Prediction Methods and Model Training Methods for the Natural Decay Capacity of Groundwater Pollution

Abstract

This invention discloses a method for predicting the natural decay capacity of groundwater pollution and a model training method. The model training method includes: S1, using polluted groundwater samples from multiple monitoring sites at different time periods as the training sample set, obtaining the microbial community composition, natural decay rate of characteristic pollutants, and natural decay significance of each sample; S2, using the microbial community composition as input and the natural decay significance as output, training a random forest classification model to obtain a first model; S3, using the microbial community composition as input and the natural decay rate as output, training an artificial neural network model to obtain a second model. The prediction method includes: T1, collecting groundwater samples from the site to be predicted, obtaining the microbial community composition, inputting it into the first model, and obtaining a prediction result on the feasibility of natural decay; T2, extracting the abundance of important species from the water samples, inputting it into the second model, and obtaining a prediction result on the natural decay rate.

Description

Technical Field

[0001] This invention relates to the field of environmental protection technology, and in particular to a model training method based on microbial community composition for predicting the natural decay capacity of groundwater pollution, as well as a method for predicting the natural decay capacity of groundwater pollution. Background Technology

[0002] Monitored Natural Attenuation (MNA) is a pollution remediation method that utilizes the natural attenuation processes present at contaminated sites to reduce the concentration and total amount of pollutants, achieving remediation goals within a reasonable timeframe. The basic principle of MNA is to reduce the concentration, toxicity, and mobility of pollutants in soil and groundwater to acceptable levels through planned monitoring strategies without human intervention, based on naturally occurring physical, chemical, and biological processes at the site (including dilution, diffusion, volatilization, adsorption, chemical or biological stabilization, biodegradation, and radioactive attenuation). MNA is suitable for a wide range of contaminated sites, including underground oil tank leak sites, landfill sites, and industrial contaminated sites. The pollutants for which MNA is applicable are primarily organic pollutants, including benzene compounds (BTEX), petroleum hydrocarbons, polycyclic aromatic hydrocarbons, and organic solvents. Compared to other active remediation methods, MNA offers advantages such as lower cost, ease of operation, minimal environmental impact, environmental safety, and thorough pollutant degradation, and is gaining increasing attention.

[0003] Site management by natural attenuation (MNA) is not a passive approach; rather, it requires a series of highly specialized technologies and methodologies, including site investigation and conceptual modeling, risk assessment, evaluation and verification of the effectiveness of natural attenuation, and the construction of monitoring systems. In practical applications, assessing the feasibility of MNA and predicting natural attenuation trends has become a major challenge for its widespread adoption. Taking groundwater pollution as an example, the natural attenuation rate of organic pollutants is influenced by factors such as pollutant properties, hydrogeological conditions, and the composition and activity of functional microorganisms. Conventional prediction methods require acquiring monitoring data over a relatively long period, and their predictions are subject to significant uncertainty. Therefore, currently, there is a lack of effective methods to accurately predict the natural attenuation capacity of groundwater pollution. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the prior art by providing an efficient and accurate method for predicting the natural decay capacity of groundwater pollution and a model training method.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A model training method for predicting the natural decay capacity of groundwater pollution, the model training method comprising the following steps:

[0007] S1. Using polluted groundwater samples from multiple locations at different times in the monitored site as the training sample set, obtain the microbial community composition, natural decay rate of characteristic pollutants, and natural decay significance of each sample.

[0008] S2. Using the microbial community composition as the input value and the natural attenuation significance as the output value, train a random forest classification model to obtain a first model for predicting whether the natural attenuation of groundwater pollution is feasible.

[0009] S3. Using the microbial community composition as the input value and the natural decay rate as the output value, train an artificial neural network model to obtain a second model for predicting the natural decay rate of groundwater pollution.

[0010] In some embodiments, in S1, the concentration of a characteristic pollutant in each sample is measured, and when calculating the natural decay rate, a reaction kinetic equation is used for fitting, outlier samples are discarded from the absolute error vector of the fitting result, and the natural decay rate constant of the characteristic pollutant is obtained, where natural decay rate = characteristic pollutant concentration × natural decay rate constant.

[0011] In some implementations, in S1, the training sample set is divided into a significantly decayed group and a non-significantly decayed group; in S3, the significantly decayed group is used as the second training sample set, and important species and their relative abundances, whose number does not exceed the size of the second training sample set, are selected as input values ​​according to the importance of microbial community composition.

[0012] In some implementations, S301 is included before S3: using the average precision in the random forest classification model to reduce parameters and evaluate the importance of microbial community composition to the significance of natural decay.

[0013] In some embodiments, in S1, the concentration of a characteristic pollutant in each sample is measured, and a reaction kinetic equation is used to fit the sample to obtain the natural decay rate constant of the characteristic pollutant. Samples with a positive natural decay rate constant and a significance P < 0.05 are classified as the significant decay group, and other samples are classified as the insignificant decay group.

[0014] In some implementations, in S3, the second training sample set is cross-validated with the training set containing 80% of the samples and the test set containing 20% ​​of the samples, and the accuracy of the second model is evaluated by mean absolute percentage error.

[0015] In some implementations, S302 is included before S3: screening the number of hidden neurons and the number of hidden layers of the artificial neural network model.

[0016] In some implementations, in S2, model cross-validation is performed synchronously using out-of-bag samples.

[0017] In some implementations, in S1, the characteristic pollutant is organic matter, and the characteristic pollutant types of each sample in the training sample set are basically the same; in S2, the microbial community composition is the relative abundance of genus-level members.

[0018] A method for predicting the natural decay capacity of groundwater pollution, the method comprising the following steps:

[0019] T1. Collect groundwater samples from the site to be predicted, obtain the microbial community composition of the water samples, input them into the first model, and obtain the prediction results of whether the natural decay of characteristic pollutants in the site to be predicted is feasible.

[0020] T2. Extract the abundance of important species in the microbial community composition of the water sample and input it into the second model to obtain the prediction results of the natural decay rate of characteristic pollutants in the site to be predicted.

[0021] The first model and the second model are both trained using the model training method for predicting the natural decay capacity of groundwater pollution as described in any one of claims 1 to 9.

[0022] Due to the application of the above technical solutions, this invention has the following advantages compared with the prior art: This invention predicts the natural decay capacity of polluted groundwater based on microbial community composition. Since the composition of the microbial community is affected by environmental factors such as the nature of pollutants and hydrogeological conditions, it can simultaneously serve as a response to environmental factors. Therefore, using microbial information to predict the natural decay capacity of pollutants has high accuracy and reliability. This invention directly uses groundwater data from monitored sites for model training, resulting in a prediction model with wide applicability. Therefore, it eliminates the need for long-term monitoring of the site to be predicted beforehand, resulting in lower cost, convenience, and time and effort savings. This invention innovatively applies a nested random forest classification model and an artificial neural network model for training, applied to the field of natural decay monitoring technology. Experimental verification shows that the trained model can accurately predict the feasibility and rate of natural decay of groundwater in polluted sites, and the prediction results are reliable. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the training and prediction methods for the groundwater pollution natural attenuation capacity prediction model in this embodiment.

[0024] Figure 2 This is an accuracy test graph for the random forest classification model in this embodiment;

[0025] Figure 3 This is a schematic diagram of the top 15 genus-level members in terms of importance of microbial community composition in this embodiment;

[0026] Figure 4 This is the accuracy test chart of the artificial neural network model in this embodiment. Detailed Implementation

[0027] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more readily understood by those skilled in the art.

[0028] This embodiment provides a method for predicting the natural decay capacity of groundwater pollution based on microbial community composition, which is particularly suitable for predicting the natural decay capacity of organically polluted groundwater. This prediction method involves training a Random Forest classification model and an Artificial Neural Network (ANN) model; therefore, this embodiment also provides a model training method for predicting the natural decay capacity of groundwater pollution.

[0029] See Figure 1 As shown, the basic steps of the model training method in this embodiment are as follows:

[0030] S1. Using polluted groundwater samples from multiple locations at different times in the monitored site as the training sample set, obtain the microbial community composition, natural decay rate of characteristic pollutants, and natural decay significance of each sample.

[0031] S2. Using microbial community composition as input and natural attenuation significance as output, train a random forest classification model to obtain the first model for predicting whether natural attenuation of groundwater pollution is feasible.

[0032] S3. Using microbial community composition as input and natural decay rate as output, train an artificial neural network model to obtain a second model for predicting the natural decay rate of groundwater pollution.

[0033] Based on the first and second models obtained using the above model training methods, the prediction method in this embodiment is as follows:

[0034] T1. Collect groundwater samples from the site to be predicted, obtain the microbial community composition of the water samples, input them into the first model, and obtain the prediction results of whether the natural decay of characteristic pollutants in the site to be predicted is feasible.

[0035] T2. Extract the abundance of important species in the microbial community composition of the water sample and input it into the second model to obtain the prediction results of the natural decay rate of characteristic pollutants in the site to be predicted.

[0036] See Figures 2 to 4 As shown below, the model training method in this embodiment will be further illustrated using benzene-contaminated groundwater as an example.

[0037] S1. Polluted groundwater samples from multiple locations at different times of the monitored site are used as the training sample set. The characteristic pollutant types of each sample in the training sample set are basically the same. In this embodiment, the long-term monitoring data of benzene series pollutant concentration and microbial community composition are specifically included, totaling 454 samples, which constitute the first training sample set.

[0038] The natural decay trend of each sample is fitted based on the reaction kinetic equation, and the natural decay trend is corrected using the outlier test method. Outlier samples are discarded from the absolute error vector of the fitting results to improve the fitting accuracy, thereby obtaining the natural decay rate constant of benzene series compounds, and then calculating the natural decay rate. In this embodiment, the natural decay trend of each sample is fitted based on the first-order reaction kinetic equation, where the natural decay rate = characteristic pollutant concentration × natural decay rate constant.

[0039] Furthermore, based on the fitting results of the natural decay trend, samples with a positive natural decay rate constant and a significance P<0.05 were classified as the significant decay group, while other samples were classified as the insignificant decay group.

[0040] S2. Using the microbial community composition of the first training sample set, i.e., the relative abundance of each genus-level member in each sample's microbial community, as the model input, and the significant attenuation of benzene series compounds in the samples as the model output, a random forest classification model is trained to obtain the first model for predicting the feasibility of natural attenuation of groundwater pollution. Simultaneously, model cross-validation is performed using out-of-bag samples. Figure 2 As shown, in the first model, 76.8% of the samples achieved an accuracy of 80% in predicting the feasibility of natural decay.

[0041] S301. Using the Mean Decrease Accuracy parameter in the random forest classification model, evaluate the importance of microbial community composition to the natural decay of benzene series compounds, and screen out key species. Figure 3 The top 15 genus-level members by importance are listed.

[0042] S302. Before training the artificial neural network model, optimize the number of hidden neurons and hidden layers to reduce the possibility of model overfitting.

[0043] S3. 204 samples showing significant decay of benzene series compounds were selected as the second training sample set. Important species and their relative abundances, ranked by importance of microbial community composition and not exceeding the size of the second training sample set, were selected as input values. Specifically, the relative abundances of the top 203 most important microbial community members were used as model input, and the natural decay rate of benzene series compounds was used as model output. This trained the artificial neural network model, resulting in the second model for predicting the natural decay rate of groundwater pollution. Cross-validation was performed with the training set containing 80% of the samples and the test set containing 20%. The model accuracy was evaluated using the Mean Absolute Percentage Error (MAPE), calculated as follows:

[0044]

[0045] Among them, A i F represents the true value of the test set sample. i This represents the predicted value of a sample in the test set, where n represents the total number of samples in the test set. For example... Figure 4 As shown, in the second model, the MAPE of the predicted natural decay rate of the test set samples is between 2% and 5%, which has high accuracy.

[0046] In summary, the prediction method and model training method for the natural decay capacity of groundwater pollution based on microbial community composition in this embodiment have wide applicability, are simple to operate, and have a short operation time. They can accurately predict the feasibility and natural decay rate of groundwater pollution of various organic pollutants and have good application prospects.

[0047] The above embodiments are only for illustrating the technical concept and features of the present invention. Their purpose is to enable those skilled in the art to understand the content of the present invention and implement it. They should not be used to limit the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A model training method for predicting the natural decay capacity of groundwater pollution, characterized in that, The model training method includes the following steps: S1. Using polluted groundwater samples from multiple monitoring sites at different times as the training sample set, the microbial community composition, natural decay rate of characteristic pollutants, and natural decay significance of each sample are obtained. Specifically, the concentration of characteristic pollutants in each sample is measured, and the reaction kinetic equation is used for fitting to obtain the natural decay rate constant of the characteristic pollutants. Samples with a positive natural decay rate constant and a significance P<0.05 are classified as significant decay group, and other samples are classified as insignificant decay group. Natural decay rate = concentration of characteristic pollutant × natural decay rate constant. S2. Using the microbial community composition as the input value and the natural attenuation significance as the output value, train a random forest classification model to obtain a first model for predicting whether the natural attenuation of groundwater pollution is feasible. S3. Using the significant decay group as the second training sample set, the microbial community composition as the input value, and the natural decay rate as the output value, train the artificial neural network model to obtain a second model for predicting the natural decay rate of groundwater pollution.

2. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 1, characterized in that, In S1, the concentration of the characteristic pollutant in each sample is measured. When calculating the natural decay rate, a reaction kinetic equation is used for fitting. Outlier samples are discarded from the absolute error vector of the fitting result to obtain the natural decay rate constant of the characteristic pollutant.

3. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 1, characterized in that, In S3, important species and their relative abundances, whose numbers do not exceed the size of the second training sample set, are selected as input values ​​according to the importance of the microbial community composition.

4. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 3, characterized in that, Before S3, there is also S301: using the average precision in the random forest classification model to reduce parameters, evaluate the importance of microbial community composition to the significance of natural decay.

5. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 3, characterized in that, In S3, cross-validation is performed on the second training sample set in such a way that the training set contains 80% of the samples and the test set contains 20% of the samples, and the accuracy of the second model is evaluated by the mean absolute percentage error.

6. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 1, characterized in that, Before S3, there is also S302: screening the number of hidden neurons and the number of hidden layers of the artificial neural network model.

7. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 1, characterized in that, In S2, model cross-validation is performed synchronously using out-of-bag samples.

8. The model training method for predicting the natural decay capacity of groundwater pollution according to claim 1, characterized in that, In S1, the characteristic pollutant is organic matter, and the types of characteristic pollutants in each sample in the training sample set are basically the same; in S2, the microbial community composition is the relative abundance of genus-level members.

9. A method for predicting the natural decay capacity of groundwater pollution, characterized in that, The prediction method includes the following steps: T1. Collect groundwater samples from the site to be predicted, obtain the microbial community composition of the water samples, input them into the first model, and obtain the prediction results of whether the natural decay of characteristic pollutants in the site to be predicted is feasible. T2. Extract the abundance of important species in the microbial community composition of the water sample and input it into the second model to obtain the prediction results of the natural decay rate of characteristic pollutants in the site to be predicted. The first model and the second model are both trained using the model training method for predicting the natural decay capacity of groundwater pollution as described in any one of claims 1 to 8.

Images