An optimization method and system for hazardous waste landfill groundwater monitoring project based on artificial intelligence

Abstract

The application discloses an optimization method for a hazardous waste landfill site groundwater monitoring project based on artificial intelligence. The method comprises the following steps: performing classification numerical coding preprocessing on groundwater monitoring well data, constructing an artificial intelligence model for predicting an over-standard state of the groundwater monitoring well, combining the artificial intelligence model to obtain an importance value and a normalized importance value of the over-standard state of the groundwater monitoring well, and determining important indexes of groundwater monitoring. A person skilled in the art can perform pollution monitoring on the hazardous waste landfill site groundwater based on the indexes. The application can quickly and efficiently optimize the hazardous waste landfill site groundwater monitoring project, reduce monitoring costs, improve monitoring efficiency, improve the implementability of a long-term groundwater monitoring system, and provide support for later management of a hazardous waste landfill pollution site.

Description

TECHNICAL FIELD

[0001] The present application relates to groundwater monitoring, in particular to an optimization method and system for hazardous waste landfill groundwater monitoring project based on artificial intelligence. BACKGROUND

[0002] Hazardous waste landfill is an important means and facility for the safe disposal of hazardous waste. Heavy metals, organic pollutants and persistent toxic substances contained in hazardous waste can migrate to soil and groundwater environment through leakage, causing soil and groundwater pollution.

[0003] Once the soil and groundwater of the hazardous waste landfill site are contaminated, the whole process risk control should be implemented, including investigation, risk assessment, engineering scheme preparation, engineering construction, effect evaluation, and post-environmental management. At present, the management and control of contaminated sites mainly focuses on investigation, risk assessment, simulation prediction, and effect evaluation. There are few studies on post-environmental management, especially long-term management after groundwater control. Due to the adoption of the risk control technology of "taking control of the source as the core and cutting off the exposure route as the basic strategy", pollutants in groundwater usually exist objectively and have significant migration and diffusion risks. Therefore, after the construction of the control project, the post-management of groundwater should be implemented to monitor the migration and diffusion risks of pollutants in groundwater for a long time. Long-term monitoring, as one of the core means of post-management of contaminated groundwater, has gradually been valued in the whole process of management and control of contaminated sites.

[0004] However, long-term groundwater monitoring is usually based on the fixed monitoring index list specified during the construction period and the effect evaluation period, and lacks adaptive adjustment and optimization mechanism for the dynamic response of pollutant migration rules, resulting in high cost and low efficiency of long-term monitoring by the pollution responsibility subject, which is difficult to meet the actual demands of the pollution responsibility subject, and finally may reduce the implementability of the long-term monitoring system.

[0005] At present, scholars at home and abroad have carried out many researches in the field of monitoring index optimization, such as the monitoring index optimization method based on principal component analysis, water quality comprehensive index, and entropy weight water quality index. However, these methods are mostly dependent on static data or single time node pollution characteristics, and it is difficult to fully tap the relevance of long-term monitoring data. In addition, the dynamic change of groundwater monitoring index is affected by many factors such as hydrogeological conditions and pollutant migration characteristics. These influencing factors often have complex nonlinear relationships, and conventional statistical methods are difficult to solve the dynamic optimization problem of long-term monitoring index. SUMMARY

[0006] To solve the above problems, the application provides an optimization method for hazardous waste landfill groundwater monitoring projects based on artificial intelligence. By classifying and numerically coding the data of the groundwater monitoring wells, an artificial intelligence model for predicting the exceeding standard state of the groundwater monitoring wells is constructed, the importance value and the normalized importance value affecting the exceeding standard state of the groundwater monitoring wells are obtained by combining the artificial intelligence model, and the important indicators for groundwater monitoring are determined. The person skilled in the art can perform pollution monitoring of the hazardous waste landfill groundwater based on the indicators.

[0007] The application provides the following technical scheme: an optimization method for hazardous waste landfill groundwater monitoring projects based on artificial intelligence, comprising the following steps:

[0008] (1) Collecting historical monitoring data of a plurality of groundwater monitoring wells to construct a groundwater monitoring data set; the groundwater monitoring data set comprises the pollution exceeding standard state of each monitoring well, the position of each monitoring well, monitoring projects and corresponding monitoring values, and the monitoring values are used to represent the numerical values or states of the monitoring projects;

[0009] (2) Taking the pollution exceeding standard state of the monitoring well as the target variable and the position and monitoring projects of the monitoring well as the independent variables, a plurality of decision tree models are constructed;

[0010] (3) Dividing the groundwater monitoring data set into a training set and a test set, training and testing the decision tree models by using the training set and the test set, and selecting the best decision tree model according to the model performance evaluation indicators, wherein the performance evaluation indicators of the decision tree model include the accuracy ACC, the precision PRE, the recall rate REC, and the harmonic mean F1 of the precision and the recall rate;

[0011] (4) Using the best decision tree model to determine the monitoring projects that have an important influence on the pollution exceeding standard of the groundwater monitoring well. When and only when the four performance evaluation indicators of one of the decision tree models are all optimal among all the decision tree models, the decision tree model is determined as the best model. The monitoring projects that have an important influence on the pollution exceeding standard of the groundwater monitoring well are determined by using the decision tree model.

[0012] In the embodiments of the application, the CART model, the CHAID model and the E-CHAID model are used. As the common knowledge in the art, CART is a binary decision tree technique that can be used for predicting classification or continuous variables. It constructs a tree structure by recursively splitting the data set until a certain stopping rule is met. For classification problems, Gini impurity is used; for regression problems, variance minimization is adopted. CHAID is also a decision tree technique, but unlike CART, it is multi-directional, meaning that each split can have more than two branches. It is based on the chi-square test to determine the best split point. E-CHAID is an extension or improved version of CHAID.

[0013] Further, the pollution exceeding standard state of the monitoring well is an unordered categorical variable, and after being coded as [0, 1] according to whether it exceeds the standard, the unordered categorical variable is input into the decision tree model as a target variable; if the pollution of the monitoring well exceeds the standard, the code is 1; if the pollution of the monitoring well does not exceed the standard, the code is 0.

[0014] Further, the monitoring items in the step 1 include key items and auxiliary items; the key items refer to characteristic toxic and harmful pollutants of the hazardous waste landfill; and the auxiliary items refer to internal and external environments where the characteristic toxic and harmful pollutants are located, including air temperature, pH, COD Mn , water temperature, turbidity, intensity level of smell and taste, presence or absence of visible matter, dissolved oxygen, water level, oxidation-reduction potential, Cl - content, Fe 3+ content, SO4 2- content, NO3 - content and the like.

[0015] Further, the monitoring well position is an unordered categorical variable, and according to the relative position in the underground water flow field, the unordered categorical variable is coded as [1, 2, 3]; if the monitoring well is located in the upstream, the code is 1; if the monitoring well is located in the middle reaches, the code is 2; and if the monitoring well is located in the downstream, the code is 3.

[0016] Further, the presence or absence of the visible matter is an unordered categorical variable, and according to the presence or absence, the unordered categorical variable is coded as [0, 1]; if there is visible matter, the code is 1; and if there is no visible matter, the code is 0.

[0017] In some preferred embodiments of the present application, for the monitoring items with direct monitoring values, a threshold judgment method is used for monitoring value coding, for example, for turbidity, high, medium and low judgment thresholds are set: high (>10 NTU), medium (3-10 NTU) and low (<3 NTU), and the high, medium and low monitoring values are coded as 3, 2 and 1 respectively; for example, for the intensity level of smell and taste, 6 levels are set: none, weak, weak, obvious, strong and very strong, and the monitoring values are coded as 0, 1, 2, 3, 4 and 5 respectively. It should be emphasized that the coding threshold and coding value setting can be set in sequence according to the actual situation, and the value size does not affect the calculation of the decision tree.

[0018] Further, the calculation method of the accuracy ACC, the precision PRE, the recall rate REC, the harmonic average F1 of the precision and the recall rate is as follows:

[0019] ACC=(TM0+TM1) / (TM0+TM1+FM0+FM1)

[0020] PRE1=(TM1) / (TM1+FM1), PRE0=(TM0) / (TM0+FM0);

[0021] REC1=(TM1) / (P1), REC0=(TM0) / (P0);

[0022] F1=(2*PRE*REC) / (PRE+REC);

[0023] Wherein: TM0 is the sample number of the monitoring well not exceeding the standard prediction; TM1 is the sample number of the monitoring well exceeding the standard prediction; FM0 is the sample number of the monitoring well not exceeding the standard prediction; FM1 is the sample number of the monitoring well exceeding the standard prediction; PRE1 represents the percentage of the sample actually exceeding the standard when the monitoring well predicts exceeding the standard; PRE0 represents the percentage of the sample actually not exceeding the standard when the monitoring well predicts not exceeding the standard; REC1 represents the percentage of the sample number of the monitoring well exceeding the standard prediction in the actual sample number; REC0 represents the percentage of the sample number of the monitoring well not exceeding the standard prediction in the actual sample number; P1 is the actual sample number of the monitoring well exceeding the standard; P0 is the actual sample number of the monitoring well not exceeding the standard.

[0024] Further, the data ratio of the training set and the test set is 7:3.

[0025] The application further provides a groundwater monitoring project optimization system, at least comprising:

[0026] The acquisition module acquires historical monitoring data of a plurality of groundwater monitoring wells.

[0027] The first processing module is used for pre-processing the groundwater monitoring well data and constructing a groundwater monitoring data set.

[0028] The second processing module is used for analyzing the groundwater monitoring data set and determining the monitoring project of the groundwater monitoring well pollution exceeding the standard.

[0029] The application further provides a computer readable storage medium, which stores computer instructions, and the instructions are executed by a processor to realize the above steps.

[0030] The application has the following beneficial effects: the application can quickly and efficiently realize optimization of the groundwater monitoring index of the hazardous waste landfill, reduce monitoring cost, improve monitoring efficiency, improve the implementability of the long-term groundwater monitoring system, and provide support for later management of the hazardous waste landfill pollution plot. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figures 1-3 The confusion matrixes of performance evaluation of the CART model, the CHAID model and the E-CHAID model are sequentially shown.

[0032] Figure 4 Importance analysis of input variables on the prediction ability of the CART model. DETAILED DESCRIPTION

[0033] In recent years, due to the powerful data mining and feature selection capabilities, decision tree (DT), support vector machine (SVM), neural network, random forest and the like as common supervised machine learning algorithms have shown application potential in the field of soil and groundwater pollution prevention and control.

[0034] The present application is found through comparative screening that the technical scenario of the present application has the advantages of strong interpretability and high calculation efficiency in judging the relevance of monitoring items compared with other algorithms.

[0035] It should be noted that the above-mentioned model is constructed based on the target variable and the independent variable, which is a common method in the art, and the present application will not be described.

[0036] In addition, the monitoring item described in the present application is a monitoring index in the art, which is described as "monitoring index" in the embodiments of the present application.

[0037] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, taking a hazardous waste landfill site in a certain province in the central region of China as an object, the intelligent optimization of the groundwater monitoring index of the hazardous waste landfill site is realized through the steps of the present application, and the specific implementation steps are as follows:

[0038] (1) Collecting the construction data of the hazardous waste landfill site in a certain province in the central region of China, including site selection, feasibility study, investigation, design, construction and completion acceptance, etc., analyzing in detail the pollution prevention and control situation in the operation process such as hidden trouble investigation, self-monitoring, pollution investigation, risk assessment, pollution control and effect evaluation, and mastering the information of pollution situation of the hazardous waste landfill site, hydrogeological conditions, monitoring well position, structure, monitoring index, etc.

[0039] (2) Obtaining the existing groundwater pollution monitoring related data, extracting the groundwater monitoring well data variables, including the pollution exceeding standard state of 73 monitoring wells, the position of 73 monitoring wells, the monitoring index and the corresponding monitoring value. The specific monitoring index includes 14 key items (fluoride, arsenic, cadmium, nickel, 1,2-dichloropropane, dichloromethane, cis-1,2-dichloroethylene, 1,2-dichloroethane, 1,2,4-trichlorobenzene, 1,2,3-trichlorobenzene, trichloromethane, dibromochloromethane, isophorone, petroleum hydrocarbon) and 13 auxiliary items (pH, COD Mn , water temperature, turbidity, smell and taste, visible matter, dissolved oxygen, water level, oxidation-reduction potential, Cl - , Fe 3+ , SO4 2- , NO3 -), and different types of variable data were implemented for classification preprocessing, unordered categorical variables (monitoring well location, monitoring well exceeding standard, visible matter), ordered categorical variables (turbidity, smell and taste), and continuous variables (fluoride, arsenic, cadmium, nickel, 1,2-dichloropropane, dichloromethane, cis-1,2-dichloroethylene, 1,2-dichloroethane, 1,2,4-trichlorobenzene, 1,2,3-trichlorobenzene, trichloromethane, dibromochloromethane, isophorone, petroleum hydrocarbon, pH, COD Mn , water temperature, dissolved oxygen, water level, oxidation-reduction potential, Cl - , Fe 3 + , SO4 2- , NO3 - ) were numerically coded according to the "Table 1 Feature Variable Preprocessing and Numerical Coding Rules" to construct the feature data set of the hazardous waste landfill groundwater monitoring index optimization decision.

[0040] Table 1 Feature Variable Preprocessing and Numerical Coding Rules

[0041]

[0042] (3) Using the constructed feature variable data set, taking "monitoring well exceeding standard" as the target variable and the remaining features as the independent variables, three different decision tree algorithms (CHAID, E-CHAID, CART) were used to build artificial intelligence models. CART model has a significant difference in the degree of full growth compared with CHAID and E-CHAID. The maximum tree depth of CART is 5, which grows more fully than CHAID and E-CHAID (the maximum tree depth is 3), which may be beneficial to the improvement of model prediction performance. Due to the more full growth of CART, the number of internal nodes and leaf nodes is more than that of CHAID and E-CHAID, and the number of internal nodes of CART, CHAID and E-CHAID is 5, 2 and 2, respectively, and the number of leaf nodes is 7, 4 and 4, respectively. The root node of CHAID is divided into two relatively pure left and right sub-trees with "1,2,4-trichlorobenzene" as the splitting variable with the largest χ 2 (232.75). The right sub-tree has been completely split and does not need to be further divided, becoming a leaf node; the left sub-tree is further divided into three "purity" higher leaf nodes by fluorine, petroleum hydrocarbon two splitting variables. The root node of E-CHAID-DT is divided into two relatively pure left and right sub-trees with "1,2,4-trichlorobenzene" as the splitting variable with the largest χ 2(237.34)The maximum split variable "1,2,4-trichlorobenzene" divides the data set into two relatively purer left and right sub-trees of "≤ detection limit to standard limit" and "> detection limit to standard limit", respectively. The right sub-tree has been completely split and does not need to be split again, becoming a leaf node; the left sub-tree is further divided into three "higher purity" leaf nodes by the fluoride and nickel split variables. The CART root node divides the data set into two relatively purer left and right sub-trees of "≤ detection limit to standard limit" and "> detection limit to standard limit" by the "impurity" decreasing fastest split variable "1,2,4-trichlorobenzene". The right sub-tree has been completely split and does not need to be split again, becoming a leaf node; the left sub-tree is further divided into six "higher purity" leaf nodes by the fluoride, nickel, petroleum hydrocarbon, and chloroform split variables. The purity of the leaf node affects the prediction performance of the model, and the lower the purity of the leaf node, the lower the prediction accuracy, recall rate, and classification accuracy of the model.

[0043] To further analyze the prediction performance of the three artificial intelligence models, the training set, the test set, and the total performance of the model (the ratio of the training set to the test set is 7:3) were evaluated, and the confusion matrix of the performance evaluation of the three artificial intelligence models is shown in Table 2. Figures 1-3 As can be seen from Table 2, the performance of CART in accuracy, precision, recall rate, and F1 value is significantly better than that of CHAID and E-CHAID, which is reflected in the following aspects: Figures 1-3

[0044] ① In terms of prediction accuracy, the accuracy of the three models in predicting "exceeding the standard" and "not exceeding the standard" in the training set and the test set is relatively high, but the prediction accuracy of the CART model is better than that of the CHAID and E-CHAID models, indicating that the CART model has stronger prediction ability for "exceeding the standard" and "not exceeding the standard". The total accuracy of the CHAID model is 95.52%, and the prediction accuracy of the training set and the test set is 95.80% and 94.85%, respectively. The total accuracy of the E-CHAID model is 95.77%, and the prediction accuracy of the training set and the test set is 96.28% and 94.56%, respectively. The total accuracy of the CART model is 98.13%, and the prediction accuracy of the training set and the test set is 98.37% and 97.63%, respectively.

[0045] ​②In terms of prediction accuracy, the total accuracy of CART model in predicting "non-exceeding" was 97.69%, and the accuracy of predicting "non-exceeding" in training set and test set was 97.99% and 97.04% respectively, which were higher than the prediction accuracy of CHAID and E-CHAID. The total accuracy of CHAID model in predicting "non-exceeding" was 94.76%, and the accuracy of predicting "non-exceeding" in training set and test set was 94.86% and 94.55% respectively; the total accuracy of E-CHAID model in predicting "non-exceeding" was 94.91%, and the accuracy of predicting "non-exceeding" in training set and test set was 95.59% and 93.26% respectively. The accuracy of CART model in predicting "exceeding" was 100%, which indicated that there was no false positive in predicting "exceeding" samples in training set and test set, while CHAID had false positive in test set.

[0046] ③In terms of prediction recall, the recall of CART model in predicting "exceeding" was 91.12%, and the recall of predicting "exceeding" in training set and test set was 92.04% and 89.29% respectively, which was significantly better than CHAID and E-CHAID (the recall of CHAID model in predicting "exceeding" was 79.29%, and the recall of predicting "exceeding" in training set and test set was 81.25% and 73.17% respectively; the recall of E-CHAID model in predicting "exceeding" was 79.88%, and the recall of predicting "exceeding" in training set and test set was 80.91% and 77.97% respectively). The recall of CART model in predicting "non-exceeding" in training set and test set was 100%, which indicated that there was no false positive in predicting "non-exceeding" samples in training set and test set, while CHAID had false positive in test set.

[0047] ④In addition, from the index F1 which comprehensively reflects the relationship between accuracy and recall, the F1 value of CART model for "non-exceeding" and "exceeding" of groundwater monitoring well (0.99, 0.95) was greater than that of CHAID algorithm (0.97, 0.88) and E-CHAID algorithm (0.97, 0.89). The greater the F1 value, the better the output result of the model.

[0048] Based on the best artificial intelligence model CART, the importance analysis of input variables on the prediction performance of the model was studied, the important indicators of groundwater monitoring of hazardous waste landfill were determined, and the optimized groundwater monitoring indicators were given. The importance analysis of input variables on the prediction ability of CART model is shown in Figure 4

[0049] Figure 4 ​​It can be seen that the change trend of the importance value and the normalized importance value is consistent in general, 1, 2, 4-trichlorobenzene and nickel have a very important influence on the prediction ability of the CART model target variable, and the normalized importance values are 1 and 0.923 respectively, and the importance values are 0.146 and 0.135 respectively; fluorides, petroleum hydrocarbons, trichloromethane, dichloromethane, cadmium and cis-1, 2-dichloroethylene have a relatively important influence on the prediction ability of the CART target variable, and the normalized importance values are 0.476, 0.445, 0.429, 0.328, 0.243 and 0.195 respectively, and the importance values are 0.069, 0.065, 0.063, 0.048, 0.035 and 0.028 respectively; arsenic, 1, 2-dichloropropane, 1, 2-dichloroethane, 1, 2, 3-trichlorobenzene, iron ions, sulfate, nitrate and dibromochloromethane also have a certain influence on the prediction ability of the CART target variable, and the normalized importance values are 0.187, 0.177, 0.173, 0.147, 0.122, 0.078, 0.023 and 0.016 respectively, and the importance values are 0.027, 0.026, 0.025, 0.021, 0.018, 0.011, 0.003 and 0.002 respectively. The importance value of 1, 2, 4-trichlorobenzene and nickel is significantly higher than that of other variables, and has a very important influence on the prediction of the over-standard situation of the groundwater monitoring well, and it is suggested that attention should be paid to the monitoring of these two indicators in the subsequent long-term monitoring of groundwater; fluorides, petroleum hydrocarbons, trichloromethane, dichloromethane, cadmium and cis-1, 2-dichloroethylene have a relatively important influence, and it is suggested that attention should be paid to the monitoring of these six indicators in the subsequent long-term monitoring of groundwater.

[0050] Through the above-mentioned artificial intelligence-based method, the groundwater monitoring index of a hazardous waste landfill in the middle part is optimized from 27 to 8.

[0051] For more information about the principles, methods and beneficial effects of the decision tree model construction in the embodiments of the present application, reference can be made to the related description of the prior art, which will not be repeated here.

[0052] The present application also provides a groundwater monitoring project optimization system, which at least comprises:

[0053] The acquisition module acquires historical monitoring data of a plurality of groundwater monitoring wells;

[0054] The first processing module is used for preprocessing the groundwater monitoring well data and constructing a groundwater monitoring data set;

[0055] The second processing module is used for analyzing the groundwater monitoring data set and determining the monitoring project of the groundwater monitoring well pollution over-standard.

[0056] The processing module of the embodiment implements the above steps, and the specific control mode can refer to the examples of the foregoing embodiments, which will not be described here.

[0057] The embodiment also provides a computer readable storage medium including a volatile or non-volatile, removable or non-removable medium implemented in any method or technology for storing information such as computer readable instructions, data structures, computer program modules or other data. The computer readable storage medium includes but is not limited to RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disc (DVD) or other optical disc storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer.

[0058] The computer readable storage medium in the embodiment can be used to store one or more computer programs, and the one or more computer programs stored therein can be executed by the processor to implement at least one step of the above monitoring index optimization.

[0059] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and any equivalent embodiments with equivalent changes or modifications are equivalent to the above embodiments. Any modification, change, equivalent change and modification of the above embodiments according to the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. An optimization method for groundwater monitoring projects at hazardous waste landfills based on artificial intelligence, characterized in that, Includes the following steps: (1) Collect historical monitoring data from multiple groundwater monitoring wells to construct a groundwater monitoring dataset; the groundwater monitoring dataset includes: the pollution exceedance status of each monitoring well, the location of each monitoring well, the monitoring items and the corresponding monitoring values, wherein the monitoring values ​​are used to represent the values ​​or status of the monitoring items; (2) Using the pollution exceedance status of the monitoring well as the target variable and the location of the monitoring well and the monitoring items as independent variables, multiple decision tree models are constructed; (3) The groundwater monitoring dataset is divided into a training set and a test set. The decision tree model is trained and tested using the training set and the test set. The best decision tree model is selected based on the model performance evaluation index. The performance evaluation index of the decision tree model includes accuracy (ACC), precision (PRE), recall (REC), and the harmonic mean of precision and recall (F1). (4) Use the best decision tree model to identify the monitoring items that have a significant impact on the pollution exceeding the standard of groundwater monitoring wells; if and only if the four performance evaluation indicators of one of the decision tree models are the best among all decision tree models, the decision tree model is determined to be the best decision tree model; use the best decision tree model to identify the monitoring items that have a significant impact on the pollution exceeding the standard of groundwater monitoring wells. The monitoring items in step (1) include key items and auxiliary items; the key items refer to the characteristics of toxic and hazardous pollutants in the hazardous waste landfill, including the content of fluoride, arsenic, cadmium, nickel, 1,2-dichloropropane, dichloromethane, cis-1,2-dichloroethylene, 1,2-dichloroethane, 1,2,4-trichlorobenzene, 1,2,3-trichlorobenzene, chloroform, dibromochloromethane, isophorone, and petroleum hydrocarbons; the auxiliary items refer to the internal and external environmental characteristics of the toxic and hazardous pollutants, including temperature, pH, and COD. Mn Water temperature, turbidity, odor and taste intensity, presence or absence of visible matter, dissolved oxygen, water level, redox potential, Cl- - Content, Fe 3+ Content, SO4 2- Content, NO3 - content; The pollution exceedance status of the monitoring well is an unordered categorical variable. It is encoded in [0,1] according to whether it exceeds the standard and then used as the target variable to input into the decision tree model; if the pollution of the monitoring well exceeds the standard, it is encoded as 1; if the pollution of the monitoring well does not exceed the standard, it is encoded as 0. The location of the monitoring well is an unordered categorical variable, and it is numerically encoded [1,2,3] according to its relative position in the groundwater flow field. If the monitoring well is located upstream, it is encoded as 1; if the monitoring well is located in the middle reaches, it is encoded as 2; if the monitoring well is located downstream, it is encoded as 3. The presence or absence of the visible object is an unordered categorical variable, and it is encoded with a value of [0,1] based on its presence or absence. If there is a visible object, it is encoded as 1; if there is no visible object, it is encoded as 0. For monitoring items with directly available monitoring values, a threshold-based method is used to encode the monitoring values. Step (2) involves constructing multiple decision tree models using three different decision tree algorithms: CHAID, E-CHAID, and CART. The maximum tree depth for CART is 5, while the maximum tree depths for CHAID and E-CHAID are both 3. The number of internal nodes for CART, CHAID, and E-CHAID are 5, 2, and 2, respectively, and the number of leaf nodes are 7, 4, and 4, respectively. The root node of CHAID is χ... 2 The largest splitting variable, "1,2,4-trichlorobenzene," divides the dataset into two relatively pure subtrees: "≤ detection limit to standard limit" and "> detection limit to standard limit." The right subtree is already completely split and does not need further splitting, becoming a leaf node. The left subtree is then further divided into three leaf nodes with higher purity by the splitting variables fluoride and petroleum hydrocarbons. The E-CHAID root node is defined by χ². 2 The largest splitting variable, "1,2,4-trichlorobenzene," divides the dataset into two relatively pure subtrees: "≤ detection limit to standard limit" and "> detection limit to standard limit." The right subtree is completely split and does not need to be split again, becoming a leaf node. The left subtree is then divided into three leaf nodes with higher purity by the splitting variables fluoride and nickel. The CART root node, with the splitting variable "1,2,4-trichlorobenzene" showing the fastest decrease in "impurity," divides the dataset into two relatively pure subtrees: "≤ detection limit to standard limit" and "> detection limit to standard limit." The right subtree is completely split and does not need to be split again, becoming a leaf node. The left subtree is then divided into six leaf nodes with higher purity by the splitting variables fluoride, nickel, petroleum hydrocarbons, and chloroform.

2. The optimization method for groundwater monitoring projects at hazardous waste landfills according to claim 1, characterized in that, The calculation methods for accuracy (ACC), precision (PRE), recall (REC), and the harmonic mean F1 of precision and recall are as follows: ACC=(TM0+TM1) / (TM0+TM1+FM0+FM1) PRE1=(TM1) / (TM1+FM1), PRE0=(TM0) / (TM0+FM0); REC1=(TM1) / (P1), REC0=(TM0) / (P0); F1 = (2 × PRE × REC) / (PRE + REC); Where: TM0 is the number of samples correctly predicted as pollution not exceeding the standard in the monitoring well; TM1 is the number of samples correctly predicted as pollution exceeding the standard in the monitoring well; FM0 is the number of samples incorrectly predicted as pollution not exceeding the standard in the monitoring well; FM1 is the number of samples incorrectly predicted as pollution exceeding the standard in the monitoring well; PRE1 represents the percentage of samples predicted as pollution exceeding the standard in the monitoring well that actually exceed the standard; PRE0 represents the percentage of samples predicted as pollution not exceeding the standard in the monitoring well that actually do not exceed the standard; REC1 represents the percentage of samples correctly predicted as pollution exceeding the standard in the monitoring well out of the actual number of samples exceeding the standard; REC0 represents the percentage of samples correctly predicted as pollution not exceeding the standard in the monitoring well out of the actual number of samples not exceeding the standard; P1 is the actual number of samples exceeding the standard in the monitoring well; P0 is the actual number of samples not exceeding the standard in the monitoring well.

3. The optimization method for groundwater monitoring projects at hazardous waste landfills according to claim 1, characterized in that, The ratio of training set to test set data is 7:

3.

4. An AI-based optimization system for groundwater monitoring projects at hazardous waste landfills, used to implement the AI-based optimization method for groundwater monitoring projects at hazardous waste landfills as described in any one of claims 1-3, characterized in that, The system includes: The data acquisition module collects historical monitoring data from multiple groundwater monitoring wells; The first processing module is used to preprocess historical monitoring data from groundwater monitoring wells to construct a groundwater monitoring dataset. The groundwater monitoring dataset includes: the pollution exceedance status of each monitoring well, the location of each monitoring well, the monitoring items and the corresponding monitoring values, wherein the monitoring values ​​are used to represent the numerical value or status of the monitoring items. The second processing module is used to analyze the groundwater monitoring dataset and identify monitoring items that have a significant impact on the pollution levels of groundwater monitoring wells.

5. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the method of any one of claims 1-3.

Images