollutants and the degree of external influence.

[0081] The present invention can also perform Pearson correlation analysis on the collected soil pollutant data. The strength of the correlation between elements in the soil can reflect the cause and homology of the pollution. If the correlation between elements is significant or extremely significant, it means that the elements have a homologous relationship and present complex pollution. Otherwise, there may be multiple different sources.

[0082] Based on the spatial location information of pollutants, the soil pollution distribution characteristics of the field soil profile are obtained to provide prior knowledge and result verification for determining the soil pollution sources under different pollution modes.

[0083] The key variables screened by the SHAP method are combined with the optimal model to finally obtain the most satisfactory model to identify the pollution sources of pollutants under different modes. The pollution sources of pollutants under different models are then verified and evaluated in combination with the prior knowledge obtained above.

[0084] S3. According to the soil pollution data of the area to be tested, the pollution source of the area to be tested is predicted using the pollution source prediction model to obtain the pollution source prediction result of the area to be tested.

[0085] Specifically include:

[0086] Divide the area to be tested into grids, obtain soil pollution data for each grid area, and determine the pollution pattern of each grid area based on the soil pollution data;

[0087] All soil pollution data and pollution patterns of the area to be tested are input into the pollution source prediction model to obtain the pollution source prediction results of the area to be tested.

[0088] The grid division of the area to be tested and the determination of the pollution mode of each grid area are the same as the above-mentioned grid division of the study area and the determination of the pollution mode of each grid area, which will not be repeated here.

[0089] The present invention constructs a pollution pattern for each grid area in the study area to characterize the development trend of soil pollution in the study area, so that the trained pollution source prediction model can better predict potential pollution sources. This solves the problem that the existing methods can only quantify or qualify the current status of soil pollutants but cannot quantify or qualify the intermediate pollution process of soil pollution.

[0090] The above are only a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for analyzing the cause pattern of regional soil pollution, characterized in that: The method comprises: S1. Divide the study area into grids, obtain soil pollution data and potential pollution sources in each grid area, and determine the pollution pattern of each grid area based on the soil pollution data; S2. Use soil pollution data, potential pollution sources and pollution patterns of all grid areas to train the machine learning model to obtain a pollution source prediction model; S3. According to the soil pollution data of the area to be tested, the pollution source of the area to be tested is predicted using the pollution source prediction model to obtain a pollution source prediction result of the area to be tested.

2. The method according to claim 1, characterized in that: Determining the pollution mode of each grid area according to the soil pollution data in S1 specifically includes: Determine the pollution code sequence of each grid area according to the soil pollution data; The pollution mode of each grid area is determined according to the pollution code sequence.

3. The method according to claim 2, characterized in that Determining the pollution code sequence of each grid area according to the soil pollution data specifically includes: Obtain the pollutant concentration in the soil pollution data of each grid area collected at preset time intervals to obtain a time series of the pollutant concentration in each grid area; The pollutant concentration greater than the threshold standard in the time series is marked as a pollution code, and the pollutant concentration less than or equal to the threshold standard is marked as a clean code, so as to obtain a pollution code sequence for each grid area.

4. The method according to claim 2, characterized in that: Determining the pollution code sequence of each grid area according to the soil pollution data specifically includes: Perform secondary grid division on each grid area to obtain multiple grid sub-areas, and determine the pollutant concentration of each grid sub-area corresponding to it according to the soil pollution data of each grid area; The pollutant concentrations of multiple grid sub-areas corresponding to each grid area are sorted according to the spatial positions to obtain the spatial sequence of the pollutant concentrations of each grid area; The pollutant concentration greater than the threshold standard in the spatial sequence is marked as a pollution code, and the pollutant concentration less than or equal to the threshold standard is marked as a clean code, so as to obtain a pollution code sequence for each grid area.

5. The method according to claim 3 or 4, characterized in that: Determining the pollution mode of each grid area according to the pollution code sequence specifically includes: Traversing the contaminated coding sequence; When all codes in the pollution code sequence are pollution codes, determining that the pollution mode of the corresponding grid area is a persistent mode; When all codes in the contaminated code sequence are clean codes, the contamination mode of the corresponding grid area is determined to be an uncontaminated mode; When the first code in the pollution code sequence is a pollution code, and the remaining codes after the first appearance of the clean code are all clean codes, it is determined that the pollution mode of the corresponding grid area is a control mode; When the first code in the contamination code sequence is a clean code, and the remaining codes after the first appearance of the contamination code are all contamination codes, it is determined that the contamination mode of the corresponding grid area is an outbreak mode; Otherwise, it is determined that the pollution mode of the corresponding grid area is a fluctuation mode.

6. The method according to claim 1, characterized in that The potential pollution sources include industrial sources, transportation sources, natural sources, domestic sources and agricultural sources.

7. The method according to claim 1, characterized in that The S2 specifically includes: Determine the training data set based on soil pollution data, potential pollution sources and pollution patterns of all grid areas; The training data set is used to train the machine learning model to obtain a pollution source prediction model.

8. The method according to claim 7, characterized in that The training data set is determined based on the soil pollution data, potential pollution sources and pollution patterns of all grid areas, specifically including: The soil pollution data, potential pollution sources and pollution patterns of all grid areas are combined into an original data set, and the original data set is subjected to maximum and minimum scaling and standard scaling respectively to obtain a maximum and minimum scaling data set and a standard scaling data set, and the original data set, the maximum and minimum scaling data set and the standard scaling data set are used as training data sets.

9. The method according to claim 1, characterized in that: The machine learning model is any one of logarithmic probability regression, decision tree, gradient boosting decision tree, random forest, multilayer perceptron and support vector machine.

10. The method according to claim 1, characterized in that The S3 specifically includes: Divide the area to be tested into grids, obtain soil pollution data for each grid area, and determine the pollution mode of each grid area according to the soil pollution data; All soil pollution data and pollution patterns of the area to be tested are input into the pollution source prediction model to obtain the pollution source prediction result of the area to be tested.

Images