Method for predicting standard exceeding of soil pollutants in coking enterprise

Abstract

The embodiment of the invention discloses a method for predicting standard exceeding of soil pollutants in coking enterprises. The method comprises the following steps: acquiring soil pollutant contents, pollution source information and natural environment information of a plurality of coking enterprises in a target area; calculating a plurality of indexes in the pollution source information and the natural environment information as independent variables, fitting a plurality of soil pollutant content prediction models by taking the soil pollutant content as a dependent variable, and optimizing parameters of each model by using a particle swarm algorithm; substituting the plurality of indexes of each coking enterprise in production into an optimal model in the plurality of models, and predicting whether the soil pollutants exceed the standard or not; wherein the indexes in the pollution source information comprise enterprise area, establishment time, production duration, ground hardening condition, staff number, pollutant emission standard comprehensive quantification result, scale, coke yield and environmental accident violation record information. According to the embodiment of the invention, the standard exceeding prediction of the soil pollutant content of the construction land in a large-scale area can be automatically realized.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of soil pollution prevention, and in particular to a method for predicting excessive soil pollutants in a coking enterprise. Background Art

[0002] Faced with the increasingly serious problem of soil pollution in industrial construction land, the "end-of-pipe treatment" approach not only consumes manpower and material resources, but also creates risks. Therefore, preventing soil pollution at the source has become a necessary measure for current production enterprises.

[0003] Due to the sheer number of operating enterprises, traditional survey methods such as sampling are unable to fully understand soil pollution and the extent of soil pollution exceeding standards across millions of operating enterprises. Therefore, a method for predicting excessive soil pollutant levels in construction land across large areas is urgently needed. Summary of the Invention

[0004] The embodiment of the present invention provides a method for predicting excessive soil pollutants in a coking enterprise in operation to solve the above technical problems.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting excessive soil pollutants in a coking enterprise, comprising:

[0006] Obtaining soil pollutant content, pollution source information, and natural environment information for multiple coking enterprises within a target area, wherein the target area includes multiple regions;

[0007] Calculating the correlation between various indicators in the pollution source information and natural environment information and the soil pollutant content, and retaining multiple indicators with the highest correlation;

[0008] Using the multiple indicators as independent variables and the soil pollutant content as the dependent variable, fitting multiple soil pollutant content prediction models, and using the particle swarm algorithm to optimize the parameters of each model;

[0009] Substituting the multiple indicators of the multiple coking enterprises in operation in the target area into the optimal model among the multiple soil pollutant content prediction models to predict whether the soil pollutants in each coking enterprise in operation exceed the standard;

[0010] Among them, the various indicators in the pollution source information include: enterprise area, establishment time, production time, ground hardening conditions, number of employees, comprehensive quantitative results of pollutant emission standards, scale, coke production and environmental accident violation record information.

[0011] In a second aspect, an embodiment of the present invention provides an electronic device, comprising:

[0012] one or more processors;

[0013] a memory for storing one or more programs,

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for predicting excessive soil pollutants in an in-production coking enterprise as described in any embodiment.

[0015] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting excessive soil pollutants in an in-production coking enterprise as described in any embodiment.

[0016] In summary, this embodiment provides a method for predicting soil pollutant levels exceeding standards at a coking enterprise. By collecting internal enterprise information and external natural environmental information, combined with the pollutant emission standards and production information implemented by the enterprise, it predicts the current and future soil pollutant levels of coking enterprises within the target area and analyzes the importance of each influencing factor. This method can achieve the following beneficial effects:

[0017] 1. Compared with the prediction of air pollutant concentrations, the formation process of on-site soil pollution is extremely complex, involving many influencing factors and having a long-term cumulative effect, which greatly increases the difficulty of prediction. This embodiment introduces a new indicator based on the concept of soil pollutant accumulation-the comprehensive quantitative result of pollutant emission standards. This indicator can more comprehensively describe the control level of coking enterprises during production and improve the accuracy of predictions. At the same time, this embodiment also uses remote sensing technology to determine whether the ground of the coking enterprise has been hardened-this is a key indicator to prevent pollutants from seeping into the soil after release. In addition, this embodiment also systematically collects environmental violation records of various enterprises-this indicator not only reveals abnormal emissions, but also reflects the environmental control capabilities of enterprises.

[0018] 2. Machine learning (ML) methods have demonstrated outstanding performance in the challenging task of predicting soil pollutant content on large-scale construction land. However, the predictive performance of ML models varies significantly under different parameter configurations, a key issue often overlooked in many studies. This example comprehensively and in-depth analyzes the predictive performance of ML models under various parameter combinations by introducing the dynamic parameter adjustment process of the particle swarm optimization (PSO) algorithm. The results show that the difference in model accuracy under different parameter configurations can reach 35% or even higher, highlighting the important role of parameter optimization in improving model performance. The PSO algorithm, with its unique swarm intelligence properties, flexibly adjusts the position and velocity of particles within the parameter search space, achieving dynamic optimization of ML model parameters, thereby ensuring efficient and accurate search for the optimal solution for each ML model. This process not only significantly improves the accuracy of machine learning predictions, but also greatly enhances the efficiency and practicality of predictions. By combining the particle swarm optimization algorithm (PSO), this example constructs an efficient and accurate machine learning model, aiming to provide strong technical support and guarantee for the prediction of soil pollutant content in enterprises.

[0019] 3. This example analyzes the importance of various influencing factors. Among them, the comprehensive quantitative results of pollutant emission standards contribute most significantly to the prediction of soil pollutant levels. A decrease in the comprehensive quantitative results of pollutant emission standards indicates that stricter pollutant emission standards have been implemented within the target area, effectively limiting pollutant emissions from enterprises. This measure significantly improves the environmental management and control capabilities of enterprises and effectively controls the accumulation of soil pollutants. In particular, because the implementation requirements of emission standards are divided by region, the quantitative results of emission standards in the same region are often similar. Therefore, the comprehensive quantitative results of pollutant emission standards are not applicable to small areas such as cities or counties, but are particularly suitable for large-scale regional soil pollution prediction, such as the national scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a flow chart of a method for predicting excessive soil pollutants in a coking enterprise provided by an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of pollutant emission standards for coking enterprises provided by an embodiment of the present invention;

[0023] Figure 3 This is a Spearman correlation analysis result between various indicators of a coking sample enterprise provided by an embodiment of the present invention;

[0024] Figure 4 This is a flow chart of a method for predicting excessive soil pollutants in a coking enterprise provided by an embodiment of the present invention;

[0025] Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0027] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0028] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0029] Figure 1 This is a flow chart of a method for predicting excessive soil pollutants in a coking enterprise in production provided by an embodiment of the present invention. This method is applicable to the situation where the current and future soil pollution conditions of multiple coking enterprises in production distributed in different regions are predicted, and is executed by electronic equipment. Figure 1 As shown, the method specifically includes:

[0030] S110. Obtain soil pollutant content, pollution source information, and natural environment information of multiple coking enterprises in a target area, wherein the target area includes multiple regions.

[0031] The target region refers to a large-scale area encompassing multiple regions. The following examples illustrate this using the entire country as the target region. This step constructs a soil pollutant content prediction index system and obtains various indicator data and soil pollutant content data (if available) from multiple coking enterprises within the target region as the data source for the entire method.

[0032] In one specific implementation, 2020 can be used as the forecast time point (i.e., predicting soil pollution after 2020). Information on various coking enterprises can be obtained from different data sources. Through multi-source data collection and collation, a total of 1,884 coking enterprises in 2020, including those in operation and those closed or relocated, were compiled. Of these, 719 active coking enterprises included information such as company name, longitude and latitude, establishment date, production hours, number of employees, compliance with pollutant emission standards, environmental accident violation records, sunshine hours, rainfall, and temperature. In addition to the aforementioned information, the 198 closed or relocated coking enterprises also included soil benzo[a]pyrene content, coke production, and closure date.

[0033] Specifically, this example used publicly available contaminated site reports and literature to collect soil benzo[a]pyrene (B[a]pyrene) data from 198 closed and relocated coking plants with soil B[a]pyrene levels. Based on relevant standards, this example considers B[a]pyrene levels greater than 1.5 mg / kg in coking plants to be excessive. The maximum B[a]pyrene level at each coking plant's soil monitoring points was used for B[a]pyrene content.

[0034] Optionally, the soil pollutant content of each coking enterprise is used as a sample to form a sample set; based on the coefficient of variation of the sample set, determine whether there are outliers; if there are outliers, fit the data distribution of the sample set; if the data distribution is positively skewed compared to the normal distribution, remove from the sample set the soil pollutant content of the outliers that is greater than the mean plus N times the standard deviation; if the data distribution is negatively skewed compared to the normal distribution, remove from the sample set the soil pollutant content of the outliers that is less than the mean minus N times the standard deviation, where N is a natural number. For example, the soil benzo[a]pyrene content of the 198 coking enterprises mentioned above varies greatly, with the maximum value being 546 mg / kg and the minimum value being 0.05 mg / kg. This results in a sample standard deviation of 71.17 and a coefficient of variation of 3.405, indicating that outliers may exist in the data. Since the distribution and skewness of benzo[a]pyrene content in coking plants are positive, and the distribution of benzo[a]pyrene data is positively skewed compared to a normal distribution, this example uses the Sigma Principle to eliminate samples from coking plants with benzo[a]pyrene levels greater than the mean plus one standard deviation. Based on the results, eight coking plants were ultimately eliminated, leaving a total of 190 plants participating in the prediction model.

[0035] In addition, since the factors affecting the degree of soil pollution in construction land are not only related to the company's own conditions but also to external natural conditions, the soil pollution impact indicators of coking enterprises included in the modeling scope of this embodiment include two major categories: one is pollution source information, and the other is natural environment information. Pollution source information mainly includes enterprise area, establishment time, production time, ground hardening, number of employees, comprehensive quantitative results of pollutant emission standards, scale, coke production (abbreviated as production), and environmental accident violation records. Natural condition information includes sunshine duration, rainfall, and temperature. Among them, the methods for obtaining each indicator are shown in Table 1.

[0036] Table 1

[0037]

[0038] Among them, large, medium, small and micro-sized enterprises can be quantified as 4, 3, 2 and 1, and those with and without ground hardening are defined as 1 and 0.

[0039] Regarding the comprehensive quantitative results of pollutant emission standards, since coking enterprises in different regions adopt different pollutant emission standards, the degree of pollution control of coking enterprises in different regions is different. In order to more accurately predict the emission of pollutants from enterprises, this embodiment proposes a method for quantifying pollutant emission standards for coking enterprises to quantify the degree of emission control of coking enterprises in different regions. For the quantification of standards, it is necessary to find the places where the standards are the same for comparative analysis, so as to quantify. Both local and national standards include emission concentration limits for three types of atmospheric pollutants: particulate matter, sulfur dioxide and nitrogen oxides from coke oven chimneys, and this limit is an indicator that directly affects the coke production of coking enterprises. Therefore, the comprehensive quantitative results of the pollutant emission standards of enterprises can be obtained in the following ways:

[0040] Step 1: Obtain multiple pollutant emission standards implemented within the target area. Specifically, the pollutant emission standards implemented by coking enterprises can be obtained through the public end of the National Pollutant Discharge Permit Platform. Generally speaking, the national standard used by coking enterprises is the Coking Chemical Industry Pollutant Emission Standard GB16171-2012, issued in 2012. Some provinces have issued a series of more stringent local standards to further strengthen the management and control of coking enterprises and reduce pollutant emissions. Figure 2 The development process of pollutant emission standards for coking enterprises across the country is shown. Table 2 summarizes the emission concentration limits of particulate matter, sulfur dioxide and nitrogen oxides in some standards.

[0041] Table 2

[0042]

[0043] Step 2: Select one of the multiple pollutant emission standards as a reference standard A, and use the emission concentration limit of each pollutant in the reference standard as the quantitative reference value of each pollutant emission concentration; based on each reference value, quantify the emission concentration limit in each pollutant emission standard. Optionally, for any pollutant emission standard B other than the reference standard, the emission concentration limit of standard B can be quantified using the following formula:

[0044]

[0045] Among them, Q is the result of standard B quantization, E i is the emission concentration limit for the i-th pollutant in Standard B, R i is the quantitative reference value for the i-th pollutant (i.e., the emission concentration limit for the i-th pollutant in reference standard A), and n is the number of pollutant types.

[0046] Performing the above operations for all standards other than the reference standard will yield a quantitative representation for each standard. For example, referring to Table 2, the Pollutant Emission Standard for the Coking Chemical Industry (GB 16171-2012) divides standards into three categories based on coke oven type and region. In this example, the standard for mechanical coke and semi-coke ovens is set to 1. Based on the quantification method, the quantified result for heat recovery coke ovens is 1.33, and the quantified result for the special emission limits for atmospheric pollutants is 0.47. The quantified results for the remaining standards are shown in Table 2.

[0047] Step 3. Determine the comprehensive quantitative results of the pollutant emission standards of multiple coking enterprises in the target area based on the quantitative expression results, start and end years, and control areas of each pollutant emission standard. Due to the differences in the start and end years of coking enterprises and the differences in regions, different coking enterprises may experience different pollutant emission standards. The pollutant emission limits specified in the pollutant emission standards are regulations that enterprises must strictly implement, which will have a great constraint on enterprises. Therefore, this embodiment calculates a comprehensive quantitative result of the pollutant emission standards for each enterprise throughout the production cycle (referred to as the comprehensive quantitative result of the pollutant emission standards) to characterize the degree of control over soil pollutant emissions by enterprises. Table 3 summarizes the time nodes and regulations of the pollutant emission standards implemented by coking enterprises across the country.

[0048] Table 3

[0049]

[0050] Based on the above, for any coking enterprise in the target area, the comprehensive quantitative result Qc of the pollutant emission standard for the entire production cycle of the enterprise can be obtained according to the following formula:

[0051]

[0052] Among them, Q j is the quantitative result of the jth pollutant emission standard implemented by the enterprise in the production cycle, x j is the number of years the standard has been implemented, and m is the number of pollutant emission standards implemented by the enterprise during its production cycle. It can be seen that the comprehensive quantitative results of pollutant emission standards for coking enterprises throughout their production cycle are closely related to the production duration.

[0053] In addition, the coke production index can be obtained in the following ways:

[0054] Step 1: Divide the multiple coking enterprises in the target area into one group of enterprises with known SO2 emissions and another group of enterprises with unknown SO2 emissions. For ease of distinction and description, the two groups of enterprises are referred to as the first enterprise and the second enterprise, respectively.

[0055] Step 2: Extract the enterprise area, pollutant emission standard quantification results, scale and establishment time of all enterprises, as well as the SO2 emissions of the first enterprise. Optionally, the enterprise name and longitude and latitude information can be obtained through the network map POI, Tianyancha and Qichacha can obtain the enterprise's longitude and latitude information, enterprise name, starting year and the number of insured persons, and then interpret the area of the coking enterprise through remote sensing, and judge the scale information by the number of insured persons. The National Pollutant Discharge Permit Management Information Platform can obtain the pollutant emission standards and SO2 permitted annual emission limits implemented by some coking enterprises. Since enterprises will strictly implement pollutant emission limits, this embodiment uses the SO2 permitted annual emission limit as the actual SO2 emissions of coking enterprises. Finally, through the mutual complementation of the four platforms, the complete information of 719 coking enterprises including enterprise name, longitude and latitude, enterprise area, scale, starting year and implemented pollutant emission standards was integrated. 361 coking enterprises with actual SO2 emissions were obtained. The scale judgment standard is based on the industrial classification standards in the "Statistical Classification Method for Large, Medium, Small and Micro Enterprises (2017)". Enterprises with more than 1,000 employees are defined as large-scale enterprises, those with 300 to 1,000 employees are defined as medium-scale enterprises, those with less than 300 but more than 20 employees are defined as small-scale enterprises, and those with less than 20 employees are defined as micro-scale enterprises. The collected number of insured persons in enterprises is used to judge the scale of enterprises.

[0056] Step 2: Using the SO2 emission information of each first enterprise as the dependent variable and the enterprise area, pollutant emission standard quantification results, scale and establishment time of each first enterprise as the independent variables, fit the SO2 emission prediction model. As mentioned above, a total of 361 coking enterprises in production with actual SO2 emissions in 2020 were collected through the public end of the National Pollutant Discharge Permit Management Information Platform. Among them, the coefficient of variation of SO2 emissions and area is relatively large. After comparing the area and SO2 emissions of coking enterprises, 17 outlier points with a large ratio between the two were eliminated from the sample points (the ratio exceeded 10,000), and the number of sample enterprises involved in the establishment of the machine learning regression prediction method was 344. The sample distribution of coking enterprises is relatively uniform, and there will be no situation where the quantification results of the pollutant emission standards of coking enterprises are relatively concentrated, which is suitable for establishing a regression prediction model. Specifically, the Pearson correlation coefficient analysis was performed between the SO2 emissions of the 344 coking sample enterprises after eliminating the outliers and the four influencing factors (enterprise area, pollutant emission standard quantification results, scale and establishment time). The results show that the four influencing factors are significantly correlated with the SO2 emissions of coking enterprises (passed the 10% significance level test). The total permitted SO2 emission information of 344 coking enterprises after removing outliers was crawled from the National Pollutant Discharge Permit Management Information Platform, combined with the area, scale, and starting year of the coking enterprises obtained through artificial remote sensing interpretation and network crawling, and the quantitative factors of coking pollutant emission standards were introduced to establish random forest regression (RFR) and support vector machine regression (SVR) prediction methods, and the root mean square error (RMSE) and R after 10-fold cross validation were used to predict the SO2 emissions of coking enterprises. 2 To determine the accuracy of the prediction, a grid search method was used to select the optimal parameters for SVR and RFR. Finally, a 100-tree random forest regression model was selected as the SO2 emission prediction model.

[0057] Step 3: Substitute the enterprise area, pollutant emission standard quantification results, scale, and establishment time of each second enterprise into the SO2 emissions prediction model to obtain the SO2 emissions of each second enterprise. That is, a 100-tree random forest regression model is used to predict the SO2 emissions of 358 coking enterprises in operation nationwide whose actual SO2 emissions in 2020 are unknown.

[0058] Step 4: Decompose the coke production of each region in the target area according to the SO2 emissions of each enterprise to obtain the coke production of each enterprise. Specifically, based on the actual sulfur dioxide emissions of coking enterprises nationwide in 2020, and taking into account the coke production of each province, the output decomposition method is used to calculate the coke production distribution of each production enterprise in proportion. Optionally, the decomposition principle is: the ratio of coke production of each enterprise in a province is equal to the ratio of sulfur dioxide emissions, and the coke production of a province is decomposed to each enterprise in the province based on this ratio.

[0059] S120: Calculate the correlation between each indicator in the pollution source information and natural environment information and the soil pollutant content, and retain the indicators with the highest correlation.

[0060] This step uses the 13 indicators of the coking enterprise sample data set with outliers removed to perform Pearson correlation analysis. The results are as follows: Figure 3 As shown in the figure, ***, **, and * represent significance levels of 1%, 5%, and 10%, respectively. Indicator 1 represents soil benzo[a]pyrene content, and indicators 2-13 represent the 12 indicators in Table 1. Ten indicators passed the 10% significance test for soil benzo[a]pyrene content in coking enterprises. The strongest correlation was with the comprehensive quantitative results of pollutant emission standards, with a correlation coefficient of 0.534. Production hours and output ranked second and third, with correlation coefficients of 0.503 and 0.466, respectively. Correlations with land area, number of employees, and scale were all greater than 0.3. Benzo[a]pyrene content showed a negative correlation with sunshine duration and ground hardening, indicating that these two indicators have an inhibitory effect on soil benzo[a]pyrene content in coking enterprises. Average annual rainfall and average annual temperature did not pass the significance test, but both had negative correlation coefficients.

[0061] Grey correlation analysis was performed on 190 coking enterprise data for 12 evaluation items (land area, establishment time, production time, ground hardening, number of employees, comprehensive quantitative results of pollutant emission standards, output, scale, number of environmental violations, average annual sunshine, average annual rainfall, and average annual temperature). Taking benzo[a]pyrene as the "reference value" (parent sequence), the correlation between the 12 evaluation items and benzo[a]pyrene was studied, and based on the correlation, an analysis reference was provided for benzo[a]pyrene in the soil of coking enterprises. Optionally, when using grey correlation analysis in this embodiment, initialization was first performed, the resolution coefficient was taken as 0.5, and the correlation coefficient calculation formula was combined to calculate the correlation value, and based on the correlation value, the correlation value was then calculated for evaluation and judgment. The correlation degree indicates the degree of similarity between each evaluation item and the "reference value" (parent sequence). It is calculated by averaging the correlation coefficients. Correlation values range from 0 to 1. A higher correlation value indicates a stronger correlation between the evaluation item and the "reference value" (parent sequence). A higher correlation degree indicates a closer relationship between the evaluation item and the "reference value" (parent sequence), and thus a higher evaluation. Using the correlation values, a comprehensive ranking of 12 relevant evaluation indicators for benzo[a]pyrene content in coking enterprises' soils was performed, as shown in Table 4. Among them, the comprehensive quantitative results of pollutant emission standards received the highest evaluation (correlation degree: 0.912), followed by output (correlation degree: 0.906) and production duration (i.e., production time in Table 4) (0.899). The number of environmental violations ranked last (correlation degree: 0.864).

[0062] Table 4

[0063]

[0064] It can be seen that the correlation between the 12 indicators and the soil pollutant content is greater than the set threshold (for example, 0.7), so all 12 indicators will be included in the subsequent modeling scope.

[0065] S130. Using the multiple indicators as independent variables and the soil pollutant content as the dependent variable, fit multiple soil pollutant content prediction models, and optimize the parameters of each model using a particle swarm algorithm.

[0066] This step builds a machine learning prediction model based on the above 13 indicators. Optionally, five machine learning regression models can be constructed to predict the benzo[a]pyrene content in the soil of coking enterprises, namely decision tree regression model, random forest regression (RFR) model, support vector machine regression (SVR) model, extreme gradient boosting (XGBoost) regression model and AdaBoost regression model. 2 , to select the optimal machine learning prediction model.

[0067] Among them, the decision tree model is an algorithm based on inductive learning. During the model construction process, branch attributes are selected from top to bottom to divide the sample set. Decision tree regression is a regression method with a binary tree structure, consisting of internal nodes and leaf nodes. The internal nodes represent the feature attributes used for regression operations, and the leaf nodes represent the values obtained by regression. The training samples are trained starting from the root node in the internal nodes, and the samples are divided into their secondary nodes according to the training results. At this time, each secondary node represents the specified value of a certain feature. The samples are trained recursively in sequence until all samples reach the leaf nodes. The regression tree divides the feature space into several units, each of which has a feature output. The decision regression tree uses the least squares method to determine the best sample splitting node and construct a binary decision tree:

[0068]

[0069] The above equation is used to find the partitioning scenario that minimizes the sum of squared errors between the two regions. j is the jth feature, s is the value of the jth feature, and C1 and C2 are the fixed output values in the two regions after the partitioning. By traversing all variables j and the optimal split point s, repeating the above partitioning process for each region until the set stopping condition is met, a decision regression tree is generated.

[0070] Support vector regression (SVR) is a type of machine learning method. Its basic idea is to use the minimization principle of structural risk control to obtain patterns and information that link descriptors with desired targets. It is particularly suitable for complex nonlinear regression. It has two important parameters, kernel function and penalty coefficient. The kernel function of SVR maps linearly inseparable samples in low-dimensional space to high-dimensional space to make them linearly separable. This embodiment uses Gaussian kernel, linear kernel, polynomial kernel and tanh kernel to perform SVR analysis to compare R 2 The appropriate kernel function is selected based on the value and mean absolute error rate. At the same time, this embodiment selects the optimal penalty coefficients of different kernel functions to construct the model method.

[0071] Random forest is an ensemble algorithm that uses a large set of regression trees to randomly select predictors for predictive analysis when solving regression problems. It has the characteristics of high predictive performance, low correlation between individual trees, and small bias. The number of trees and the randomly selected predictors are the two main parameters of the random forest algorithm. The random forest regression prediction result is determined by the average value of the set of decision trees. In this embodiment, the model selection does not limit the depth of the trees.

[0072] The XGBoost regression gradient boosting decision tree is an improvement on the GBDT (Gradient Boosting Decision Tree) algorithm. The gradient boosting algorithm trains newly added weak classifiers based on the negative gradient of the model's loss function. After training, the Nth weak classifier is cumulatively combined with the first N-1 weak classifiers. Unlike traditional gradient boosting algorithms, the XGBoost algorithm uses a second-order expansion of the loss function after Taylor expansion, and incorporates a regularization term to control model complexity. This second-order expansion is faster in finding optimal parameters than the first-order derivative gradient boosting in GBDT.

[0073] AdaBoost regression is a method that improves the accuracy of a given learning algorithm by combining hypotheses created by the learning algorithm. This algorithm is a further derivative of the decision tree algorithm. In AdaBoost, each decision tree is a weak learner. The basic principle is to rationally combine weak learners to generate a strong learner. AdaBoost uses an iterative approach, training only one weak learner at a time. Then, based on the error between the training results of the weak learner and the true value, the next iteration is entered, and the next weak learner is trained based on the original weak learner. The training process of the Nth weak learner focuses more on data that was misclassified or had large regression errors in the first N-1 weak learners. By reducing the error of individual learners each time and increasing the weight of learners with good learning results, the AdaBoost model is eventually integrated.

[0074] Furthermore, the model evaluation method of this embodiment uses a holdout method, which directly divides the dataset into two mutually exclusive sets, 70% of which serves as the training set and 30% as the validation set. After removing outliers, there are 190 coking enterprise sample data, of which 133 serve as the training set and 57 serve as the validation set. Machine learning algorithms can exhibit significant differences in model performance due to different parameter configurations. Therefore, parameter adjustments are necessary during model performance evaluation and selection to achieve optimal model performance. Particle swarm optimization, which starts from a random solution and continuously iterates to find the optimal solution, is characterized by fast convergence and high accuracy.

[0075] In addition, this embodiment utilizes a particle swarm optimization algorithm to optimize the parameters of five machine learning regression methods. The particle swarm optimization algorithm is a hyperparameter optimization algorithm for machine learning models based on the predation behavior of bird flocks. The particle swarm optimization algorithm simulates birds in a flock by designing a group of particles. Each particle is assigned two attributes: speed and position. Speed represents the speed of movement, and position represents the direction of movement. Each particle moves in space and records the extreme value it finds. The optimal individual extreme value among all particles is regarded as the "global optimal solution" for the entire particle swarm. If the "global optimal solution" obtained through weak search does not meet the set conditions, the speed and position of each particle are updated, and finally a "global optimal solution" that meets the conditions is obtained. The particle swarm optimization algorithm does not perform crossover operations during the search for the optimal solution, relying on particle speed to complete the search. During the iterative search, particles only transmit their own optimal solutions to other particles, resulting in a faster search speed. It also requires fewer parameters to adjust, has a simple structure, and is easy to implement in engineering.

[0076] Specifically, the particle swarm algorithm process mainly includes setting the particle swarm size, position and speed, setting the inertia weight, maximum speed, acceleration constant, maximum number of iterations, etc. This section sets the initialization particle swarm size to 20, the inertia weight w to 0.5, the acceleration constants c1 and c2 to 2, the maximum number of iterations to 1000, and the incremental threshold alpha of the optimal fitness value to 0.0001.

[0077] There are three main optimization parameters for decision tree regression: min simple split, max depth, and ccp alpha, all of which are pruning parameters used to limit tree overfitting. The min simple split specifies that if the sample size of an internal node is less than this value, no further splits will be made; the max depth specifies that all values exceeding the specified depth will be pruned; and the larger the ccp alpha, the more nodes will be pruned. Using the particle swarm optimization algorithm, the optimal decision tree regression model was determined to be achieved when the min simple split was 5, the max depth was 12, and the ccp alpha was 0.9899. Other parameters were min sample leaf = 1, criterion = mse, and min weight fraction leaf = 0.

[0078] Random forest regression is a combination of multiple decision trees. This example primarily optimizes three parameters: nestimators represents the number of decision trees; max depth and ccp alpha are the same as for decision trees. The optimization results show that the optimal solutions for these parameters are 263, 50, and 0.465. Other parameters are set to 5 for the minimum simple split, 1 for the minimum sample leaf, 1 for the criterion, and 0 for the minimum weight fraction leaf.

[0079] Support vector machine regression (SVR) is a process that uses nonlinear mapping to map data into a high-dimensional space so that the independent and dependent variables have good linear regression characteristics, and then returns to the original space after fitting. Using the particle swarm optimization algorithm, the polynomial kernel function is the optimal kernel function. It has three main parameters that need to be tuned, where degree is the order of the polynomial kernel function, gamma is the kernel function coefficient, and C is the penalty coefficient. In this example, the three optimal parameters are 9, 0, and 0.988, respectively.

[0080] AdaBoost regression generates a strong learner by assigning high weights to learners with low error rates and low weights to learners with high error rates. Two parameters require optimization: n estimators, which represents the number of learners. Increasing the number improves fitting ability but increases computation time and the probability of overfitting; and learningrate, which controls the degree to which each learner corrects errors from the previous learning period. In this example, the optimal parameters for AdaBoost regression are n estimators of 50 and learning rate of 0.

[0081] XGBoost regression is an efficient implementation of gradient boosting tree (GBDT) regression, which adds a regularization term to the loss function. XGBoost uses the second-order Taylor expansion of the loss function as the loss function fit, which solves the problem that some loss functions are difficult to calculate derivatives. This embodiment adjusts five parameters, including n estimators, max depth, and regalpha. Among them, n estimators represents the number of base learners, max depth represents the maximum depth of the tree, and regalpha is the regularization constant. The optimal parameters of the three are 773, 50, and 0.2924 found by the particle swarm optimization algorithm. The other two parameters are learning rate, whose size can control the convergence speed and find the optimal value; and reglambda regularization term. The two optimal parameters are 0.1 and 1. In addition, this study uses gbtree as the base learner of XGBoost regression.

[0082] The five best machine learning regression models constructed using the training set are compared with the goodness of fit R of the five machine learning regression models in the test set. 2 It was found that the best prediction effect of benzo[a]pyrene in soil of coking enterprises was achieved by using random Senli regression. 2 The maximum value was 0.7714. The prediction performance rankings for the other four machine learning regression algorithms were XGBoost regression, AdaBoost regression, decision tree regression, and support vector machine regression. By comparing the predicted results of benzo[a]pyrene content in soil from coking enterprises in the validation set with the actual values, we were able to determine the accuracy of each machine learning regression model in predicting excess levels.

[0083] After the optimal model is determined, the importance of each indicator in the optimal model can also be analyzed. Among them, feature importance ranking is one of the methods of machine learning model interpretation. This embodiment mainly uses the random forest feature importance method to calculate the importance of factors affecting the content of benzo[a]pyrene in the soil of coking enterprises. Among them, the most important factor is the comprehensive quantitative result of the pollutant emission standard, with an importance of 24.4%, followed by production time and output, with importance of 0.8% and 15% respectively. The importance of establishment time also exceeds 10%. The three natural factors of average annual rainfall, average annual temperature and average annual sunshine are relatively not important, at less than 4%. In addition, the relative influence of ground hardening and the number of environmental violations on the content of benzo[a]pyrene in the soil of coking enterprises is about 4%, and the influence of the number of employees is 5.7%.

[0084] Furthermore, if the importance of a particular indicator falls below a certain threshold and the number of possible values for the indicator falls below another threshold, the possible values for that indicator can be refined. For example, categorical variables such as scale and ground hardening have limited contributions. This is because the method constructed in this paper is a regression method, and for categorical variables, the limited number of possible values makes it difficult to effectively improve the prediction accuracy of the regression model. This demonstrates that the possible values for this variable can be refined, such as reducing ground hardening from two values of 0 / 1 to four, thereby enhancing its contribution to the regression model.

[0085] S140: Substitute the multiple indicators of the multiple coking enterprises in operation in the target area into the optimal model among the multiple soil pollutant content prediction models to predict whether the soil pollutants in each coking enterprise in operation exceed the standard.

[0086] Continuing with the above-mentioned specific implementation example, the optimal random forest regression model and parameters selected above were used to predict the current (2020) and next 20-year excess levels of benzo[a]pyrene in the soil of coking plants operating nationwide. A benzo[a]pyrene content in the soil of a coking plant exceeding the screening value of 1.5 mg / kg was considered to exceed the standard. When predicting the future benzo[a]pyrene content in the soil of coking plants, the future production hours, the comprehensive quantitative results of the pollutant emission standards, the annual average sunshine, rainfall, and temperature were changed over time, while other indicators were assumed to remain unchanged. The sunshine hours, rainfall, and temperature from 2021 to 2040 were calculated based on 2020, and the annual averages were recalculated. The comprehensive quantitative results of the pollutant emission standards changed according to the time provisions of the pollutant emission standards implemented by the enterprise. In summary, the future changes in the predictive factors for benzo[a]pyrene content in the soil of coking plants are shown in Table 5.

[0087] Table 5

[0088]

[0089] The above-mentioned whole method process can also be combined with Figure 4 Based on the above method and steps, by changing the prediction time node to 2025 and collecting data for the corresponding year, it is possible to predict the soil pollutant exceeding the standard in coking enterprises in operation in 2025 and beyond. The prediction results can provide data support for the decision-making of soil pollution control priorities and control measures in the target area now and in the future.

[0090] To better illustrate the method provided in this embodiment, the reasons for selecting the above 12 indicators as factors affecting soil pollutant content are explained below:

[0091] 1. Enterprise area: The area occupied by a coking enterprise may directly affect the degree of soil pollution. The larger the area, the more possible pollution sources there are and the more complicated the pollution situation.

[0092] 2. Establishment Date: A company's establishment date may reflect its historical accumulation of production technology and environmental protection measures. Companies established earlier may use more backward production technology, while companies established later may adopt more environmentally friendly production methods.

[0093] 3. Production duration: The length of production may affect the cumulative pollutant emissions. Long-term continuous production may increase the risk of soil contamination.

[0094] 4. Ground hardening: The degree of ground hardening affects the penetration and diffusion of soil pollutants. Hardened ground may reduce the penetration of pollutants into the deep soil, but it may also lead to the accumulation of surface pollutants.

[0095] 5. Number of employees: The number of employees may be related to the scale of the enterprise and the intensity of production activities, thereby indirectly affecting the degree of soil pollution.

[0096] 6. Comprehensive quantitative results of pollutant emission standards: This indicator comprehensively reflects the control role of pollutant emission standards in the entire production cycle of an enterprise and is a comprehensive indicator that affects pollutant generation.

[0097] 7. Scale: Enterprise scale is closely related to factors such as output and number of employees. The number of employees is an important indicator for assessing an enterprise's environmental performance.

[0098] 8. Output: Output reflects the production scale of an enterprise. The larger the output, the larger the production scale of the enterprise, which may mean more raw material input and waste output, which may increase the risk of soil pollution.

[0099] 9. Number of Environmental Violations: The number of environmental violations directly reflects a company's compliance with environmental protection. Companies that repeatedly violate regulations may lack effective environmental and pollution control measures.

[0100] 10. Historical sunshine duration: Sunshine duration may affect photochemical reactions and the degradation of pollutants in the soil. Sufficient sunlight may help reduce the levels of certain pollutants.

[0101] 11. Historical rainfall: Rainfall directly affects soil moisture and the leaching of pollutants. High rainfall may accelerate the migration and dispersion of these pollutants.

[0102] 12. Historical Temperatures: Temperature is a key factor affecting degradation rates. Biochemical reactions occur in the soil. Suitable temperatures may promote the degradation of pollutants, while excessively high or low temperatures may inhibit this process.

[0103] In summary, this example provides a method for predicting soil pollutant levels exceeding standards at a coking plant. By collecting internal enterprise information and external natural environmental information, combined with the pollutant emission standards and production data implemented by the enterprise, this method predicts the current and future benzo[a]pyrene content in the soil of the coking plant in the target area and analyzes the importance of each influencing factor. This method can achieve the following beneficial effects:

[0104] 1. Compared with the prediction of air pollutant concentrations, the formation process of soil pollution in construction land is extremely complex, involving many influencing factors and has a long-term cumulative effect, which greatly increases the difficulty of prediction. This embodiment introduces a new indicator based on the concept of soil pollutant accumulation-the comprehensive quantitative result of pollutant emission standards. This indicator can more comprehensively describe the control level of coking enterprises during production and improve the accuracy of predictions. At the same time, this embodiment also uses remote sensing technology to determine whether the ground of the coking enterprise has been hardened-this is a key indicator to prevent pollutants from seeping into the soil after release. In addition, this embodiment also systematically collects environmental violation records of various enterprises-this indicator not only reveals abnormal emissions, but also reflects the environmental control capabilities of enterprises.

[0105] 2. Machine learning (ML) methods have demonstrated outstanding performance in the challenging task of predicting soil pollutant content on large-scale construction land. However, the predictive performance of ML models varies significantly under different parameter configurations, a key issue often overlooked in many studies. This example comprehensively and in-depth analyzes the predictive performance of ML models under various parameter combinations by introducing the dynamic parameter adjustment process of the particle swarm optimization (PSO) algorithm. The results show that the difference in model accuracy under different parameter configurations can reach 35% or even higher, highlighting the important role of parameter optimization in improving model performance. The PSO algorithm, with its unique swarm intelligence properties, flexibly adjusts the position and velocity of particles within the parameter search space, achieving dynamic optimization of ML model parameters, thereby ensuring efficient and accurate search for the optimal solution for each ML model. This process not only significantly improves the accuracy of machine learning predictions, but also greatly enhances the efficiency and practicality of predictions. By combining the particle swarm optimization algorithm (PSO), this example constructs an efficient and accurate machine learning model, aiming to provide strong technical support and guarantee for the prediction of soil pollutant content in enterprises.

[0106] 3. In order to solve the problem of model accuracy and robustness, this example introduces the Sigma principle to reduce the variability of samples, thus laying a solid foundation for subsequent model construction. Among the four machine learning regression models developed, the random forest model performed outstandingly, especially in predicting the concentration of benzopyrene in the soil of the coking enterprise. Its R 2The value reaches 0.771, and the root mean square error is 2.1.

[0107] 4. This example analyzes the importance of various influencing factors. Among them, the comprehensive quantitative results of pollutant emission standards contribute most significantly to the prediction of soil pollutant levels. A decrease in the comprehensive quantitative results of pollutant emission standards indicates that stricter pollutant emission standards have been implemented in the target area, effectively limiting pollutant emissions from enterprises. This measure significantly improves the environmental management and control capabilities of enterprises and effectively controls the accumulation of soil pollutants. In particular, because the implementation requirements of emission standards are divided by region, the quantitative results of emission standards in the same region are often similar. Therefore, the comprehensive quantitative results of pollutant emission standards are not applicable to small areas such as cities or counties, but are particularly suitable for large-scale regional soil pollution prediction, such as the national scale.

[0108] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the device includes a processor 60, a memory 61, an input device 62 and an output device 63; the number of processors 60 in the device can be one or more. Figure 5 In the embodiment, a processor 60 is used as an example; the processor 60, the memory 61, the input device 62 and the output device 63 in the device can be connected by a bus or other means. Figure 5 The bus connection is taken as an example.

[0109] Memory 61, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the method for predicting excessive soil pollutants in an operating coking enterprise, as described in the embodiments of the present invention. Processor 60 executes the software programs, instructions, and modules stored in memory 61 to perform various functional applications and data processing of the device, thereby implementing the aforementioned method for predicting excessive soil pollutants in an operating coking enterprise.

[0110] The memory 61 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal. Furthermore, the memory 61 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory 61 may further include memory remotely located relative to the processor 60, and these remote memories may be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0111] The input device 62 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 63 may include a display device such as a display screen.

[0112] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for predicting excessive soil pollutants in an in-production coking enterprise according to any embodiment is implemented.

[0113] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by an instruction execution system, device or device or used in combination with it.

[0114] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0115] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0116] Computer program code for performing the operations of the present invention can be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting excessive soil pollutants in a coking enterprise, characterized in that: include: Obtaining soil pollutant content, pollution source information, and natural environment information for multiple coking enterprises within a target area, wherein the target area includes multiple regions; Calculating the correlation between various indicators in the pollution source information and natural environment information and the soil pollutant content, and retaining multiple indicators with the highest correlation; Using the multiple indicators as independent variables and the soil pollutant content as the dependent variable, fitting multiple soil pollutant content prediction models, and using the particle swarm algorithm to optimize the parameters of each model; Substituting the multiple indicators of the multiple coking enterprises in operation in the target area into the optimal model among the multiple soil pollutant content prediction models to predict whether the soil pollutants in each coking enterprise in operation exceed the standard; Among them, the various indicators in the pollution source information include: enterprise area, establishment time, production time, ground hardening conditions, number of employees, comprehensive quantitative results of pollutant emission standards, scale, coke production and environmental accident violation record information.

2. The method according to claim 1, characterized in that The acquisition of soil pollutant content, pollution source information, and natural environment information of multiple coking enterprises in the target area includes: Obtain multiple pollutant emission standards implemented in the target area; Selecting a reference standard from the plurality of pollutant emission standards, and using the emission concentration limit of each pollutant in the reference standard as a quantitative reference value of the emission concentration of each pollutant; According to the reference values, the emission concentration limits in the emission standards of each pollutant are quantified; Based on the quantitative expression results, start and end years and control areas of each pollutant emission standard, the comprehensive quantitative results of the pollutant emission standards of multiple coking enterprises in the target area are determined.

3. The method according to claim 2, characterized in that The quantitative expression of the emission concentration limit values in each pollutant emission standard according to each reference value includes: quantitative expression of the emission concentration limit values in any pollutant emission standard by the following formula: Among them, Q is the result after quantization, E i is the emission concentration limit of the i-th pollutant, R i is the quantitative reference value of the emission concentration of the i-th pollutant, and n is the number of pollutant types; Accordingly, determining the comprehensive quantitative results of pollutant emission standards for multiple coking enterprises within the target area based on the quantitative expression results, start and end years, and control areas of each pollutant emission standard includes: determining the comprehensive quantitative results of pollutant emission standards for any coking enterprise within the target area according to the following formula: Among them, Q C It is the comprehensive quantitative result of the pollutant emission standards for the entire production cycle of the enterprise, Q j is the quantitative expression result of the jth pollutant emission standard implemented by the enterprise in the production cycle, x j is the length of years for the implementation of the jth pollutant emission standard, and m is the number of pollutant emission standards implemented by the enterprise during its production cycle.

4. The method according to claim 1, wherein The acquisition of soil pollutant content, pollution source information, and natural environment information of multiple coking enterprises in the target area includes: Divide multiple coking enterprises in the target area into a first enterprise with known SO2 emissions and a second enterprise with unknown SO2 emissions; The SO2 emission prediction model was fitted using the SO2 emission information of each first-tier enterprise as the dependent variable and the enterprise area, pollutant emission standard quantification results, scale and establishment time of each first-tier enterprise as the independent variables; Substituting the enterprise area, pollutant emission standard quantification results, scale and establishment time of each second enterprise into the SO2 emission prediction model to obtain the SO2 emission of each second enterprise; The coke production of each region in the target area is decomposed according to the SO2 emission of each enterprise to obtain the coke production of each enterprise.

5. The method according to claim 1, wherein The multiple indicators are used as independent variables and the soil pollutant content is used as the dependent variable to fit multiple soil pollutant content prediction models, including: Taking the multiple indicators as independent variables and the soil pollutant content as the dependent variable, decision tree regression, random forest regression, support vector machine regression, extreme gradient boosting regression and AdaBoost regression were performed respectively to obtain multiple soil pollutant content prediction models; By comparing the R 2 , select the optimal model.

6. The method according to claim 1, characterized in that The acquisition of soil pollutant content, pollution source information, and natural environment information of multiple coking enterprises in the target area includes: Obtain soil pollutant content from multiple coking enterprises within the target area to form a sample set; Determining whether there are outliers based on the coefficient of variation of the sample set; If there are outliers, fitting the data distribution of the sample set; If the data distribution is positively skewed compared to the normal distribution, the soil pollutant content of abnormal enterprises that is greater than the mean plus N times the standard deviation will be eliminated from the sample set; If the data distribution is negatively skewed compared to the normal distribution, the soil pollutant content of abnormal enterprises that is less than the mean value minus N times the standard deviation will be eliminated from the sample set; Wherein, N is a natural number.

7. The method according to claim 1, characterized in that After fitting multiple soil pollutant content prediction models using the multiple indicators as independent variables and the soil pollutant content as the dependent variable, and optimizing the parameters of each model using the particle swarm algorithm, the method further includes: determining an optimal model among the plurality of soil pollutant content prediction models; Performing importance analysis on various indicators in the optimal model; If the importance of a certain indicator is lower than the first threshold and the number of indicator values is less than the second threshold, the value range of the certain indicator is refined.

8. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for predicting excessive soil pollutants in an in-production coking enterprise as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, which, when executed by a processor, implements the method for predicting excessive soil pollutants in an in-production coking enterprise as described in any one of claims 1-7.

Images