Underground water pollution monitoring natural attenuation early warning and repairing method based on machine learning

Abstract

The invention discloses an underground water pollution monitoring natural attenuation early warning and repairing method based on machine learning, and the method comprises the following specific steps: laying a polluted site underground water monitoring network, and carrying out the underground water flow direction and water quality multi-parameter monitoring; collecting an underground water sample of the polluted site; reconstructing a time sequence array; constructing a response model of migration of the underground water pollutants along the flow direction; evaluating the natural attenuation performance of groundwater pollutants by adopting a trend analysis method, and estimating the occurrence interval and threshold value of the natural attenuation of the groundwater polluted site through change-point analysis; and formulating a polluted site underground water in-situ classification and grading remediation method based on an early warning result. According to the method, targeted remediation is carried out through establishment of the response model of migration of the underground water pollutants along with the flow direction, analysis of the change trend and interval threshold value of natural attenuation of the pollutants and grading and classification, and the early warning threshold value and the in-situ remediation strategy of the underground water monitoring natural attenuation of the polluted site can be scientifically and efficiently revealed; and accurate implementation of in-situ remediation of underground water pollution is ensured.

Description

Technical Field

[0001] This invention relates to the field of groundwater monitoring technology for contaminated sites, and in particular to a groundwater pollution monitoring, natural attenuation early warning, and remediation method based on machine learning. Background Technology

[0002] Natural attenuation of pollutants in groundwater is an important means of groundwater pollution remediation. However, due to the complexity of the site, the high heterogeneity of hydrogeological conditions, and the nonlinearity of groundwater flow, the attenuation performance and occurrence area of ​​groundwater pollutants in the site are unclear. As a result, the reliability and feasibility of natural attenuation in the application of groundwater pollution cannot be accurately applied. There is an urgent need for a response model that can scientifically and efficiently monitor and provide early warning of the natural attenuation performance of groundwater in the site, and thereby formulate more accurate remediation methods. Summary of the Invention

[0003] This invention provides a machine learning-based method for monitoring, monitoring, early warning, and remediation of groundwater pollution. It addresses technical issues such as establishing response models for the migration of groundwater pollutants with the flow direction, analyzing the changing trends and interval thresholds of natural pollutant decay, and implementing targeted remediation based on classification and grading.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: The specific steps of the machine learning-based groundwater pollution monitoring, natural attenuation, early warning, and remediation method are as follows: Step 1: Deploy a groundwater monitoring network for the contaminated site, conduct multi-parameter monitoring of groundwater flow direction and water quality, and collect groundwater samples from the contaminated site; Step 2: Reconstruct the time series array and perform data preprocessing, including reading monitoring data in text format, processing mixed data types, finding missing values, removing error values, and standardizing and removing dimensions from the data. Step 3: Based on the groundwater flow direction, the Support Vector Regression (SVR) algorithm is used, with historical time step data as input, to predict the pollutant concentration in future time steps, thereby constructing a response model for the migration of groundwater pollutants with the flow direction. Step 4: Assess the natural decay performance of groundwater pollutants using trend analysis, calculate the trend of average concentration of each pollutant over time, and use linear regression analysis to calculate the trend slope and statistical significance. p value); Step 5: Estimate the occurrence range and threshold of natural decay of groundwater contaminated sites through variable point analysis, and use nonparametric self-sampling to estimate the probability of natural decay of pollutants with groundwater flow direction. Step 6: Draw intuitive and visual charts to represent the probability, cumulative distribution, and concentration change points of pollutants as they naturally decay with water flow, and formulate in-situ classification and grading remediation methods for groundwater at contaminated sites based on the early warning results.

[0005] Furthermore, in step one, a groundwater monitoring network for the contaminated site is established. Based on the geological and hydrogeological conditions of the site and the characteristics of groundwater pollution, natural decay monitoring wells are set up around the delineated pollution range, along with background monitoring wells, pollution monitoring wells, and control monitoring wells. Among them, the pollution monitoring wells include those set up at the pollution source, upstream of the pollution source, and downstream of the pollution source, in order to depict the dynamics of key areas of the pollution plume in detail, while making full use of the monitoring wells set up during the site environmental survey.

[0006] Furthermore, in the multi-parameter water quality monitoring in step one, the water level is monitored through groundwater level observation wells, while water temperature, dissolved oxygen, pH, redox potential, and conductivity are measured on-site using portable instruments.

[0007] Furthermore, the groundwater samples collected in step one from the contaminated site were analyzed and tested. Based on the rainfall and pollution characteristics of the site, typical groundwater pollution indicators were selected and seasonally sampled and analyzed. Before sample collection, the wells were cleaned three times using a handheld peristaltic pump. Sampling began after the cleaning was deemed satisfactory.

[0008] Furthermore, in step two, during the monitoring data processing, time and well number information are extracted, and the data is reorganized into a three-dimensional array structure of {well number × pollutant × time step}; among them, the reconstructed time series array is used as input for the historical time step data in step three.

[0009] Furthermore, for the response model based on the migration of groundwater pollutants with the flow direction in step three, a prediction model is trained separately for each pollutant, and the model performance is evaluated based on the mean square error. The Gaussian kernel function is used to handle nonlinear relationships and automatically optimize the kernel parameters.

[0010] Furthermore, in step four, when using trend analysis to assess the natural decay performance of groundwater pollutants, at a 95% confidence level, the natural decay probability is calculated based on the decrease rate and trend significance. Pollutants are then categorized according to their decay probability. p It is divided into three levels: high (>0.7), medium (0.4-0.7), and low (<0.4).

[0011] Furthermore, in step five, the occurrence range and threshold of natural decay of groundwater contaminated sites are estimated through change point analysis. For pollutants with significant decay trends, concentration change point detection is performed. The Bayesian change point detection method is used to identify key time points of concentration change. The change point positions are marked on the time series to determine the occurrence range of natural decay and multi-level early warning thresholds.

[0012] Furthermore, in step five, nonparametric auto-sampling is used to estimate the probability of pollutants naturally decaying with the direction of groundwater flow. Specifically, 1000 uniform samples with replacement are taken from the given training set. That is, whenever a sample is selected, it is selected again with equal probability and added to the training set again. Specifically, the process involves: resampling with replacement from the original sample to obtain a resampled set of the same size as the original sample; using these resampled data sets to estimate the bias; repeating this process 1000 times to obtain a large number of estimates, thus obtaining the sampling distribution of the statistic; and using these sampling distributions to estimate the properties of the population distribution and determine the frequency of occurrence.

[0013] Furthermore, based on the early warning results of natural attenuation of groundwater pollutants revealed by machine learning in step six, and in conjunction with the site's groundwater function requirements, key areas and indicators for in-situ groundwater remediation are identified, and remediation methods for pollutant classification and grading are formulated.

[0014] The beneficial effects of this invention are reflected in: By deploying a groundwater monitoring well network and conducting water quality monitoring and analysis, the real-time nature and accuracy of water quality indicators are ensured. A multi-method fusion assessment system is formed by constructing a response model for the migration of groundwater pollutants along the flow direction, combining linear regression, hypothesis testing statistical analysis, and machine learning methods (SVR). Based on the reconstruction of historical time data, the scientific rigor and accuracy of natural degradation assessment are improved. The efficiency of machine learning is utilized to quantify the uncertainty of prediction results, and trend analysis, change point assessment, and intuitive display provide more comprehensive information support for decision-making. This invention automates and intelligentizes monitoring and early warning, greatly improving work efficiency; it generates rich visualization results and detailed assessment reports, facilitating classification, grading, understanding, and decision-making. It provides a scientific basis and technical support for groundwater monitoring and remediation of contaminated sites.

[0015] This invention effectively solves the technical problems of establishing response models for the migration of groundwater pollutants with the flow direction during groundwater pollution, analyzing the changing trends and interval thresholds of natural pollutant decay, and implementing targeted remediation based on classification and grading. It can scientifically and efficiently reveal the early warning thresholds for natural decay of groundwater at contaminated sites and the in-situ remediation strategies, ensuring the precise implementation of in-situ remediation of groundwater at contaminated sites. Other features and advantages of this 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 this invention can be realized and obtained by means of the methods particularly pointed out in the description. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the groundwater monitoring well network layout at a landfill site; Figure 2 This is a table analyzing the decline trends of various pollutants; Figure 3 This is a diagram showing the ranking of the natural decay probabilities of various pollutants. Figure 4 This is a diagram illustrating the probability assessment levels of natural decay. Figure 5 This is a cumulative distribution curve of the natural decay of hexavalent chromium; Figure 6 This is a point analysis chart showing the concentration variation of hexavalent chromium. Detailed Implementation

[0017] like Figure 1 As shown, taking a certain landfill as an example, according to the contaminated site exploration data, the groundwater aquifer in the area where this valley-type landfill is located is mainly a bedrock fissure aquifer, and the rock mass is mainly early Mesozoic Jurassic quartz syenite porphyry. The bedrock fissures at the bottom are widely distributed, and the natural flow direction of groundwater is from northeast to southwest.

[0018] Combination Figures 1 to 6 As shown, the method for monitoring, predicting, and remediating groundwater pollution based on machine learning is further illustrated. The specific steps are as follows: Step 1: Deploy a groundwater monitoring network for the contaminated site, conduct multi-parameter monitoring of groundwater flow direction and water quality, and collect groundwater samples from the contaminated site.

[0019] In step one, a groundwater monitoring network for the contaminated site is established. Based on the geological and hydrogeological conditions of the site and the characteristics of groundwater pollution, natural decay monitoring wells are set up around the delineated pollution range. Background monitoring wells, pollution monitoring wells, and control monitoring wells are also set up. Among them, the pollution monitoring wells include those set up at the pollution source, upstream of the pollution source, and downstream of the pollution source, so as to depict the dynamics of key areas of the pollution plume in detail. At the same time, the monitoring wells set up in the site environmental survey are fully utilized.

[0020] In step one, the water level is monitored through groundwater level observation wells, while water temperature, dissolved oxygen, pH, redox potential, and conductivity are measured on-site using portable instruments.

[0021] For the groundwater samples collected in step one from the contaminated site, analysis and testing were conducted. Based on the rainfall and pollution characteristics of the site, typical groundwater pollution indicators were selected and seasonally sampled and analyzed. Before sample collection, the wells were cleaned three times using a handheld peristaltic pump. Sampling began after the cleaning was deemed satisfactory.

[0022] In this implementation, monitoring wells were constructed in sections upstream, midstream, and downstream of the groundwater flow, based on the regional topography and hydrogeological characteristics. Figure 1The upstream control well (G1) is located 30 m east of the landfill. In the midstream section, based on pollution diffusion patterns, five pollution diffusion monitoring wells (G2, G3, G4, G5, G6) are deployed in a quadrilateral pattern at 30 m intervals around the landfill. Furthermore, to further investigate the attenuation trend of groundwater pollutants along the landfill route, four additional monitoring wells (NG8, NG9, NG10, NG11) are constructed in the downstream section approximately 600 m from the landfill in the first quarter of 2022, following the groundwater flow direction, for a total of 10 groundwater monitoring wells.

[0023] Given the significant seasonal influence of groundwater in the study area, samples were collected separately for the rainy and dry seasons based on rainfall, with a sampling frequency of once per quarter. Before sampling, the wells were cleaned three times using a handheld peristaltic pump. Sampling began only after the cleaning was deemed satisfactory. Sampling bottles were rinsed three times with polyethylene tubing. Water samples were collected using a Bayer tube, and the water level was monitored through a groundwater level observation well. Water temperature, dissolved oxygen, pH, redox potential, and conductivity were measured on-site using portable instruments. Water samples used for inorganic anion testing were filtered through a 0.22 μm filter to remove impurities. Water samples used for metal element testing were filtered through a 0.45 μm filter, and 1–2 drops of concentrated nitric acid were added to adjust the pH to below 2. After collection, the water samples were refrigerated and brought back to the laboratory for analysis.

[0024] The test indicators include: total hardness (TH), total dissolved solids (TDS), chloride (Cl), iron (Fe), manganese (Mn), copper (Cu), zinc (Zn), volatile phenols (VPs), oxygen consumption (COD), and ammonia nitrogen (NH4). + -N), fecal coliform (FC), nitrite (NO2-N), nitrate (NO3-N), cyanide (CN), mercury (Hg), arsenic (As), chromium (Cr), hexavalent chromium (Cd) 6+ ) and lead (Pb).

[0025] Step 2: Reconstruct the time series array and perform data preprocessing, including reading monitoring data in text format, processing mixed data types, finding missing values, removing error values, and standardizing and removing dimensions. During monitoring data processing, time and well number information are extracted, and the data is reconstructed into a three-dimensional array structure of {well number × pollutant × time step}. The reconstructed time series array is used as input for the historical time step data in Step 3.

[0026] Step 3: Based on the groundwater flow direction, the Support Vector Regression (SVR) algorithm is used, with historical time-step data as input, to predict the pollutant concentration in future time steps, thereby constructing a response model for groundwater pollutant migration along the flow direction. For each pollutant, a separate prediction model is trained, and the model performance is evaluated based on the mean squared error. A Gaussian kernel function is used to handle nonlinear relationships and automatically optimize the kernel parameters.

[0027] Step 4: Assess the natural decay performance of groundwater pollutants using trend analysis, calculate the trend of average concentration of each pollutant over time, and use linear regression analysis to calculate the trend slope and statistical significance. p value).

[0028] In step four, when using trend analysis to assess the natural decay performance of groundwater pollutants, the probability of natural decay is calculated based on the rate of decrease and the significance of the trend at a 95% confidence level. Analysis of the natural decay performance of each pollutant as follows: Figure 2 As shown, half of the 19 indicators tested showed a decrease in concentration, with one indicator (hexavalent chromium) exhibiting a 95% probability of a true decline. P >0.7); Nine indicators, including total dissolved solids, chloride, manganese, oxygen consumption, ammonia nitrogen, nitrate, chromium, lead, and mercury, showed a moderate probability of natural decay. P The value is between 0.4 and 0.7; while the total hardness, iron, copper, zinc, volatile phenols, fecal coliforms, nitrite, cyanide, and arsenic are nine pollutants with low natural decay probability. P <0.4). Pollutants are classified according to their decay probability. p The scores are divided into three levels: high (>0.7), medium (0.4-0.7), and low (<0.4) and ranked accordingly. Figure 3 and Figure 4 As shown.

[0029] Step 5: Estimate the occurrence range and threshold of natural decay of groundwater contaminated sites through change point analysis. Use nonparametric bootstrapping to estimate the probability of natural decay of pollutants along groundwater flow. Estimate the occurrence range and threshold of natural decay of groundwater contaminated sites through change point analysis. For pollutants with significant decay trends, conduct concentration change point detection. Use the Bayesian change point detection method to identify key time points of concentration changes. Mark the change point locations on the time series to determine the occurrence range of natural decay and multi-level early warning thresholds.

[0030] For step five, nonparametric bootstrap sampling is used to estimate the probability of pollutants naturally decaying with groundwater flow. Specifically, this involves: uniformly sampling with replacement 1000 times from a given training set, meaning that whenever a sample is selected, it is selected again with equal probability and added back to the training set; specifically, resampling with replacement is performed from the original sample to obtain a resampled set of the same size as the original sample; these resampled data sets are used to estimate the bias; by repeating this process 1000 times, a large number of estimates can be obtained, thus obtaining the sampling distribution of the statistic; these sampling distributions are used to estimate the properties of the population distribution and determine the frequency of occurrence.

[0031] Step Six: Create intuitive and visual charts representing the probability, cumulative distribution, and concentration change points of pollutants as they naturally decay with water flow to characterize the early warning results. Based on these results, develop in-situ remediation methods for groundwater at contaminated sites, categorized and graded according to the early warning results. For the early warning results of natural decay of groundwater pollutants revealed by machine learning in Step Six, and in conjunction with the site's groundwater functional requirements, identify key areas and indicators for in-situ groundwater remediation, and develop remediation methods categorized and graded according to the pollutants.

[0032] like Figure 5 and Figure 6 As shown, where Figure 6 In this study, 2020, 2021, and 2022 represent the years; S1, S2, S3, and S4 represent the first, second, third, and fourth quarters, respectively. Change point analysis was used to estimate the natural decay intervals of groundwater contaminated sites. The results showed that the concentration of hexavalent chromium in groundwater continuously decreased over time. When the pollutant concentration decreased by 0.00596 mg / L per quarter, the cumulative probability of decay reached 80%; when the quarterly concentration decrease exceeded 0.00706 mg / L, the pollutant was essentially completely decayed. Three change point locations (change point 1, change point 2, and change point 3) were identified for the natural decay intervals of hexavalent chromium during the study period, occurring in the first, second, and fourth quarters of 2021, respectively.

[0033] Based on the site's groundwater function requirements, a classification and grading method for in-situ groundwater remediation under natural degradation should be defined. For hexavalent chromium, which has a high probability of natural degradation, the natural degradation process should continue to be monitored. For total dissolved solids, chloride, manganese, oxygen consumption, ammonia nitrogen, nitrate, chromium, lead, and mercury, which have a medium probability of natural degradation, it is recommended to increase the monitoring frequency. For total hardness, iron, copper, zinc, volatile phenols, fecal coliforms, nitrite, cyanide, and arsenic, which have a low probability of natural degradation, active remediation measures should be considered.

[0034] 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 machine learning based groundwater pollution monitoring natural attenuation early warning remediation method, characterized in that, The specific steps are as follows: Step one, layout of the contaminated site groundwater monitoring network, to carry out groundwater flow and water quality monitoring of multiple parameters; and collect groundwater samples from contaminated sites; Step two, reconstruct the time series array, data preprocessing, including reading the text format of monitoring data, processing mixed data types, finding missing values, error values, and standardizing the dimensionless; Step three, according to the groundwater flow direction, using support vector regression (SVR) algorithm, taking historical time step data as input, predicting the concentration of pollutants in the future time step, and building a response model of groundwater pollutant migration along the flow direction; Step four, using trend analysis method to evaluate the natural attenuation performance of groundwater pollutants, calculate the average concentration of each pollutant with time trend, using linear regression analysis to calculate the trend slope and statistical significance (P value) p value); Step five, estimate the occurrence interval and threshold of natural attenuation of groundwater pollution sites through change point analysis, and estimate the probability of natural attenuation of pollutants along the groundwater flow direction using non-parametric bootstrap sampling; Step six, draw intuitive visualization charts of the probability of natural attenuation of pollutants along the water flow, cumulative distribution and concentration change point to represent the early warning results, and develop in-situ classification and grading remediation methods for contaminated sites based on the early warning results.

2. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 1, wherein, In step one, the groundwater monitoring network of the contaminated site is laid out, and according to the geological and hydrogeological conditions of the site, the groundwater pollution characteristics, the natural attenuation monitoring wells are laid out around the delineated pollution range, the background monitoring wells, the pollution monitoring wells and the control monitoring wells are set up, among which the pollution monitoring wells include the monitoring wells set up at the pollution source, the upstream and downstream of the pollution source, to depict the dynamics of the key area of the pollution plume in detail, and make full use of the monitoring wells set up during the site environmental investigation.

3. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 2, wherein, For the water quality multi-parameter monitoring in step one, water level is monitored through groundwater level observation hole, and water temperature, dissolved oxygen, pH, oxidation reduction point, and conductivity are measured on site using portable instruments.

4. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 3, wherein, For the collection of groundwater samples from contaminated sites for analysis and testing in step one, according to the rainfall characteristics and pollution characteristics of the site, typical groundwater pollution indicators are selected for seasonal sampling and analysis. Before sample collection, the well is cleaned three times using a handheld peristaltic pump, and sampling begins after the well is cleaned and qualified.

5. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 4, wherein, In step two, when processing monitoring data, extract time and well number information, and reorganize the data into a three-dimensional array structure of {well number x pollutant x time step}; among which, the reconstructed time series array is used as the input of historical time step data in step three.

6. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 5, wherein, For the response model of groundwater pollutant migration along the flow direction in step three, a prediction model is trained for each pollutant, and the model performance is evaluated based on mean square error. The Gaussian kernel function is used to handle nonlinear relationships and automatically optimize kernel parameters.

7. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 6, wherein, For the trend analysis method in step four, the natural attenuation probability was calculated based on the decline ratio and the significance of the trend at the 95% confidence level. The pollutants were divided into three levels according to the natural attenuation probability: high (>0.7), medium (0.4-0.7), and low (<0.4). p For the trend analysis method in step four, the natural attenuation probability was calculated based on the decline ratio and the significance of the trend at the 95% confidence level. The pollutants were divided into three levels according to the natural attenuation probability: high (>0.7), medium (0.4-0.7), and low (<0.4).

8. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 7, wherein, For step five, through change point analysis to estimate the occurrence interval and threshold of natural attenuation of groundwater pollution sites, concentration change point detection is carried out for pollutants with significant attenuation trend, Bayesian change point detection method is used to identify the key time points of concentration change; mark the change point position on the time series, and determine the occurrence interval and multi-level early warning threshold of natural attenuation.

9. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 8, wherein, For step five, the probability of natural attenuation of pollutants along the groundwater flow direction is estimated using non-parametric bootstrap sampling. Specifically, 1000 uniform resamples with replacement are taken from the given training set, i.e. each time a sample is selected, it is equally likely to be selected again and added to the training set; Specifically, resample with replacement from the original sample to obtain a resample set of the same size as the original sample; use these resample data sets to estimate the bias; by repeating this process 1000 times, a large number of estimates can be obtained, and the sampling distribution of the statistic can be obtained; Using these sampling distributions, the properties of the population distribution are estimated, and the frequency of occurrence is determined.

10. The machine learning based groundwater pollution monitoring natural attenuation early warning remediation method of claim 9, wherein, For step six, based on the machine learning revealed site groundwater pollutant natural attenuation early warning results, combined with the site groundwater function requirements, the key area and attention index of groundwater in-situ remediation are determined, and the remediation method of pollutant classification and grading is formulated.

Images