A method, medium and system for determining health risk deviation of groundwater organic pollution

Abstract

The present invention provides a method, medium and system for determining health risk deviation of groundwater organic pollution, which belongs to the technical field of groundwater organic pollution, and comprises: first, multiple groups of verified groundwater organic pollution data are collected, including groundwater parameters, organic pollutant parameters and climate precipitation parameters. Then, the probability dependence relationship between these parameters and health risk vectors (carcinogenic risk rate, non-carcinogenic risk rate and exposure dose) is analyzed by using Bayesian network, and the main factors affecting health risk are selected. Based on this, a group of health risk equations considering these influencing factors is established. A neural network prediction model with a multi-branch structure is trained by fitting the model and generating a large amount of data by Monte Carlo simulation. The model can predict the health risk vector and risk deviation vector of the groundwater to be tested according to the organic pollution data of the groundwater to be tested, and solves the technical problem that it is difficult to evaluate the deviation of health risk in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of groundwater organic pollution, and in particular, relates to a method, medium and system for determining health risk deviation of groundwater organic pollution. Background Art

[0002] Groundwater pollution is a widespread environmental problem. Due to the impact of industrial development, agricultural activities and domestic water use, many groundwater bodies are damaged by organic pollutants, which seriously threaten the health of local residents. Common groundwater organic pollutants include pesticides, industrial solvents, fuel additives, etc. Once these pollutants penetrate into the groundwater system, they may cause harm to the human body through drinking water or contact. Therefore, accurately assessing the health risks caused by organic pollution in groundwater is crucial to the development of effective pollution prevention and control strategies and risk management plans.

[0003] The existing health risk assessment methods for groundwater organic pollution mainly include the following: 1) Empirical models, which directly calculate health risks based on simple factors such as pollutant concentration and exposure time; 2) Chemical kinetic models, which establish the migration and transformation process of pollutants in groundwater and combine them with exposure assessment models to predict health risks; 3) Statistical models, which use regression analysis and other methods to establish quantitative relationships between pollutant characteristics and health effects. These methods can reflect the impact of groundwater organic pollution on human health to a certain extent, but there are still some limitations:

[0004] 1) The empirical model is too simplistic and cannot fully consider the impact of complex factors such as groundwater quality and climate on health risks.

[0005] 2) Chemical kinetic models require a large number of parameter inputs, making data acquisition difficult in practical applications and limiting model accuracy.

[0006] 3) The statistical model is based on a limited data sample and has weak predictive power for new regions or pollution situations.

[0007] In addition, existing methods usually only give deterministic health risk assessment results, and it is difficult to quantitatively describe the prediction uncertainty. However, in practical applications, due to the complexity of the groundwater system and the limitations of monitoring data, the health risk assessment results often have a certain degree of deviation, which is generally used to evaluate the reliability of health risk prediction. However, existing groundwater organic pollution health risk assessment technologies often find it difficult to evaluate health risk deviations, making it difficult to analyze the reliability of health risk predictions. Summary of the invention

[0008] In view of this, the present invention provides a method, medium and system for determining the deviation of health risk of groundwater organic pollution, which can solve the technical problem that the existing groundwater organic pollution health risk assessment technology is often difficult to evaluate the deviation of health risk.

[0009] The present invention is achieved in that:

[0010] A first aspect of the present invention provides a method for determining health risk deviation of groundwater organic pollution, which comprises the following steps:

[0011] S10, obtaining multiple sets of groundwater organic pollution data whose accuracy has been verified and their corresponding health risk vectors and deviation vectors, wherein the organic pollution data include groundwater parameters, organic pollutant parameters, and climate precipitation parameters;

[0012] S20, using a Bayesian network to establish a probability dependency between each groundwater parameter, organic pollutant parameter, climate precipitation parameter and the health risk vector, and selecting multiple parameters that affect the health risk according to a preset dependency threshold, recorded as an influencing factor group;

[0013] S30. Based on the probability dependence of Bayesian network, a group of groundwater organic pollution health risk equations considering the influencing factor group is established, including: exposure dose equation, carcinogenic risk equation, non-carcinogenic risk equation, comprehensive health risk equation, and health risk deviation equation;

[0014] S40, fitting the groundwater organic pollution health risk equation group using the multiple groups of groundwater organic pollution health risk data to obtain a fitted health risk equation group;

[0015] S50, using the Monte Carlo simulation method, randomly sampling the parameters in the fitted health risk equation group to generate a large amount of simulation data;

[0016] S60, using the large amount of simulation data as a training set and the multiple groups of groundwater organic pollution health risk data as a validation set, training a neural network with a multi-branch structure to obtain a groundwater organic pollution health risk prediction model, which is used to calculate the health risk vector and deviation vector of the groundwater to be tested based on the organic pollution data of the groundwater to be tested;

[0017] S70. Using a pre-trained health risk assessment model, inputting the health risk vector and deviation vector of the groundwater to be tested, obtaining the health risk deviation of the groundwater to be tested, which is used to describe the reliability of the health risk vector.

[0018] The multi-branch neural network includes a groundwater parameter subnetwork, an organic pollutant parameter subnetwork, a climate precipitation parameter subnetwork and a fusion subnetwork.

[0019] The groundwater parameter subnetwork is used to process groundwater parameters. The training input is pH value, dissolved oxygen, conductivity, and redox potential. The training output is the contribution of groundwater parameters to health risks. The structure is a 3-layer fully connected network with 8 hidden layer neurons.

[0020] The organic pollutant parameter subnetwork is used to process organic pollutant parameters. The training input is concentration, molecular weight, octanol-water partition coefficient, and water solubility. The training output is the contribution of organic pollutant parameters to health risks. The structure is a 3-layer fully connected network with 16 hidden layer neurons.

[0021] The climate precipitation parameter subnetwork is used to process climate precipitation parameters. The training input is the average annual precipitation, precipitation intensity, precipitation frequency, and evaporation. The training output is the contribution of climate precipitation parameters to health risks. The structure is a 3-layer fully connected network with 8 hidden layer neurons.

[0022] The fusion sub-network is used to synthesize the outputs of each sub-network, the training input is the output of each sub-network, the training output is the final health risk vector and deviation vector, the structure is a 2-layer fully connected network, and the number of hidden layer neurons is 32;

[0023] The training data of the health risk assessment model are health risk vectors and deviation vectors corresponding to multiple groups of verified groundwater organic pollution data, as well as the reliability of the health risk vectors; wherein the input for training is the health risk vector and deviation vector, and the output of training is the reliability of the health risk vector;

[0024] The structure of the health risk assessment model is a 3-layer fully connected neural network, the number of neurons in the input layer is 6 (corresponding to the dimensions of the health risk vector and the deviation vector), the number of neurons in the hidden layer is 64, and the number of neurons in the output layer is 1 (corresponding to the final health risk deviation);

[0025] The training steps of the health risk assessment model include:

[0026] Step 1, data preprocessing: normalize the training data;

[0027] Step 2, model initialization: randomly initialize model parameters;

[0028] Step 3, forward propagation: calculate the model prediction output;

[0029] Step 4, loss calculation: use mean square error as the loss function;

[0030] Step 5, back propagation: calculate gradients and update model parameters;

[0031] Step 6, iterative training: repeat steps 3-5 until the model converges or reaches the preset number of iterations;

[0032] Step 7: Model Validation: Use the validation set to evaluate model performance and perform fine-tuning if necessary.

[0033] Wherein, the groundwater parameters include pH value, dissolved oxygen, conductivity, and redox potential.

[0034] The organic pollutant parameters include concentration, molecular weight, octanol-water partition coefficient, and water solubility.

[0035] Among them, the climate precipitation parameters include average annual precipitation, precipitation intensity, precipitation frequency, and evaporation.

[0036] The health risk vector is a two-dimensional vector including carcinogenic risk rate, non-carcinogenic risk rate and exposure dose.

[0037] Furthermore, the dependency threshold is preset to 0.6.

[0038] Among them, the method for determining the types and concentrations of organic pollutants in groundwater uses a method of comparing surface enhanced Raman spectroscopy images with standard images.

[0039] Specifically, the following steps are included:

[0040] Preparation of high concentration surface enhanced Raman spectroscopy active substrates:

[0041] Select appropriate substrate materials (such as silver, gold and other metal nanostructures) to prepare a substrate with high surface enhancement effect. This can be achieved through chemical deposition, electrochemical deposition and other methods.

[0042] Establish characteristic peak intensity-concentration standard curve:

[0043] Using standard organic solution samples, Raman spectral data at different concentrations were obtained on the prepared high-concentration surface-enhanced Raman spectroscopy active substrate.

[0044] Analyze the intensity of the characteristic peaks in the Raman spectrum and draw a standard curve of the characteristic peak intensity and concentration. This can be used for subsequent quantitative analysis of samples of unknown concentration.

[0045] Separation and stabilization of organic pollutants:

[0046] Chromatographic separation and other methods are used to separate and extract the target components in complex organic pollutant samples.

[0047] The separated organic pollutants are stably fixed on the prepared surface-enhanced Raman active substrate to facilitate subsequent Raman spectroscopy testing.

[0048] Obtain surface enhanced Raman spectra of organic pollutants:

[0049] The Raman spectrometer was used to test the organic pollutant samples fixed on the substrate to obtain their surface enhanced Raman spectra.

[0050] Data analysis and interpretation of results:

[0051] The measured surface enhanced Raman spectrum of organic pollutants is compared and analyzed with the previously established surface enhanced Raman spectrum fingerprint and standard curve.

[0052] Through comparative analysis, the types and concentrations of organic pollutants in groundwater were finally determined and quantitatively analyzed.

[0053] Wherein, the step S10 includes: first, it is necessary to collect several groups of groundwater organic pollution data whose accuracy has been verified. Each group of data includes groundwater parameters, organic pollutant parameters, climate precipitation parameters, health risk vectors and deviation vectors. Groundwater parameters include pH value, dissolved oxygen content, conductivity, and redox potential; organic pollutant parameters include concentration, molecular weight, octanol-water partition coefficient, and water solubility; climate precipitation parameters include average annual precipitation, precipitation intensity, precipitation frequency, and evaporation. The health risk vector includes carcinogenic risk rate, non-carcinogenic risk rate, and exposure dose. Select several groups of data from the existing verification data, screen and clean them, and input them as basic data.

[0054] Wherein, the step S20 includes: first, constructing a Bayesian probability network model, taking the aforementioned various parameters and health risk vectors as network nodes, and establishing a probability dependency relationship based on the potential causal relationship between them. Through Bayesian inference, the conditional probability distribution of the health risk vector corresponding to each parameter can be obtained, reflecting the degree of influence of each parameter on the health risk. According to the preset dependency threshold, the parameters with greater influence on the health risk are selected as the main influencing factors, which are included in the subsequent health risk equation.

[0055] Wherein, the step S30 includes: based on the probabilistic dependency relationship of the aforementioned Bayesian network, establishing a group of health risk equations for groundwater organic pollution that considers the main influencing factors. The group of equations includes an exposure dose equation, a carcinogenic risk equation, a non-carcinogenic risk equation, a comprehensive health risk equation, and a health risk deviation equation. The parameters involved in each equation have clear physical or chemical meanings, and the calculation steps are also described in detail, which can comprehensively characterize the health risks caused by groundwater organic pollution.

[0056] 1. Exposure dose equation:

[0057] ;

[0058] In the formula, is the exposure dose (mg / kg / day); is the pollutant concentration (mg / L); is the intake rate (L / day); is the exposure frequency (days / year); is the duration of exposure (years); is body weight (kg); is the average time (days); is the partial derivative of the octanol-water partition coefficient with respect to temperature; is the temperature change (°C); is an imaginary unit; is the phase angle (rad); is the water solubility at the jth sampling point (mg / L); is the pH value of the jth sampling point; is the dissolved oxygen at the jth sampling point (mg / L); is the number of sampling points;

[0059] The calculation steps of the exposure dose equation are as follows:

[0060] 1. Calculate the basic exposure dose: ;

[0061] 2. Calculate the temperature correction factor: ;

[0062] 3. Calculate the complex correction factor: ;

[0063] 4. Calculate the water quality correction factor: ;

[0064] 5. Final exposure dose: .

[0065] 2. Carcinogenic risk equation:

[0066] ;

[0067] In the formula, For cancer risk; is the exposure dose (mg / kg / day); is the slope factor (unit ); is the wavelength (m); is the angular frequency (rad / s); is time (s); is the concentration or intensity of the kth environmental factor; is the reference value of environmental factors; is the weight coefficient of the kth environmental factor; is the number of environmental factors; is the steepness coefficient of the logistic function; is the average annual precipitation (mm); is the precipitation threshold (mm);

[0068] Calculation steps:

[0069] Calculate the basic cancer risk: ;

[0070] Calculate the periodic correction factor: ;

[0071] Calculate environmental factor correction: ;

[0072] Calculate precipitation impact factor: ;

[0073] Ultimate cancer risk: .

[0074] 3. Non-carcinogenic risk equation:

[0075] ;

[0076] In the formula, is the hazard quotient (non-carcinogenic risk); is the exposure dose (mg / kg / day); is the reference dose (mg / kg / day); is an imaginary unit; is the molecular weight of the pollutant (g / mol); is the reference molecular weight (g / mol); is the intensity value of the lth precipitation intensity level (mm / h); is the frequency of the lth precipitation intensity level; is the evaporation of the lth precipitation intensity level (mm); is the number of precipitation intensity levels; is the redox potential (mV); is the reference redox potential (mV); is the redox potential normalization factor (mV);

[0077] Calculation steps:

[0078] Calculate the basic hazard quotient: ;

[0079] Calculate the molecular weight correction factor: ;

[0080] Calculate precipitation impact factor: ;

[0081] Calculate the redox potential influencing factor: ;

[0082] Final Hazard Quotient: .

[0083] 4. Comprehensive health risk equation:

[0084]

[0085] In the formula, For comprehensive health risks; For cancer risk; is the hazard quotient (non-carcinogenic risk); is the partial derivative of the distribution coefficient with respect to conductivity; is the conductivity change (μS / cm); is the dissolved oxygen concentration (mg / L); is the reference dissolved oxygen concentration (mg / L); is the dissolved oxygen impact index; is the exposure time (years); is the attenuation coefficient ( );

[0086] Calculation steps:

[0087] Calculate basic comprehensive health risk: ;

[0088] Calculate the conductivity correction factor: ;

[0089] Calculate the dissolved oxygen impact factor: ;

[0090] Compute the time integration factor: ;

[0091] Final comprehensive health risks: .

[0092] 5. Health Risk Bias Equation:

[0093]

[0094] In the formula, for health risk bias; For the predicted overall health risk; is the observed overall health risk; is the partial derivative of water solubility with respect to temperature; is the temperature change (°C); is the octanol-water partition coefficient; is the reference octanol-water partition coefficient; is the steepness coefficient of the logistic function; is the weight of the h-th precipitation feature; is the observed value of the h-th precipitation feature; is the average value of precipitation characteristics; is the standard deviation of precipitation characteristics; is the number of precipitation features;

[0095] Calculation steps:

[0096] Calculate the base deviation: ;

[0097] Calculate the temperature correction factor: ;

[0098] Calculate the factors affecting the octanol-water partition coefficient: ;

[0099] Calculate the factors affecting precipitation characteristics: ;

[0100] Final health risk deviation: .

[0101] Wherein, the step S40 includes: using the multiple groups of groundwater organic pollution health risk data obtained in the above step S10 as training sets, and inputting them into the equation group established in step S30. Using optimization algorithms such as nonlinear least squares method, the parameters in each equation are fitted so that the predicted value is as close to the measured value as possible. Repeat the above optimization process until the fitting effect of each equation reaches the satisfactory accuracy requirement, thereby obtaining a well-fitted health risk equation group.

[0102] Wherein, the step S50 includes: first, determining the probability distribution model of each parameter, and specifying a suitable probability distribution function for each parameter according to the statistical characteristics of the original data. Then, using the Monte Carlo sampling algorithm, a large number of random samplings are performed according to the probability distribution of each parameter to generate tens of thousands of sets of simulation data, covering the main areas of the parameter space. Finally, these simulation data are substituted into the health risk equation group of step S30, and the corresponding health risk vector and deviation vector are calculated as sample data for neural network training.

[0103] Wherein, the step S60 includes: using a neural network with a multi-branch structure, including a groundwater parameter sub-network, an organic pollutant parameter sub-network, a climate precipitation parameter sub-network and a fusion sub-network. The input and output variables and network structure of each sub-network have been clearly described. The large amount of simulated health risk data generated in step S50 is used as a training set, and the multi-branch neural network model is trained using a back propagation algorithm, and finally a prediction model that can predict health risk vectors and deviation vectors based on groundwater, organic pollutants and climate parameters is obtained.

[0104] Wherein, step S70 includes: pre-training a 3-layer fully connected neural network as a health risk assessment model, with the input of health risk vector and deviation vector, and the output of final health risk deviation. In model training, mean square error is used as loss function, and the network parameters are updated by back propagation algorithm until the model performance meets the requirements. The health risk vector and deviation vector of the groundwater to be tested are input into the trained model to obtain the health risk deviation value of the groundwater sample, which is used to describe the reliability of the health risk prediction results.

[0105] The training data of the health risk assessment model are health risk vectors and deviation vectors corresponding to multiple groups of verified groundwater organic pollution data, as well as the reliability of the corresponding health risk vectors. The input for training is the health risk vector and the deviation vector, and the output of training is the reliability of the health risk vector.

[0106] Optionally, the structure of the health risk assessment model is a 3-layer fully connected neural network, the number of neurons in the input layer is 6 (corresponding to the dimensions of the health risk vector and the deviation vector), the number of neurons in the hidden layer is 64, and the number of neurons in the output layer is 1 (corresponding to the final health risk deviation). The training steps of the model include: data preprocessing, model initialization, forward propagation, loss calculation, back propagation, and model verification.

[0107] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining the health risk deviation of groundwater organic pollution.

[0108] A third aspect of the present invention provides a system for determining health risk deviation of organic pollution in groundwater, which includes the above-mentioned computer-readable storage medium.

[0109] Compared with the prior art, the method, medium and system for determining health risk deviation of organic pollution in groundwater provided by the present invention have the following beneficial effects:

[0110] 1. A set of comprehensive health risk assessment equations considering groundwater parameters, organic pollutant parameters and climate conditions was established. These equations not only cover the calculation of exposure dose, carcinogenic risk and non-carcinogenic risk, but also introduce the influence of water quality factors such as temperature, pH value, dissolved oxygen and precipitation characteristics on health risks. Compared with the existing single factor model, this method can more comprehensively characterize the health threat caused by organic pollution in groundwater.

[0111] 2. The Bayesian network method was used to determine the main parameters affecting health risks, providing a basis for the establishment of the health risk equation. The Bayesian network can mine the complex interactive relationships between the parameters and provide an effective analytical method for screening out key influencing factors. This avoids the limitations of relying on subjective experience to select parameters.

[0112] 3. Monte Carlo simulation technology was introduced to generate a large amount of simulated health risk data covering various parameter combinations. These data provide a wide range of sample sets for the subsequent training of health risk prediction models based on machine learning, improving the generalization performance of the model.

[0113] 4. A multi-branch neural network model was constructed, which can output the health risk level and its reliability deviation based on the physical and chemical parameters of groundwater, pollutant characteristics, and climatic conditions. This model takes into account the personalized modeling and comprehensive prediction of each influencing factor, and is superior to the existing single model in terms of accuracy and robustness.

[0114] 5. A health risk reliability assessment model is proposed to quantitatively describe the uncertainty of health risk assessment results by predicting health risk deviations. This provides an important basis for subsequent risk management and helps to formulate more prudent and reliable decision-making plans.

[0115] In general, the present invention solves the technical problem that the existing groundwater organic pollution health risk assessment technology is often difficult to evaluate the deviation of health risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Figure 1 A flow chart of the method provided by the present invention; DETAILED DESCRIPTION

[0117] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0118] like Figure 1 As shown, it is a flow chart of a method for determining health risk deviation of groundwater organic pollution provided by the first aspect of the present invention, and the method comprises the following steps:

[0119] S10, obtaining multiple sets of groundwater organic pollution data with verified accuracy and their corresponding health risk vectors and deviation vectors, the organic pollution data including groundwater parameters, organic pollutant parameters, and climate precipitation parameters;

[0120] S20, using the Bayesian network to establish the probability dependence between each groundwater parameter, organic pollutant parameter, climate precipitation parameter and the health risk vector, and according to the preset dependence threshold, select multiple parameters that affect the health risk and record them as the influencing factor group;

[0121] S30. Based on the probability dependence of Bayesian network, a group of groundwater organic pollution health risk equations considering the influencing factor group is established, including: exposure dose equation, carcinogenic risk equation, non-carcinogenic risk equation, comprehensive health risk equation, and health risk deviation equation;

[0122] S40, fitting a group of health risk equations for organic pollution of groundwater using multiple groups of health risk data for organic pollution of groundwater to obtain a group of fitted health risk equations;

[0123] S50, using the Monte Carlo simulation method, randomly sampling the parameters in the fitted health risk equation group to generate a large amount of simulation data;

[0124] S60, using a large amount of simulated data as a training set and multiple groups of groundwater organic pollution health risk data as a validation set, training a neural network with a multi-branch structure to obtain a groundwater organic pollution health risk prediction model, which is used to calculate the health risk vector and deviation vector of the groundwater to be tested based on the organic pollution data of the groundwater to be tested;

[0125] S70. Using a pre-trained health risk assessment model, inputting a health risk vector and a deviation vector of the groundwater to be tested, and obtaining a health risk deviation of the groundwater to be tested, which is used to describe the reliability of the health risk vector.

[0126] The specific implementation methods of the above steps are described in detail below:

[0127] Step S10: Acquire multiple sets of groundwater organic pollution data whose accuracy has been verified and their corresponding health risk vectors and deviation vectors.

[0128] The specific implementation method of this step is as follows:

[0129] First, it is necessary to collect several sets of groundwater organic pollution data whose accuracy has been verified. Each set of data includes the following parameters:

[0130] 1. Groundwater parameters: including pH value, dissolved oxygen content, conductivity, and redox potential. These parameters reflect the physical and chemical properties of groundwater.

[0131] 2. Organic pollutant parameters: including pollutant concentration, molecular weight, octanol-water partition coefficient, and water solubility. These parameters describe the chemical properties of the pollutants.

[0132] 3. Climate precipitation parameters: including average annual precipitation, precipitation intensity, precipitation frequency, and evaporation. These parameters reflect the local climate conditions.

[0133] 4. Health risk vector: including carcinogenic risk rate, non-carcinogenic risk rate, and exposure dose. These parameters describe the health risks of the population to the pollutant.

[0134] 5. Health risk deviation vector: quantitatively describe the reliability of the health risk vector.

[0135] Select several groups of data from the existing validation data to ensure that they cover the reasonable range of the above parameters and provide basic data for subsequent model establishment and validation. These data need to be screened and cleaned to ensure their accuracy and completeness.

[0136] Step S20: Use the Bayesian network to establish the probabilistic dependency relationship between each parameter and the health risk vector, and select the main parameters that affect the health risk.

[0137] The specific implementation method of this step is as follows:

[0138] 1. Construct a Bayesian probabilistic network model. The aforementioned parameters and health risk vectors are used as network nodes, and probabilistic dependencies are established based on the potential causal relationship between them. The Bayesian network can characterize the complex interactions between the parameters and reflect the intensity of their impact on health risks.

[0139] 2. Calculate the conditional probability distribution between each parameter and the health risk vector. Through Bayesian inference, the conditional probability of the health risk vector corresponding to each parameter can be obtained. These conditional probabilities reflect the degree of influence of each parameter on the health risk.

[0140] 3. According to the preset dependency threshold, select the parameters with greater impact on health risk as the main influencing factors. Generally speaking, parameters with conditional probability greater than 0.6 can be considered as the main influencing factors. These main influencing factors will be used as input variables for subsequent health risk equations.

[0141] Through the above steps, the probability dependence relationship between groundwater parameters, organic pollutant parameters, climate parameters and health risks was established, and the main parameters affecting health risks were determined, which laid the foundation for the subsequent risk equation establishment.

[0142] Step S30: Based on the probabilistic dependency relationship of the Bayesian network, a set of groundwater organic pollution health risk equations is established taking into account the main influencing factors.

[0143] The specific implementation method of this step is as follows:

[0144] 1. Establish the exposure dose equation. The exposure dose is the basis of health risk assessment. Its calculation needs to consider factors such as pollutant concentration, intake rate, exposure time, etc. It is also affected by water quality parameters such as groundwater pH value, dissolved oxygen, and temperature changes. Therefore, the exposure dose equation can be expressed as:

[0145] ;

[0146] in, is the exposure dose, (mg / kg / day), is the pollutant concentration, (mg / L), is the intake rate, (L / day), is the exposure frequency, (days / year), is the duration of exposure, (years), is body weight, (kg), is the average time, (days), is the partial derivative of the octanol-water partition coefficient with respect to temperature, (dimensionless), is the temperature change, (°C), is an imaginary unit, (dimensionless), is the phase angle, (rad), For the Water solubility at each sampling point, (mg / L), For the pH value at each sampling point, (dimensionless), For the Dissolved oxygen at each sampling point, (mg / L), is the number of sampling points, (dimensionless).

[0147] 2. Establish a carcinogenic risk equation. The calculation of carcinogenic risk needs to consider the exposure dose, slope factor, periodic changes in environmental factors (such as temperature and precipitation), and the impact of precipitation on risk. Therefore, the carcinogenic risk equation can be expressed as:

[0148] ;

[0149] in, is the cancer risk, (dimensionless), is the exposure dose, (mg / kg / day), is the slope factor, ( ), is the wavelength, (m), is the angular frequency, (rad / s), is time,(s), For the The concentration or intensity of an environmental factor, (dimensionless), is the reference value of the environmental factor (dimensionless), For the The weight coefficient of each environmental factor, (dimensionless), is the number of environmental factors (dimensionless), is the steepness coefficient of the logistic function, (dimensionless), is the average annual precipitation, (mm), is the precipitation threshold, (mm).

[0150] 3. Establish a non-carcinogenic risk equation. The calculation of non-carcinogenic risk (hazard quotient) needs to consider the exposure dose, reference dose, and the influence of factors such as pollutant molecular weight, precipitation intensity, evaporation, and redox potential. Therefore, the non-carcinogenic risk equation can be expressed as:

[0151] ;

[0152] in, is the hazard quotient, (dimensionless), is the exposure dose, (mg / kg / day), is the reference dose, (mg / kg / day), is an imaginary unit, (dimensionless), is the molecular weight of the pollutant, (g / mol), is the reference molecular weight, (g / mol), For the The intensity value of each precipitation intensity level, (mm / h), For the The frequency of precipitation intensity levels, (dimensionless), For the Evaporation of each precipitation intensity level, (mm), is the number of precipitation intensity levels (dimensionless), is the redox potential, (mV), is the reference redox potential, (mV), is the redox potential normalization factor, (mV).

[0153] 4. Establish a comprehensive health risk equation. The comprehensive health risk is a combination of carcinogenic risk and non-carcinogenic risk, and the effects of conductivity changes, dissolved oxygen content, and exposure time must also be considered. Therefore, the comprehensive health risk equation can be expressed as:

[0154] ;

[0155] in, is the comprehensive health risk, (dimensionless), is the cancer risk, (dimensionless), is the hazard quotient, (dimensionless), is the partial derivative of the distribution coefficient with respect to conductivity, (L / μS), is the conductivity change, (μS / cm), is the dissolved oxygen concentration, (mg / L), is the reference dissolved oxygen concentration, (mg / L), is the dissolved oxygen impact index, (dimensionless), is the exposure time, (years), is the attenuation coefficient, (1 / year).

[0156] 5. Establish a health risk bias equation. The health risk bias quantifies the reliability of predicting health risks, and needs to consider the difference between the predicted value and the measured value, temperature changes, octanol-water partition coefficient, and the influence of precipitation characteristics. Therefore, the health risk bias equation can be expressed as:

[0157] ;

[0158] in, is the health risk deviation, (dimensionless), is the predicted comprehensive health risk, (dimensionless), is the observed comprehensive health risk, (dimensionless), is the partial derivative of water solubility with respect to temperature, (1 / °C), is the temperature change, (°C), is the octanol-water partition coefficient, (dimensionless), is the reference octanol-water partition coefficient, (dimensionless), is the steepness coefficient of the logistic function, (dimensionless), For the The weight of each precipitation feature, (dimensionless), For the Observed values ​​of precipitation characteristics (mm or mm / h, etc.), is the average value of precipitation characteristics (mm or mm / h, etc.), is the standard deviation of precipitation characteristics (mm or mm / h, etc.), is the number of precipitation characteristics, (dimensionless).

[0159] By establishing the above five equations, the health risks caused by organic pollution in groundwater can be fully described, including exposure dose, carcinogenic risk, non-carcinogenic risk, comprehensive health risk and reliability of health risk. These equations cover multiple influencing factors such as groundwater parameters, organic pollutant characteristics, climate conditions, etc., and mathematical models are used to characterize the interaction between them.

[0160] Step S40: Use the verified groundwater organic pollution health risk data to fit the above equation group to obtain a fitted health risk equation group.

[0161] The specific implementation method of this step is as follows:

[0162] 1. The multiple groups of groundwater organic pollution health risk data obtained in the above step S10 are used as training sets and input into the equation group established in step S30.

[0163] 2. Use optimization algorithms such as nonlinear least squares method to fit the parameters in each equation so that the predicted value is as close to the measured value as possible.

[0164] 3. Repeat the above optimization process until the fitting effect of each equation reaches satisfactory accuracy requirements.

[0165] 4. Obtain a fitted set of health risk equations for subsequent health risk prediction.

[0166] Through this step, the theoretically established health risk equations were calibrated and fitted with the actual observation data, ensuring that the equations can accurately describe the health risks caused by organic pollution in groundwater. This laid the foundation for the subsequent prediction model training.

[0167] Step S50: Using the Monte Carlo simulation method, randomly sampling the parameters in the fitted health risk equation group to generate a large amount of simulation data.

[0168] The specific implementation method of this step is as follows:

[0169] 1. Determine the probability distribution model of each parameter. According to the statistical characteristics of the original data, specify a suitable probability distribution function for each parameter, such as normal distribution, lognormal distribution, etc.

[0170] 2. Monte Carlo sampling algorithm is used to perform a large number of random sampling according to the probability distribution of each parameter. Tens of thousands of sets of simulation data are generated, covering the main areas of the parameter space.

[0171] 3. Substitute these simulation data into the health risk equation group established in step S30 to calculate the corresponding health risk vector and deviation vector.

[0172] 4. Organize and save these simulated health risk data as sample data for subsequent neural network training.

[0173] Through Monte Carlo simulation, a large amount of health risk data covering various parameter combinations was generated. These simulated data will provide a wide range of training sets for the next step of neural network training, which will help improve the generalization ability of the model.

[0174] Step S60: Train a multi-branch neural network to obtain a groundwater organic pollution health risk prediction model.

[0175] The specific implementation method of this step is as follows:

[0176] 1. Neural network architecture design:

[0177] - A multi-branch neural network is used, including groundwater parameter sub-network, organic pollutant parameter sub-network, climate precipitation parameter sub-network and fusion sub-network.

[0178] - Groundwater parameter subnetwork: The input is pH value, dissolved oxygen, conductivity, and redox potential, and the output is the contribution of groundwater parameters to health risks. A 3-layer fully connected network is used with 8 hidden layer neurons.

[0179] - Organic pollutant parameter subnetwork: The input is concentration, molecular weight, octanol-water partition coefficient, and water solubility. The output is the contribution of organic pollutant parameters to health risks. A 3-layer fully connected network is used with 16 hidden layer neurons.

[0180] - Climate precipitation parameter subnetwork: The input is the average annual precipitation, precipitation intensity, precipitation frequency, and evaporation, and the output is the contribution of climate precipitation parameters to health risks. A 3-layer fully connected network is used with 8 hidden layer neurons.

[0181] - Fusion sub-network: The input is the output of each sub-network, and the output is the final health risk vector and deviation vector. A 2-layer fully connected network is used, and the number of hidden layer neurons is 32.

[0182] 2. Data preprocessing:

[0183] - Normalize the large amount of simulated health risk data generated in step S50 to improve the training effect.

[0184] 3. Model training:

[0185] - Divide the simulated data into training set and validation set, and train the multi-branch neural network model.

[0186] - The mean square error loss function is used to update the network parameters using the back propagation algorithm.

[0187] - Repeat the steps of forward propagation, loss calculation, and back propagation until the model converges or reaches the preset number of iterations.

[0188] - Use the validation set to evaluate model performance and make adjustments and fine-tuning if necessary.

[0189] 4. Model output:

[0190] - After training, a neural network model is obtained that can predict health risk vectors and deviation vectors based on groundwater parameters, organic pollutant parameters and climate precipitation parameters.

[0191] - The model can be used to predict health risks under any groundwater data.

[0192] Through this step, a neural network prediction model integrating groundwater characteristics, pollutant properties and climate factors was constructed, which can effectively predict the health risks caused by organic pollution in groundwater. This provides a reliable tool for subsequent health risk assessment.

[0193] Step S70: Using a pre-trained health risk assessment model, input the health risk vector and deviation vector of the groundwater to be tested to obtain the health risk deviation of the groundwater.

[0194] The specific implementation method of this step is as follows:

[0195] 1. Health risk assessment model structure:

[0196] - A 3-layer fully connected neural network structure is used, with 6 neurons in the input layer (corresponding to the dimensions of the health risk vector and the deviation vector), 64 neurons in the hidden layer, and 1 neuron in the output layer (corresponding to the final health risk deviation).

[0197] 2. Model training process:

[0198] - Data preprocessing: Normalize the training data.

[0199] - Model initialization: Randomly initialize model parameters.

[0200] - Forward propagation: compute the model's predicted output.

[0201] - Loss calculation: Use mean square error as the loss function.

[0202] - Backpropagation: Calculate gradients and update model parameters.

[0203] - Iterative training: Repeat the forward propagation, loss calculation, and back propagation steps until the model converges or reaches the preset number of iterations.

[0204] - Model Validation: Use the validation set to evaluate model performance and perform fine-tuning if necessary.

[0205] 3. Calculation of health risk deviation:

[0206] - Input the health risk vector and deviation vector of the groundwater to be tested into the trained health risk assessment model.

[0207] - The model output is the health risk deviation of the groundwater to be tested, which is used to describe the reliability of the health risk vector.

[0208] Through this step, the pre-trained health risk assessment model can be used to quickly calculate the health risk deviation of any groundwater sample. The deviation value reflects the reliability of the health risk prediction results and provides a basis for subsequent risk management.

[0209] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for determining the health risk deviation of groundwater organic pollution.

[0210] A third aspect of the present invention provides a system for determining health risk deviation of organic pollution in groundwater, which includes the above-mentioned computer-readable storage medium.

[0211] Specifically, the principle of the present invention is:

[0212] 1. Multi-factor coupling modeling: The formation and evolution of health risks are often affected by multiple factors such as the physical and chemical properties of groundwater, the properties of pollutants, and climatic conditions. In view of this feature, the present invention constructs a set of equations including exposure dose, carcinogenic risk, non-carcinogenic risk, comprehensive health risk, and health risk deviation, which comprehensively describes the impact mechanism of various factors on health risks. For example, water quality parameters such as groundwater pH and dissolved oxygen will affect the migration and transformation process of pollutants, thereby changing the actual exposure dose of the human body; temperature changes will change the physical and chemical properties of pollutants, thereby affecting their toxic effects; precipitation intensity and frequency will change the migration path and exposure pathway of pollutants. These complex interactions make the accurate assessment of health risks very critical.

[0213] 2. Probabilistic graphical model analysis: When establishing the health risk equation, the present invention uses Bayesian network technology to explore the probabilistic dependency relationship between various parameters. Bayesian network is a directed acyclic graph that can effectively characterize the causal relationship and conditional probability distribution between variables. By introducing nodes such as groundwater parameters, organic pollutant parameters and climate parameters into the network and establishing connections between them and the health risk vector, the relative importance of each parameter to health risk can be analyzed. This provides a basis for parameter selection in subsequent equation establishment and avoids the drawbacks of relying on subjective experience. At the same time, the inference mechanism of the Bayesian network also provides support for the interpretability of health risk prediction.

[0214] 3. Machine learning model training: In order to improve the accuracy and generalization ability of health risk prediction, the present invention adopts a neural network model based on deep learning. The model adopts a multi-branch structure to extract the contribution of groundwater parameters, organic pollutant parameters and climate parameters to health risks respectively, and integrates the output of each branch in the fusion layer to obtain the final risk prediction result. Such a design can not only capture the personalized characteristics of various influencing factors, but also make full use of the complex interactive relationship between them. In order to train this complex neural network model, the present invention uses the Monte Carlo simulation method to generate a large amount of simulated health risk data as training samples. This not only expands the coverage of the training set, but also improves the prediction ability of the model in new regions or pollution situations.

[0215] 4. Health risk reliability assessment: In addition to predicting the health risk level itself, the present invention also proposes a method for assessing health risk deviation. This method establishes an independent neural network model, with the predicted health risk vector and deviation vector as input, and the final health risk deviation value as output. This deviation value quantitatively describes the reliability of the health risk prediction results and provides valuable reference information for decision makers. The introduction of this reliability assessment makes up for the deficiency of existing methods that only give deterministic results, and helps decision makers make more prudent risk control measures.

[0216] In general, the method for determining the health risk of organic pollution in groundwater proposed in this paper fully integrates advanced technologies such as probabilistic graphical models, mathematical analysis and machine learning, and has made great innovations in parameter selection, modeling mechanism and result interpretation. Through technical means such as multi-factor coupling modeling, probabilistic inference analysis and deep learning training, this method can more comprehensively and accurately assess the health risks caused by organic pollution in groundwater, and quantitatively give the reliability of risk prediction.

[0217] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention is provided below: This embodiment 1 includes the following steps:

[0218] Step S10: Acquire multiple sets of groundwater organic pollution data whose accuracy has been verified and their corresponding health risk vectors and deviation vectors.

[0219] The specific implementation method of this step is as follows:

[0220] First, it is necessary to collect several sets of groundwater organic pollution data whose accuracy has been verified. Each set of data includes the following parameters:

[0221] 1. Groundwater parameters: including pH value ( ), dissolved oxygen content ( ), conductivity( ), redox potential ( ). These parameters reflect the physical and chemical properties of groundwater.

[0222] 2. Organic pollutant parameters: including pollutant concentration ( )、Molecular weight( ), octanol-water partition coefficient ( ), water solubility ( ). These parameters describe the chemical nature of the pollutant.

[0223] 3. Climate precipitation parameters: including annual average precipitation ( ), precipitation intensity ( )、Precipitation frequency( )、evaporation( ). These parameters reflect local climatic conditions.

[0224] 4. Health risk vector: including cancer risk rate ( ), non-cancer risk rate ( ), exposure dose ( ). These parameters describe the health risk to a population from exposure to the pollutant.

[0225] 5. Health risk deviation vector: quantitatively describe the reliability of the health risk vector, denoted as .

[0226] Select several groups of data from the existing validation data to ensure that they cover the reasonable range of the above parameters and provide basic data for subsequent model establishment and validation. These data need to be screened and cleaned to ensure their accuracy and completeness.

[0227] Step S20: Use the Bayesian network to establish the probabilistic dependency relationship between each parameter and the health risk vector, and select the main parameters that affect the health risk.

[0228] The specific implementation method of this step is as follows:

[0229] 1. Construct a Bayesian probabilistic network model. The aforementioned parameters and health risk vectors are used as network nodes, and probabilistic dependencies are established based on the potential causal relationship between them. The Bayesian network can characterize the complex interactions between the parameters and reflect the intensity of their impact on health risks.

[0230] 2. Calculate the conditional probability distribution between each parameter and the health risk vector. Through Bayesian inference, the conditional probability distribution of the health risk vector corresponding to each parameter can be obtained. ,in represents groundwater parameters, organic pollutant parameters or climate precipitation parameters. These conditional probabilities reflect the impact of each parameter on health risks.

[0231] 3. According to the preset dependency threshold , select the parameters that have a greater impact on health risks as the main influencing factors. Generally speaking, when the conditional probability is greater than When the parameter is , it can be considered as the main influencing factor. These main influencing factors will be used as input variables for the subsequent health risk equation.

[0232] Through the above steps, the probability dependence relationship between groundwater parameters, organic pollutant parameters, climate parameters and health risks was established, and the main parameters affecting health risks were determined, which laid the foundation for the subsequent risk equation establishment.

[0233] Step S30: Based on the probabilistic dependency relationship of the Bayesian network, a set of groundwater organic pollution health risk equations is established taking into account the main influencing factors.

[0234] The specific implementation method of this step is as follows:

[0235] 1. Establish the exposure dose equation. The exposure dose is the basis of health risk assessment. Its calculation needs to consider factors such as pollutant concentration, intake rate, exposure time, etc. It is also affected by water quality parameters such as groundwater pH value, dissolved oxygen, and temperature changes. Therefore, the exposure dose equation can be expressed as:

[0236]

[0237] in, is the exposure dose (mg / kg / day), is the pollutant concentration (mg / L), is the intake rate (L / day), is the exposure frequency (days / year), is the exposure duration (years), is body weight (kg), is the average time (days), is the partial derivative of the octanol-water partition coefficient with respect to temperature (dimensionless), is the temperature change (°C), is an imaginary unit (dimensionless), is the phase angle (rad), For the Water solubility at each sampling point (mg / L), For the pH value of each sampling point (dimensionless), For the Dissolved oxygen at each sampling point (mg / L), is the number of sampling points (dimensionless).

[0238] 2. Establish a carcinogenic risk equation. The calculation of carcinogenic risk needs to consider the exposure dose, slope factor, periodic changes in environmental factors (such as temperature and precipitation), and the impact of precipitation on risk. Therefore, the carcinogenic risk equation can be expressed as:

[0239]

[0240] in, is the cancer risk (dimensionless), is the exposure dose (mg / kg / day), is the slope factor ( ), is the wavelength (m), is the angular frequency (rad / s), is time (s), For the The concentration or intensity of an environmental factor (dimensionless), is the reference value of the environmental factor (dimensionless), For the The weight coefficient of each environmental factor (dimensionless), is the number of environmental factors (dimensionless), is the steepness coefficient of the logistic function (dimensionless), is the average annual precipitation (mm), is the precipitation threshold (mm).

[0241] 3. Establish a non-carcinogenic risk equation. The calculation of non-carcinogenic risk (hazard quotient) needs to consider the exposure dose, reference dose, and the influence of factors such as pollutant molecular weight, precipitation intensity, evaporation, and redox potential. Therefore, the non-carcinogenic risk equation can be expressed as:

[0242]

[0243] in, is the hazard quotient (dimensionless), is the exposure dose (mg / kg / day), is the reference dose (mg / kg / day), is an imaginary unit (dimensionless), is the molecular weight of the pollutant (g / mol), is the reference molecular weight (g / mol), For the The intensity value of each precipitation intensity level (mm / h), For the The frequency of precipitation intensity levels (dimensionless), For the Evaporation of each precipitation intensity level (mm), is the number of precipitation intensity levels (dimensionless), is the redox potential (mV), is the reference redox potential (mV), is the redox potential normalization factor (mV).

[0244] 4. Establish a comprehensive health risk equation. The comprehensive health risk is a combination of carcinogenic risk and non-carcinogenic risk, and the effects of conductivity changes, dissolved oxygen content, and exposure time must also be considered. Therefore, the comprehensive health risk equation can be expressed as:

[0245]

[0246] in, is the comprehensive health risk (dimensionless), is the cancer risk (dimensionless), is the hazard quotient (dimensionless), is the partial derivative of the distribution coefficient with respect to conductivity (L / μS), is the conductivity change (μS / cm), is the dissolved oxygen concentration (mg / L), is the reference dissolved oxygen concentration (mg / L), is the dissolved oxygen impact index (dimensionless), is the exposure time (years), is the attenuation coefficient (1 / year).

[0247] 5. Establish a health risk bias equation. The health risk bias quantifies the reliability of predicting health risks, and needs to consider the difference between the predicted value and the measured value, temperature changes, octanol-water partition coefficient, and the influence of precipitation characteristics. Therefore, the health risk bias equation can be expressed as:

[0248]

[0249] in, is the health risk deviation (dimensionless), is the predicted comprehensive health risk (dimensionless), is the observed comprehensive health risk (dimensionless), is the partial derivative of water solubility with respect to temperature (1 / °C), is the temperature change (°C), is the octanol-water partition coefficient (dimensionless), is the reference octanol-water partition coefficient (dimensionless), is the steepness coefficient of the logistic function (dimensionless), For the The weight of each precipitation feature (dimensionless), For the Observed values ​​of precipitation characteristics (mm or mm / h, etc.), is the average value of precipitation characteristics (mm or mm / h, etc.), is the standard deviation of precipitation characteristics (mm or mm / h, etc.), is the number of precipitation characteristics (dimensionless).

[0250] By establishing the above five equations, the health risks caused by organic pollution in groundwater can be fully described, including exposure dose, carcinogenic risk, non-carcinogenic risk, comprehensive health risk and reliability of health risk. These equations cover multiple influencing factors such as groundwater parameters, organic pollutant characteristics, climate conditions, etc., and mathematical models are used to characterize the interaction between them.

[0251] Step S40: Use the verified groundwater organic pollution health risk data to fit the above equation group to obtain a fitted health risk equation group.

[0252] The specific implementation method of this step is as follows:

[0253] 1. The multiple groups of groundwater organic pollution health risk data obtained in the above step S10 are used as training sets and input into the equation group established in step S30.

[0254] 2. Use optimization algorithms such as nonlinear least squares method to fit the parameters in each equation so that the predicted value As close to the measured value as possible The optimization objective can be expressed as:

[0255]

[0256] in, represents the set of parameters in the equation, is the number of training samples.

[0257] 3. Repeat the above optimization process until the fitting effect of each equation reaches the satisfactory accuracy requirement, that is, ,in is the preset accuracy threshold.

[0258] 4. Obtain a fitted set of health risk equations for subsequent health risk prediction.

[0259] Through this step, the theoretically established health risk equations were calibrated and fitted with the actual observation data, ensuring that the equations can accurately describe the health risks caused by organic pollution in groundwater. This laid the foundation for the subsequent prediction model training.

[0260] Step S50: Using the Monte Carlo simulation method, randomly sampling the parameters in the fitted health risk equation group to generate a large amount of simulation data.

[0261] The specific implementation method of this step is as follows:

[0262] 2. Use the Monte Carlo sampling algorithm to perform a large number of random samplings based on the probability distribution of each parameter. Parameters , can be generated by the following formula Random samples:

[0263]

[0264] in, For parameters The inverse function of the cumulative distribution function, To obey uniform distribution A random number.

[0265] 3. Substitute these simulated parameter data into the health risk equations established in step S30 to calculate the corresponding health risk vector and the deviation vector .

[0266] 4. Organize and save these simulated health risk data as sample data for subsequent neural network training.

[0267] Through Monte Carlo simulation, a large amount of health risk data covering various parameter combinations was generated. These simulated data will provide a wide range of training sets for the next step of neural network training, which will help improve the generalization ability of the model.

[0268] Step S60: Train a multi-branch neural network to obtain a groundwater organic pollution health risk prediction model.

[0269] The specific implementation method of this step is as follows:

[0270] 1. Neural network architecture design:

[0271] A neural network with a multi-branch structure is adopted, including groundwater parameter sub-network, organic pollutant parameter sub-network, climate precipitation parameter sub-network and fusion sub-network.

[0272] Groundwater parameter subnetwork: Input is , , , , the output is the contribution of groundwater parameters to health risks, a 3-layer fully connected network is used, and the number of hidden layer neurons is 8. The output of this sub-network is recorded as .

[0273] Organic pollutant parameter subnetwork: Input is , , , , the output is the contribution of organic pollutant parameters to health risks, a 3-layer fully connected network is used, and the number of hidden layer neurons is 16. The output of this sub-network is recorded as .

[0274] Climate precipitation parameter subnetwork: Input is , , , , the output is the contribution of climate precipitation parameters to health risks, using a 3-layer fully connected network with 8 hidden layer neurons. The output of this subnetwork is .

[0275] Fusion sub-network: input is the output of each sub-network , , , the output is the final health risk vector and the deviation vector ,A 2-layer fully connected network is adopted, and the number of neurons in the hidden layer is 32.

[0276] 2. Data preprocessing:

[0277] The large amount of simulated health risk data generated in step S50 is normalized to improve the training effect. The normalized data is recorded as and .

[0278] 3. The steps of model training adopt the common model training steps in the prior art, which are specifically described as follows:

[0279] Divide the simulated data into training set and validation set;

[0280] Using mean square error loss function ,in, Represents the predicted output vector of the network on the i-th training sample, and uses the back propagation algorithm to update the network parameters , the specific update rules are:

[0281] ,in, is the learning rate;

[0282] Repeat the steps of forward propagation, loss calculation, and back propagation until the model converges or reaches the preset number of iterations;

[0283] Use the validation set to evaluate model performance and make adjustments and fine-tuning if necessary.

[0284] 4. Model output:

[0285] After training, a neural network model is obtained that can predict health risk vectors and deviation vectors based on groundwater parameters, organic pollutant parameters and climate precipitation parameters.

[0286] The model can be used to predict health risks under any groundwater data.

[0287] Through this step, a neural network prediction model integrating groundwater characteristics, pollutant properties and climate factors was constructed, which can effectively predict the health risks caused by organic pollution in groundwater. This provides a reliable tool for subsequent health risk assessment.

[0288] Step S70: Using a pre-trained health risk assessment model, input the health risk vector and deviation vector of the groundwater to be tested to obtain the health risk deviation of the groundwater.

[0289] The specific implementation method of this step is as follows:

[0290] 1. Health risk assessment model structure:

[0291] A three-layer fully connected neural network structure is adopted, with 6 neurons in the input layer (corresponding to the dimensions of the health risk vector and the deviation vector), 64 neurons in the hidden layer, and 1 neuron in the output layer (corresponding to the final health risk deviation).

[0292] The parameters of this model are .

[0293] 2. Model training process:

[0294] Data preprocessing: Normalize the training data.

[0295] Model initialization: Randomly initialize model parameters .

[0296] Forward propagation: computing the model’s predicted output ,in are the input health risk vector and deviation vector.

[0297] Loss calculation: Use mean square error as the loss function , Represents the predicted output vector of the network on the i-th training sample.

[0298] Backpropagation: Calculating Gradients , and update the model parameters .

[0299] Iterative training: Repeat the forward propagation, loss calculation, and back propagation steps until the model converges or reaches the preset number of iterations.

[0300] Model Validation: Use the validation set to evaluate model performance and perform fine-tuning if necessary.

[0301] 3. Calculation of health risk deviation:

[0302] The health risk vector of the groundwater to be tested and the deviation vector Input into the trained health risk assessment model.

[0303] The model output is the health risk deviation value of the groundwater to be tested. , used to describe the reliability of health risk vectors.

[0304] Through this step, the pre-trained health risk assessment model can be used to quickly calculate the health risk deviation of any groundwater sample. The deviation value reflects the reliability of the health risk prediction results and provides a basis for subsequent risk management.

[0305] In summary, the method for determining the health risk of organic pollution in groundwater proposed in the present invention includes the following 7 steps:

[0306] 1. Obtain verified data on organic pollution in groundwater and its health risk;

[0307] 2. Establish a Bayesian network model to determine the main parameters affecting health risks;

[0308] 3. Based on the Bayesian network, establish a set of equations including exposure dose, carcinogenic risk, non-carcinogenic risk, comprehensive health risk and health risk deviation;

[0309] 4. Use the verified data to fit the equations to obtain the final health risk equations;

[0310] 5. Use Monte Carlo simulation to generate a large amount of simulated health risk data;

[0311] 6. Train a multi-branch neural network model to predict health risk vectors and deviation vectors;

[0312] 7. Use the trained health risk assessment model to calculate the health risk deviation of the groundwater to be tested.

[0313] This method uses probabilistic graphical models, mathematical modeling, machine learning and other technologies to comprehensively and accurately assess the health risks caused by organic pollution in groundwater, and provide a scientific basis for groundwater risk management. Each step is closely linked to each other, forming a complete health risk determination framework.

[0314] In step S50, a large amount of simulated health risk data is generated using the Monte Carlo sampling method. Parameters , first determine its probability distribution function , and then generate it by inverse transform sampling Random samples:

[0315]

[0316] in, For parameters The inverse function of the cumulative distribution function, To obey uniform distribution In this way, a set of simulated data covering the main areas of the parameter space is obtained.

[0317] In step S60, a multi-branch neural network structure is constructed to predict health risks. Among them, the groundwater parameter subnetwork, organic pollutant parameter subnetwork and climate precipitation parameter subnetwork respectively extract the contribution of these three types of parameters to health risks, and the fusion subnetwork integrates the outputs of each subnetwork to generate the final health risk vector and deviation vector. The training of the entire neural network adopts the standard back propagation algorithm, and the network parameters are updated by minimizing the mean square error loss function, and finally a model that can accurately predict health risks is obtained.

[0318] In step S70, a 3-layer fully connected neural network is pre-trained as a health risk assessment model. The input of the model is the health risk vector and the deviation vector, and the output is the final health risk deviation. During the training process, the mean square error is also used as the loss function, and the network parameters are updated using the back propagation algorithm until the model performance meets the requirements. The health risk vector and deviation vector of the groundwater to be tested are input into the trained model to obtain the health risk deviation value of the groundwater sample, which is used to describe the reliability of the health risk prediction results.

[0319] In summary, the method for determining the health risk of organic pollution in groundwater proposed in this Example 1 includes the following 7 steps:

[0320] 1. Obtain verified data on organic pollution in groundwater and its health risk;

[0321] 2. Establish a Bayesian network model to determine the main parameters affecting health risks;

[0322] 3. Based on the Bayesian network, establish a set of equations including exposure dose, carcinogenic risk, non-carcinogenic risk, comprehensive health risk and health risk deviation;

[0323] 4. Use the verified data to fit the equations to obtain the final health risk equations;

[0324] 5. Use Monte Carlo simulation to generate a large amount of simulated health risk data;

[0325] 6. Train a multi-branch neural network model to predict health risk vectors and deviation vectors;

[0326] 7. Use the trained health risk assessment model to calculate the health risk deviation of the groundwater to be tested.

[0327] Example 1 Through the above steps, a comprehensive and accurate groundwater organic pollution health risk assessment framework was established. The framework combines Bayesian networks, mathematical modeling, and deep learning technologies, and can predict the corresponding health risk level and the reliability of risk prediction based on groundwater, pollutants, and climate parameters.

[0328] In order to better understand and implement the present invention, the following is an example 2 of a specific application scenario of the present invention: A groundwater source in a certain area has long been polluted by a variety of organic compounds, mainly including pesticides, industrial solvents, and fuel additives. These pollutants enter the groundwater system through leakage, overflow, etc., posing a serious threat to the water safety and health of local residents. In order to evaluate the health risks caused by this organic pollution of groundwater, the local environmental protection department decided to use the health risk determination method proposed in the present invention for analysis.

[0329] First, the environmental protection department organized a professional team to monitor and sample the groundwater in the area for half a year. Groundwater samples were collected from different geographical locations and hydrogeological conditions. Each sample was tested for comprehensive physical and chemical indicators and organic pollutant composition, including:

[0330] Groundwater parameters:

[0331] pH value:

[0332] Dissolved Oxygen: mg / L

[0333] Conductivity: μS / cm

[0334] Redox potential: mV

[0335] Organic pollutant parameters:

[0336] Pollutant concentration ( ):

[0337] Pesticide A: mg / L

[0338] Industrial solvent B: mg / L

[0339] Fuel Additive C: mg / L

[0340] Molecular weight ( ):

[0341] Pesticide A: g / mol

[0342] Industrial solvent B: g / mol

[0343] Fuel Additive C: g / mol

[0344] Octanol-water partition coefficient ( ):

[0345] Pesticide A:

[0346] Industrial solvent B:

[0347] Fuel Additive C:

[0348] water solubility ( ):

[0349] Pesticide A: mg / L

[0350] Industrial solvent B: mg / L

[0351] Fuel Additive C: mg / L

[0352] Climate precipitation parameters:

[0353] Average annual precipitation: mm

[0354] Precipitation intensity: mm / h

[0355] Precipitation frequency:

[0356] Evaporation capacity: mm

[0357] Among them, precipitation intensity, frequency and evaporation correspond to precipitation events of three different intensity levels.

[0358] After collecting the above groundwater organic pollution data, the environmental protection department then evaluated the health risks. First, the Bayesian network was used to analyze the probability dependence between each parameter and the health risk vector. Based on the network analysis results, the following six main influencing factors were identified: pollutant concentration ( ), octanol-water partition coefficient ( )、pH value( )、Dissolved oxygen( ), average annual precipitation ( ) and temperature changes ( ). These parameters will serve as input variables for subsequent health risk equations.

[0359] Next, based on the identified main influencing factors, the environmental protection department established a health risk assessment equation group for groundwater organic pollution, including:

[0360] 1. Exposure dose equation:

[0361]

[0362] in, is the exposure dose (mg / kg / day), is the pollutant concentration (mg / L), is the intake rate (L / day), is the exposure frequency (day / year), is the exposure duration (year), is body weight (kg), is the average time (day), is the partial derivative of the octanol-water partition coefficient with respect to temperature (dimensionless), is the temperature change (°C), is an imaginary unit (dimensionless), is the phase angle (rad), For the Water solubility at each sampling point (mg / L), For the pH value of each sampling point (dimensionless), For the Dissolved oxygen at each sampling point (mg / L), is the number of sampling points (dimensionless).

[0363] 2.Carcinogenic risk equation:

[0364]

[0365] in, is the cancer risk (dimensionless), is the exposure dose (mg / kg / day), is the slope factor ( ), is the wavelength (m), is the angular frequency (rad / s), is time (s), For the The concentration or intensity of an environmental factor (dimensionless), is the reference value of the environmental factor (dimensionless), For the The weight coefficient of each environmental factor (dimensionless), is the number of environmental factors (dimensionless), is the steepness coefficient of the logistic function (dimensionless), is the average annual precipitation (mm), is the precipitation threshold (mm).

[0366] 3. Non-carcinogenic risk equation:

[0367]

[0368] in, is the hazard quotient (dimensionless), is the exposure dose (mg / kg / day), is the reference dose (mg / kg / day), is an imaginary unit (dimensionless), is the molecular weight of the pollutant (g / mol), is the reference molecular weight (g / mol), For the The intensity value of each precipitation intensity level (mm / h), For the The frequency of precipitation intensity levels (dimensionless), For the Evaporation of each precipitation intensity level (mm), is the number of precipitation intensity levels (dimensionless), is the redox potential (mV), is the reference redox potential (mV), is the redox potential normalization factor (mV).

[0369] 4. Comprehensive health risk equation:

[0370]

[0371] in, is the comprehensive health risk (dimensionless), is the cancer risk (dimensionless), is the hazard quotient (dimensionless), is the partial derivative of the distribution coefficient with respect to conductivity (L / μS), is the conductivity change (μS / cm), is the dissolved oxygen concentration (mg / L), is the reference dissolved oxygen concentration (mg / L), is the dissolved oxygen impact index (dimensionless), is the exposure time (year), is the attenuation coefficient (1 / year).

[0372] 5. Health Risk Bias Equation:

[0373]

[0374] in, is the health risk deviation (dimensionless), is the predicted comprehensive health risk (dimensionless), is the observed comprehensive health risk (dimensionless), is the partial derivative of water solubility with respect to temperature (1 / °C), is the temperature change (°C), is the octanol-water partition coefficient (dimensionless), is the reference octanol-water partition coefficient (dimensionless), is the steepness coefficient of the logistic function (dimensionless), For the The weight of each precipitation feature (dimensionless), For the Observed values ​​of precipitation characteristics (mm or mm / h, etc.), is the average value of precipitation characteristics (mm or mm / h, etc.), is the standard deviation of precipitation characteristics (mm or mm / h, etc.), is the number of precipitation characteristics (dimensionless).

[0375] Next, the environmental protection department will The monitoring data of groundwater samples were substituted into the above health risk equation, and the optimal values ​​of each parameter were obtained through nonlinear fitting. Taking pesticide A as an example, the fitting results are as follows:

[0376] Exposure dose: mg / kg / day

[0377] Cancer Risk:

[0378] Non-cancer risks:

[0379] Comprehensive health risks:

[0380] Health risk bias:

[0381] Through similar calculations, the environmental protection department obtained the health risk assessment results of all three pollutants at various groundwater points. In order to further expand the data set, the Monte Carlo simulation method was used to generate These data cover various combinations of groundwater parameters, organic pollutant parameters and climate conditions, providing a wide range of sample sets for subsequent neural network model training.

[0382] With the above basic data, the environmental protection department began to build a groundwater organic pollution health risk prediction model. A multi-branch neural network structure was used, including groundwater parameter sub-network, organic pollutant parameter sub-network, climate precipitation parameter sub-network and fusion sub-network. The specific input and output of each sub-network are as follows:

[0383] - Groundwater parameters subnetwork:

[0384] Input - pH, dissolved oxygen, conductivity, redox potential

[0385] Output - Contribution of groundwater parameters to health risks

[0386] - Organic pollutants parameters sub-network:

[0387] Input - Contaminant concentration, molecular weight, octanol-water partition coefficient, water solubility

[0388] Output - Contribution of organic pollutant parameters to health risks

[0389] - Climate precipitation parameter sub-network:

[0390] Input - Average annual precipitation, precipitation intensity, precipitation frequency, evaporation

[0391] Output - Contribution of climate precipitation parameters to health risks

[0392] - Fusion sub-network:

[0393] Input - Output of each sub-network

[0394] Output - Final health risk vector (CR, HQ, ED) and deviation vector (ΔR)

[0395] The M=10,000 data sets generated by the Monte Carlo simulation were divided into training sets and validation sets, and the multi-branch neural network model was trained using the back propagation algorithm. After multiple rounds of iterative optimization, the prediction effect of the model on the validation set reached satisfactory accuracy requirements. In this way, the environmental protection department obtained a reliable health risk prediction model for groundwater organic pollution.

[0396] Finally, the environmental protection department also trained a three-layer fully connected neural network as an evaluation model for the reliability of health risk prediction results. The input of the model is the predicted health risk vector and deviation vector, and the output is the final health risk deviation value. In the model training, the environmental protection department used the n=30 groups of measured health risk data accumulated in the early stage as samples, used the mean square error as the loss function, and continuously optimized the network parameters through back propagation until the accuracy of the predicted deviation value reached the standard.

[0397] With the above two prediction models, the environmental protection department can carry out organic pollution health risk assessment for the groundwater in the area. For any groundwater sampling point, the monitored groundwater parameters, organic pollutant parameters and climate conditions can be input into the multi-branch neural network model to predict the health risk level of the groundwater, including carcinogenic risk, non-carcinogenic risk and comprehensive health risk. At the same time, the predicted health risk vector and deviation vector can also be input into a special assessment model to obtain the reliability index of the health risk prediction results.

[0398] Taking a groundwater sampling point as an example, its monitoring data is as follows:

[0399] - Groundwater parameters:

[0400] pH = 6.9, dissolved oxygen = 4.5 mg / L, conductivity = 620 μS / cm, redox potential = 340mV

[0401] - Organic pollutants parameters:

[0402] Pesticide A concentration = 0.13 mg / L, molecular weight = 269 g / mol, octanol-water partition coefficient = 3.1, water solubility = 118 mg / L

[0403] Industrial solvent B concentration = 0.32 mg / L, molecular weight = 162 g / mol, octanol-water partition coefficient = 1.8, water solubility = 960 mg / L

[0404] Fuel Additive C concentration = 0.17 mg / L, molecular weight = 201 g / mol, octanol-water partition coefficient = 2.5, water solubility = 195 mg / L

[0405] - Climate precipitation parameters:

[0406] Average annual precipitation = 920 mm, precipitation intensity = (2.8, 5.3, 8.0) mm / h, precipitation frequency = (0.35, 0.32, 0.23), evaporation = (0.9, 1.3, 1.5) mm

[0407] By inputting these data into the trained multi-branch neural network model, the health risk prediction results of the groundwater can be obtained:

[0408] - Health risks of pesticide A:

[0409] Exposure dose mg / kg / day

[0410] Cancer risk

[0411] Non-cancer risk

[0412] Comprehensive health risks

[0413] Health risks of industrial solvent B:

[0414] Exposure dose mg / kg / day

[0415] Cancer risk

[0416] Non-cancer risk

[0417] Comprehensive health risks

[0418] Health Risks of Fuel Additive C:

[0419] Exposure dose mg / kg / day

[0420] Cancer risk

[0421] Non-cancer risk

[0422] Comprehensive health risks

[0423] Taking these three pollutants into consideration, the overall health risk of the groundwater sample is:

[0424] Comprehensive health risks:

[0425] ;

[0426] Next, the predicted health risk vector and the deviation vector By inputting it into the health risk assessment model, the health risk deviation of the groundwater sample can be obtained:

[0427] Health risk bias

[0428] This shows that the health risk prediction results for the groundwater sampling point have a deviation range of about 20%. This small deviation value indicates that the prediction results are highly reliable.

[0429] Through similar calculations, environmental protection departments can conduct a comprehensive health risk assessment of groundwater samples from all monitoring points in the area.

[0430] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for determining health risk deviation of groundwater organic pollution, characterized in that: The following steps are involved: S10, obtaining multiple sets of groundwater organic pollution data whose accuracy has been verified and their corresponding health risk vectors and deviation vectors, wherein the organic pollution data include groundwater parameters, organic pollutant parameters, and climate precipitation parameters; S20, using a Bayesian network to establish a probability dependency between each groundwater parameter, organic pollutant parameter, climate precipitation parameter and the health risk vector, and selecting multiple parameters that affect the health risk according to a preset dependency threshold, recorded as an influencing factor group; S30. Based on the probability dependence of Bayesian network, a group of groundwater organic pollution health risk equations considering the influencing factor group is established, including: exposure dose equation, carcinogenic risk equation, non-carcinogenic risk equation, comprehensive health risk equation, and health risk deviation equation; S40, fitting the groundwater organic pollution health risk equation group using the multiple groups of groundwater organic pollution health risk data to obtain a fitted health risk equation group; S50, using the Monte Carlo simulation method, randomly sampling the parameters in the fitted health risk equation group to generate a large amount of simulation data; S60, using the large amount of simulation data as a training set and the multiple groups of groundwater organic pollution health risk data as a validation set, training a neural network with a multi-branch structure to obtain a groundwater organic pollution health risk prediction model, which is used to calculate the health risk vector and deviation vector of the groundwater to be tested based on the organic pollution data of the groundwater to be tested; S70, using a pre-trained health risk assessment model, inputting the health risk vector and deviation vector of the groundwater to be tested, and obtaining the health risk deviation of the groundwater to be tested, which is used to describe the reliability of the health risk vector; Wherein, the exposure dose equation is specifically: ； In the formula, is the exposure dose (mg / kg / day); is the pollutant concentration (mg / L); is the intake rate (L / day); is the exposure frequency (days / year); is the duration of exposure (years); is body weight (kg); is the average time (days); is the partial derivative of the octanol-water partition coefficient with respect to temperature; is the temperature change (°C); is an imaginary unit; is the phase angle (rad); is the water solubility at the jth sampling point (mg / L); is the pH value of the jth sampling point; is the dissolved oxygen at the jth sampling point (mg / L); is the number of sampling points; The carcinogenic risk equation is specifically: ； In the formula, For cancer risk; is the exposure dose (mg / kg / day); is the slope factor (unit ); is the wavelength (m); is the angular frequency (rad / s); is time (s); is the concentration or intensity of the kth environmental factor; is the reference value of environmental factors; is the weight coefficient of the kth environmental factor; is the number of environmental factors; is the steepness coefficient of the logistic function; is the average annual precipitation (mm); is the precipitation threshold (mm); The non-carcinogenic risk equation is specifically: ； In the formula, is the hazard quotient (non-carcinogenic risk); is the exposure dose (mg / kg / day); is the reference dose (mg / kg / day); is an imaginary unit; is the molecular weight of the pollutant (g / mol); is the reference molecular weight (g / mol); is the intensity value of the lth precipitation intensity level (mm / h); is the frequency of the lth precipitation intensity level; is the evaporation of the lth precipitation intensity level (mm); is the number of precipitation intensity levels; is the redox potential (mV); is the reference redox potential (mV); is the redox potential normalization factor (mV); The comprehensive health risk equation is specifically: ； In the formula, For comprehensive health risks; For cancer risk; is the hazard quotient (non-carcinogenic risk); is the partial derivative of the distribution coefficient with respect to conductivity; is the conductivity change (μS / cm); is the dissolved oxygen concentration (mg / L); is the reference dissolved oxygen concentration (mg / L); is the dissolved oxygen impact index; is the exposure time (years); is the attenuation coefficient ( ); The health risk deviation equation is specifically: ； In the formula, for health risk bias; For the predicted overall health risk; is the observed overall health risk; is the partial derivative of water solubility with respect to temperature; is the temperature change (°C); is the octanol-water partition coefficient; is the reference octanol-water partition coefficient; is the steepness coefficient of the logistic function; is the weight of the h-th precipitation feature; is the observed value of the h-th precipitation feature; is the average value of precipitation characteristics; is the standard deviation of precipitation characteristics; is the number of precipitation features.

2. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The multi-branch neural network includes a groundwater parameter subnetwork, an organic pollutant parameter subnetwork, a climate precipitation parameter subnetwork and a fusion subnetwork.

3. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The groundwater parameters include pH value, dissolved oxygen, conductivity, and redox potential.

4. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The organic pollutant parameters include concentration, molecular weight, octanol-water partition coefficient, and water solubility.

5. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The climate precipitation parameters include average annual precipitation, precipitation intensity, precipitation frequency, and evaporation.

6. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The health risk vector is a two-dimensional vector including a carcinogenic risk rate, a non-carcinogenic risk rate, and an exposure dose.

7. A method for determining health risk deviation of groundwater organic pollution according to claim 6, characterized in that: The dependency threshold is preset to 0.

6.

8. A method for determining health risk deviation of groundwater organic pollution according to claim 1, characterized in that: The method for determining the types and concentrations of organic pollutants in groundwater is to compare surface enhanced Raman spectroscopy images with standard images.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, which, when executed, are used to execute the method for determining health risk deviation of groundwater organic pollution according to any one of claims 1 to 8.

10. A system for determining health risk deviation of groundwater organic pollution, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images