A Collaborative Optimization Method for Soil and Groundwater Monitoring Networks Based on Multi-Source Information and Machine Learning

Abstract

This invention discloses a collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning. The method optimizes the network layout by zoning the study area under multi-source information, analyzing spatial variability and uncertainty, generating a preliminary set of candidate monitoring points, and implementing multi-objective optimization. This invention integrates multi-source prior information to achieve scientific zoning of the study area, overcoming the shortcomings of traditional methods that neglect spatial heterogeneity, and making the monitoring network layout more consistent with the actual geological background. It quantifies the uncertainty of spatial prediction, overcoming the limitations of relying solely on subjective experience and achieving a quantitative balance between uncertainty reduction and monitoring costs. Simultaneously, by analyzing the correlation between soil and groundwater pollution, collaborative monitoring is achieved, avoiding redundant monitoring points and significantly reducing monitoring costs. The introduction of an automated machine learning model fully utilizes multi-source heterogeneous data, improving the accuracy of pollution risk prediction; combined with a multi-objective optimization algorithm, it provides decision-makers with diverse options.

Description

Technical Field

[0001] This invention relates to the field of soil and groundwater monitoring technology, and in particular to a collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning. Background Technology

[0002] Soil and groundwater are interdependent, and accurately understanding their pollution status is a prerequisite and core element for conducting site environmental risk assessments, developing scientific remediation plans, and implementing effective environmental management. For a long time, the deployment and optimization of soil and groundwater monitoring networks have relied primarily on traditional technologies, which have many insurmountable drawbacks. Early monitoring well deployment methods often depended on the subjective experience of hydrogeological experts or employed simple uniform grid layouts, lacking quantitative consideration of the spatial heterogeneity of the geological and hydrogeological background conditions of the study area. This resulted in insufficient rationality in the layout of monitoring points: including deploying too many redundant points in low-pollution-risk areas, wasting monitoring costs such as well construction and maintenance; and sparsely deploying points in high-pollution-risk areas, creating monitoring blind spots, failing to accurately capture pollution distribution characteristics and diffusion trends, and failing to meet the needs of precise monitoring.

[0003] Current technologies still have limitations, often focusing on a single objective: monitoring the current state of groundwater pollution. They fail to adequately consider the synergistic and holistic nature of soil and groundwater pollution, and lack optimization of the integrated soil-groundwater monitoring system. Static analysis, primarily based on historical or current data, fails to effectively incorporate advanced technologies like machine learning to predict the spatial distribution of pollutants, thus hindering proactive guidance for monitoring site deployment. Optimization is also limited, with targets often confined to cost or a few technical indicators, making it difficult to balance conflicting objectives such as monitoring coverage, uncertainty reduction, cost control, and regional representativeness. Therefore, a targeted monitoring network solution is needed that can synergistically consider soil and groundwater media, overcome spatial heterogeneity, quantify spatial prediction uncertainties, and achieve multi-objective optimization. Summary of the Invention

[0004] This invention provides a collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning, which can solve technical problems such as collaboratively considering soil and groundwater media, overcoming spatial heterogeneity defects, quantifying the uncertainty of spatial prediction, and multi-objective optimization monitoring.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The specific steps of the collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning are as follows:

[0007] Step 1: Zoning of the study area based on multi-source information: Collect multi-source information on soil type, hydrological conditions, land use type, and historical pollution monitoring of the study area; Divide the study area into multiple zones with similar geological background characteristics based on the multi-source information;

[0008] Step 2: Spatial Variability and Uncertainty Analysis: Simultaneously analyze the spatial distribution of soil and groundwater monitoring indicators within each zone, calculate the prediction variance and information entropy value at each location to quantify the spatial variability and information acquisition value of the monitoring data; then identify hotspot areas with high spatial variability and large prediction uncertainty, and analyze the spatial correlation between soil and groundwater pollution.

[0009] Step 3: Construct a machine learning model: Using multi-source information, hotspot areas obtained from analysis, and soil-groundwater pollution correlation parameters as input features, construct a machine learning model accordingly;

[0010] Step 4: Generate a preliminary candidate site set: Using the constructed machine learning model, predict the spatial distribution probability of pollutants in the entire study area, and generate a preliminary candidate site set of monitoring wells covering the monitoring needs of soil and groundwater based on the prediction results.

[0011] Step 5: Multi-objective optimization layout: Construct a multi-objective optimization model with the objectives of minimizing monitoring costs, minimizing overall prediction uncertainty, and maximizing spatial coverage uniformity, and then optimize the preliminary candidate site set; from this, determine the optimal layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater.

[0012] Furthermore, the study area in step one is divided into zones based on similar geological background characteristics, and the K-means clustering algorithm is used to perform spatial clustering of the study area.

[0013] Furthermore, the feature parameters of the multi-source information input in the K-means clustering algorithm include at least one or more combinations of soil type, aquifer lithology, groundwater depth, land use type, and distribution of historical pollution sites. The number of clusters K is determined by the silhouette coefficient method to ensure maximum similarity within each partition and maximum difference between partitions.

[0014] Furthermore, in step two, within each study area, the spatial distribution of soil and groundwater monitoring indicators concentrations is analyzed simultaneously using Kriging interpolation and information entropy theory.

[0015] Furthermore, the formula for calculating the information entropy value is:

[0016]

[0017] In the formula, H(X) represents the information entropy at spatial location x, and p(x) represents the information entropy at spatial location x. ip(x) represents the probability that the pollutant concentration at this location is at the i-th level, where n is the total number of concentration levels. i (Obtained via the instruction Kriging method.)

[0018] Furthermore, the spatial correlation between soil and groundwater pollution in step two is achieved by calculating the correlation coefficient between soil pollutant concentration and groundwater pollutant concentration. When the correlation coefficient is greater than 0.6, it is determined to be a strongly correlated area, and soil monitoring points and groundwater monitoring wells will be prioritized for deployment in subsequent site selection.

[0019] Furthermore, for the machine learning model constructed in step three, a random forest model is adopted, using geological, hydrological, and land use characteristics as input variables to predict the distribution and uncertainty of pollutant concentrations.

[0020] Furthermore, for step five, the multi-objective optimization model is expressed as:

[0021]

[0022]

[0023]

[0024]

[0025] In the formula, X is the decision variable vector, x i The value is 0 or 1, indicating whether the i-th candidate point is selected. 1 indicates that the point is selected, and 0 indicates that it is not selected.

[0026] f1(X) is the total monitoring cost, c i Let f2(X) be the well construction and maintenance cost at the i-th point; f2(X) be the average prediction variance; N be the total number of grid points in the study area; J be the set of all grid points in the study area; and Var j (X) represents the Kriging prediction variance of the j-th grid point under the current monitoring layout X; f3(X) is the negative value of spatial coverage uniformity, and Coverage(X) is an index reflecting the degree of spatial distribution uniformity of monitoring points, which is constructed based on the area variation coefficient of Thiessen polygons.

[0027] Furthermore, in step five, after constructing the multi-objective optimization model, the non-dominated sorting genetic algorithm (NSGA-II) with elitist strategy is used to optimize the preliminary candidate point set to obtain the Pareto optimal solution set, from which the optimized layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater is determined.

[0028] The multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm with an elite strategy (NSGA-II). Its decision variables are encoded in binary. Each gene position corresponds to a monitoring well position in the initial candidate position set. A gene value of "1" indicates that the position is selected, and "0" indicates that it is not selected. During the population evolution process, a partition constraint processing mechanism is introduced to ensure that at least one monitoring well position is selected in each partition.

[0029] The beneficial effects of this invention are reflected in:

[0030] 1) This invention integrates multi-source prior information such as geology, hydrogeology, and land use, realizes the scientific zoning of the study area, overcomes the shortcomings of traditional methods that ignore spatial heterogeneity, and makes the layout of the monitoring network more in line with the actual geological background.

[0031] 2) This invention quantifies the uncertainty of spatial prediction, overcomes the need to rely solely on subjective experience for judgment, and also assesses the value of information acquisition, achieving a quantitative balance between uncertainty reduction and monitoring costs; at the same time, by analyzing the correlation between soil and groundwater pollution, it achieves collaborative monitoring, avoids redundant site deployment, and significantly reduces monitoring costs.

[0032] 3) This invention introduces an automatic machine learning model, which makes full use of multi-source heterogeneous data, improves the accuracy of pollution risk prediction, and provides reliable prior knowledge for monitoring network optimization; combined with the NSGA-II multi-objective optimization algorithm, it can find the optimal solution set that takes into account cost, accuracy and spatial representativeness on the Pareto front, providing decision-makers with diversified options.

[0033] 4) The method of this invention achieves synergistic optimization of soil and groundwater monitoring. By sharing monitoring wells to obtain dual-media data, it can save 30%-50% of the cost compared with the traditional independent deployment method while ensuring monitoring accuracy. It has good economic and environmental benefits and is suitable for the construction of monitoring networks for large-scale contaminated sites.

[0034] Therefore, this application solves the technical problems of synergistically considering soil and groundwater media, overcoming spatial heterogeneity defects, quantifying the uncertainty of spatial prediction, and multi-objective optimized monitoring. Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention; the main objectives and other advantages of the invention can be realized and obtained by means of the methods particularly pointed out in the description. Attached Figure Description

[0035] Figure 1 This is a technical roadmap for a collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning.

[0036] Figure 2This is a K-means spatial clustering result diagram. The left image is in latitude and longitude coordinate system, and the right image is in metric coordinate system. Different colors represent different clustering partitions (6 classes in total).

[0037] Figure 3 These are the results of Kriging interpolation. The left figure shows the spatial distribution of the predicted risk scores, and the right figure shows the spatial distribution of the predicted variance (uncertainty).

[0038] Figure 4 This is a graph showing the information entropy distribution of the Kriging variance;

[0039] Figure 5 It is a multi-source characteristic correlation heatmap, used to show the correlation coefficients between various water quality indicators and hydrogeological parameters;

[0040] Figure 6 This is a bar chart showing the distribution of well counts in each cluster region;

[0041] Figure 7 This is the Pareto front plot obtained by the NSGA-II algorithm, showing the trade-off between information entropy and cost;

[0042] Figure 8 This is a schematic diagram of the monitoring network distribution before optimization;

[0043] Figure 9 This is a schematic diagram of the distribution after the monitoring network optimization;

[0044] Figure 10 This is a comparison chart of prediction uncertainty. Detailed Implementation

[0045] The technical solutions of the present invention will be described in detail below through embodiments. The following embodiments are merely exemplary and can only be used to explain and illustrate the technical solutions of the present invention, and should not be construed as limiting the technical solutions of the present invention.

[0046] like Figure 1 As shown, the collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning is further illustrated. The specific steps are as follows:

[0047] Step 1: Zoning of the study area based on multi-source information: Collect multi-source information on soil type, hydrological conditions, land use type, and historical pollution monitoring of the study area; based on the multi-source information, divide the study area into multiple zones with similar geological background characteristics.

[0048] For the study area in Step 1, the area is divided into zones based on similar geological background characteristics, and the K-means clustering algorithm is used to perform spatial clustering of the study area. The feature parameters of the multi-source information input in the K-means clustering algorithm include at least one or more combinations of soil type, aquifer lithology, groundwater level depth, land use type, and distribution of historical pollution sites. The number of clusters K is determined by the silhouette coefficient method to ensure maximum similarity within each zone and maximum difference between zones.

[0049] In step two, within each study area, the spatial distribution of soil and groundwater monitoring indicators is analyzed simultaneously using Kriging interpolation and information entropy theory.

[0050] The formula for calculating information entropy is:

[0051]

[0052] In the formula, H(X) represents the information entropy at spatial location x, and p(x) represents the information entropy at spatial location x. i p(x) represents the probability that the pollutant concentration at this location is at the i-th level, where n is the total number of concentration levels. i (Obtained via the instruction kriging method.)

[0053] The spatial correlation between soil and groundwater pollution in step two is achieved by calculating the correlation coefficient between soil pollutant concentration and groundwater pollutant concentration. When the correlation coefficient is greater than 0.6, it is determined to be a strongly correlated area, and soil monitoring points and groundwater monitoring wells will be prioritized for deployment in subsequent site selection.

[0054] Step 2: Spatial Variability and Uncertainty Analysis: Simultaneously analyze the spatial distribution of soil and groundwater monitoring indicators within each zone, calculate the prediction variance and information entropy value at each location to quantify the spatial variability and information acquisition value of the monitoring data; then identify hotspot areas with high spatial variability and large prediction uncertainty, and analyze the spatial correlation between soil and groundwater pollution.

[0055] Step 3: Construct a machine learning model: Using multi-source information, hotspot areas obtained from analysis, and soil-groundwater pollution correlation parameters as input features, construct a machine learning model.

[0056] For the machine learning model constructed in step three, a random forest model is adopted, using geological, hydrological, and land use characteristics as input variables to predict the distribution and uncertainty of pollutant concentrations.

[0057] Step 4: Generate a preliminary candidate site set: Using the constructed machine learning model, predict the spatial distribution probability of pollutants in the entire study area, and generate a preliminary candidate site set of monitoring wells covering the monitoring needs of soil and groundwater based on the prediction results.

[0058] Step 5: Multi-objective optimization layout: Construct a multi-objective optimization model with the objectives of minimizing monitoring costs, minimizing overall prediction uncertainty, and maximizing spatial coverage uniformity, and then optimize the preliminary candidate site set; from this, determine the optimal layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater.

[0059] For step five, the multi-objective optimization model is expressed as:

[0060]

[0061]

[0062]

[0063]

[0064] In the formula, X is the decision variable vector, x i The value is 0 or 1, indicating whether the i-th candidate point is selected. 1 indicates that the point is selected, and 0 indicates that it is not selected.

[0065] f1(X) is the total monitoring cost, c i Let f2(X) be the well construction and maintenance cost at the i-th point; f2(X) be the average prediction variance; N be the total number of grid points in the study area; J be the set of all grid points in the study area; and Var j (X) represents the Kriging prediction variance of the j-th grid point under the current monitoring layout X; f3(X) is the negative value of spatial coverage uniformity, and Coverage(X) is an index reflecting the degree of spatial distribution uniformity of monitoring points, which is constructed based on the area variation coefficient of Thiessen polygons.

[0066] In step five, after constructing the multi-objective optimization model, the non-dominated sorting genetic algorithm with elitist strategy (NSGA-II) is used to optimize the preliminary candidate point set to obtain the Pareto optimal solution set, from which the optimal layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater is determined.

[0067] The multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm with an elite strategy (NSGA-II). Its decision variables are encoded in binary. Each gene position corresponds to a monitoring well position in the initial candidate position set. A gene value of "1" indicates that the position is selected, and "0" indicates that it is not selected. During the population evolution process, a partition constraint processing mechanism is introduced to ensure that at least one monitoring well position is selected in each partition.

[0068] A plain area in a certain city was selected as a typical region, including 86 historical monitoring wells. The main monitoring indicators included groundwater chemical components such as sulfate, chloride, fluoride, sodium, manganese, and arsenic, as well as hydrogeological parameters such as ground elevation, stable water level elevation, and groundwater depth. Digital elevation models (DEMs), hydrogeological survey reports, and historical water quality monitoring data were collected for the study area. Feature vectors were extracted for each monitoring point, including soil type, aquifer lithology, groundwater depth, and spatial coordinates. K-means clustering was used for spatial clustering, and the optimal number of clusters, K=6, was determined using the silhouette coefficient method, dividing the study area into 6 geological and hydrological units. Figure 2 As shown, the cluster regions are spatially distributed in a regular manner. Cluster 3 (central region) contains 25 wells, and cluster 5 (southeast) contains 18 wells. The geological backgrounds within each region are highly similar.

[0069] Based on historical data from 86 existing monitoring wells, the generalized kriging method was used to spatially interpolate the comprehensive risk score. For example... Figure 3 As shown, the left figure shows the spatial distribution of Kriging interpolation predictions, and the right figure shows the distribution of prediction variance (uncertainty). In the densely populated monitoring well region (near the center coordinates), the prediction variance is low (0.063-0.075); in the boundary region, the uncertainty increases significantly. Simultaneously, the information entropy distribution is calculated based on the indicator Kriging method (…). Figure 4 The high information entropy area (>0.28nat) is mainly concentrated in the northeast of the study area, indicating that the concentration level of pollutants in this area is highly uncertain and the information acquisition value is high.

[0070] Through feature correlation analysis ( Figure 5 A risk prediction model was constructed using the Random Forest algorithm as input variables, including sulfate (mg / L), chloride (mg / L), fluoride (mg / L), sodium (mg / L), manganese (mg / L), arsenic (mg / L), spatial coordinates (X,Y), groundwater depth (m), ground elevation, stable water level elevation, pH (dimensionless), color (degrees), turbidity (NTU), total hardness (mg / L), and total dissolved solids (mg / L). Cross-validation results showed R² = 0.7582, significantly outperforming traditional geostatistical methods in prediction accuracy. Feature importance analysis indicated that chloride and the TDS / TDS ratio were the most important predictors, with importance scores of 0.93. At least 8 monitoring wells were required to be retained within each cluster partition, and collaborative monitoring wells were mandatory in strongly correlated regions (r > 0.6). Figure 6As shown, the distribution of well numbers in each cluster region is as follows: Cluster 0 (8 wells), Cluster 1 (15 wells), Cluster 2 (10 wells), Cluster 3 (25 wells), Cluster 4 (10 wells), and Cluster 5 (18 wells). Based on the machine learning prediction results, 13 high-risk blank areas were identified, and a preliminary candidate point set containing 120 candidate points was generated (twice the final number of points).

[0071] A multi-objective optimization model is constructed with the objectives of minimizing monitoring costs, minimizing prediction uncertainty (information entropy), and maximizing spatial coverage uniformity. The NSGA-II algorithm is used to solve the model, with a population size of 100, 500 generations, a crossover probability of 0.9, and a mutation probability of 0.1. Figure 7 As shown, the Pareto front clearly demonstrates the trade-off between information entropy and cost: when the cost decreases from 55 to 12, the information entropy increases from -0.07 to 0.06. Decision-makers choose the inflection point solution based on budget constraints (cost ≈ 30, information entropy ≈ -0.02).

[0072] Optimization results are as follows Figure 8 and Figure 9 As shown, 50 wells were selected and retained from the original 86 wells, and 13 new optimized wells were added, ultimately forming an optimized layout network of 63 monitoring wells. The optimized monitoring wells are more evenly distributed in space, eliminating the densely overlapping areas (well spacing <500m in the original central area) and monitoring blind areas (boundary areas) of the original monitoring network.

[0073] Optimization effect verification ( Figure 10 The results show that the average kriging variance (0.065) after optimization is basically the same as before optimization, but the number of monitoring wells is reduced by 26.7%, the cost is reduced by 26.7%, the coefficient of variation of spatial coverage (CV) is reduced from 0.45 to 0.32, and the densely overlapping area of ​​the original monitoring network (well spacing in the original central area is <500m) and monitoring blind area are eliminated.

[0074] Comparative Example 1

[0075] This comparative example uses the traditional expert experience method to deploy the monitoring network, eliminating the quantitative analysis process and relying solely on expert interpretation to evenly distribute monitoring points in suspected contaminated areas. Under the same budget constraints as the example, 63 monitoring wells were deployed, but the spatial coverage uniformity was poor (coefficient of variation CV=0.45), and three actual contaminated hotspots were missed (the coverage rate of areas with information entropy > 0.3 was only 60%).

[0076] Comparative Example 2

[0077] This comparative example employs a simple spatial coverage optimization method (maximizing spatial representativeness without considering geological zoning and pollution correlation), solved using an integer programming algorithm. While 63 monitoring points were deployed, achieving uniform spatial coverage (CV=0.32), insufficient point deployment in geologically complex areas (cluster 3) resulted in a 35% higher prediction variance compared to the previous example. Furthermore, the lack of consideration for soil-groundwater synergy led to a 15% higher overall cost.

[0078] The method of this invention achieves synergistic optimization of soil and groundwater monitoring. By acquiring dual-media data through shared monitoring wells, it can save 30%-50% of the cost compared with the traditional independent deployment method while ensuring monitoring accuracy. It has good economic and environmental benefits and is suitable for the construction of monitoring networks for large-scale contaminated sites.

[0079] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative optimization method for soil and groundwater monitoring networks based on multi-source information and machine learning, characterized in that, The specific steps are as follows: Step 1: Zoning of the study area based on multi-source information: Collect multi-source information on soil type, hydrological conditions, land use type, and historical pollution monitoring of the study area; Divide the study area into multiple zones with similar geological background characteristics based on the multi-source information; Step 2: Spatial Variability and Uncertainty Analysis: Simultaneously analyze the spatial distribution of soil and groundwater monitoring indicators within each zone, calculate the prediction variance and information entropy value at each location to quantify the spatial variability and information acquisition value of the monitoring data; then identify hotspot areas with high spatial variability and large prediction uncertainty, and analyze the spatial correlation between soil and groundwater pollution. Step 3: Construct a machine learning model: Using multi-source information, hotspot areas obtained from analysis, and soil-groundwater pollution correlation parameters as input features, construct a machine learning model accordingly; Step 4: Generate a preliminary candidate site set: Using the constructed machine learning model, predict the spatial distribution probability of pollutants in the entire study area, and generate a preliminary candidate site set of monitoring wells covering the monitoring needs of soil and groundwater based on the prediction results. Step 5: Multi-objective optimization layout: Construct a multi-objective optimization model with the objectives of minimizing monitoring costs, minimizing overall prediction uncertainty, and maximizing spatial coverage uniformity, and then optimize the preliminary candidate site set; from this, determine the optimal layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater.

2. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 1, characterized in that, For the study area in step one, the study area is divided into zones with similar geological background characteristics, and the K-means clustering algorithm is used to perform spatial clustering of the study area.

3. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 2, characterized in that, The feature parameters of the K-means clustering algorithm that are input from multiple sources include at least one or more combinations of soil type, aquifer lithology, groundwater depth, land use type, and distribution of historical pollution sites. The number of clusters K is determined by the silhouette coefficient method to ensure that the similarity within each partition is maximized and the difference between partitions is maximized.

4. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 1, characterized in that, In step two, within each study area, the spatial distribution of soil and groundwater monitoring indicators is analyzed simultaneously using Kriging interpolation and information entropy theory.

5. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 4, characterized in that, The formula for calculating information entropy is: In the formula, H(X) represents the information entropy at spatial location x, and p(x) represents the information entropy at spatial location x. i p(x) represents the probability that the pollutant concentration at this location is at the i-th level, where n is the total number of concentration levels. i (Obtained via the instruction Kriging method.) 6. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 1, characterized in that, The spatial correlation between soil and groundwater pollution in step two is achieved by calculating the correlation coefficient between soil pollutant concentration and groundwater pollutant concentration. When the correlation coefficient is greater than 0.6, it is determined to be a strongly correlated area, and soil monitoring points and groundwater monitoring wells are deployed in a coordinated manner in subsequent site selection.

7. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 1, characterized in that, For the machine learning model constructed in step three, a random forest model is adopted, using geological, hydrological, and land use characteristics as input variables to predict the distribution and uncertainty of pollutant concentrations.

8. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 1, characterized in that, For step five, the multi-objective optimization model is expressed as: In the formula, X is the decision variable vector, x i The value is 0 or 1, indicating whether the i-th candidate point is selected. 1 indicates that the point is selected, and 0 indicates that it is not selected. f1(X) is the total monitoring cost, c i Let f2(X) be the well construction and maintenance cost at the i-th point; f2(X) be the average prediction variance; N be the total number of grid points in the study area; J be the set of all grid points in the study area; and Var j (X) represents the Kriging prediction variance of the j-th grid point under the current monitoring layout X; f3(X) is the negative value of spatial coverage uniformity, and Coverage(X) is an index reflecting the degree of spatial distribution uniformity of monitoring points, which is constructed based on the area variation coefficient of Thiessen polygons.

9. The method for collaborative optimization of soil and groundwater monitoring networks based on multi-source information and machine learning as described in claim 8, characterized in that, In step five, after constructing the multi-objective optimization model, the non-dominated sorting genetic algorithm with elitist strategy (NSGA-II) is used to optimize the preliminary candidate point set to obtain the Pareto optimal solution set, from which the optimal layout scheme of the monitoring network that takes into account the needs of coordinated monitoring of soil and groundwater is determined. The multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm with an elite strategy (NSGA-II). Its decision variables are encoded in binary. Each gene position corresponds to a monitoring well position in the initial candidate position set. A gene value of "1" indicates that the position is selected, and "0" indicates that it is not selected. During the population evolution process, a partition constraint processing mechanism is introduced to ensure that at least one monitoring well position is selected in each partition.

Images