Water environment pollution tracing system and method

Abstract

The application discloses a water environment pollution tracing system and method, and the system comprises sequentially connected multilevel monitoring modules, a flow field reconstruction module, an inversion solving module, an intelligent identification module and a decision-making module; the multilevel monitoring module collects time series of pollutant concentration and three-dimensional flow field parameters in real time, and performs spatial interpolation and time filtering processing on the flow field data; the flow field reconstruction module establishes a dynamic source-receptor response matrix according to a pollutant transmission trajectory network; the inversion solving module calculates a parameter posterior probability distribution; the intelligent identification module identifies a pollution source type through a pre-trained classification model; and the decision-making module comprehensively determines an effective pollution source according to posterior probability confidence and spatiotemporal evolution characteristics. The application fuses flow field reconstruction and probability inference technologies, realizes high-precision identification and parameter estimation of a water body dynamic pollution source, and solves problems that a mobile pollution source and intermittent emission cannot be effectively processed in the prior art.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, specifically to a water pollution source tracing system and method. Background Technology

[0002] In water pollution supervision, accurately identifying the location, intensity, and type of pollution sources is a key prerequisite for achieving precise governance. Existing pollution source tracing technologies mainly rely on two types of methods: the first type is the water quality fingerprint-based source tracing method, which identifies the source by comparing the characteristics of pollutant components, but this method has poor applicability to mixed pollution and dynamic release situations; the second type is the numerical model-based source tracing method, which reverse-engineers by establishing pollutant transport and diffusion models, but most existing methods assume that the pollution source is a static point source and cannot effectively handle dynamic pollution situations such as mobile pollution sources and intermittent emissions.

[0003] Pollution in real-world aquatic environments often exhibits significant spatiotemporal dynamics, such as mobile pollution caused by ship navigation, intermittent illegal discharges from factories, and pipeline leaks. Identifying these dynamic pollution sources often lacks accurate three-dimensional flow field information, leading to transmission paths and large calculation errors. Furthermore, traditional inversion algorithms make overly simplistic assumptions and fail to capture the dynamic release characteristics of pollution sources, resulting in high parameter space dimensions and difficulties in optimization solutions under multi-source pollution conditions.

[0004] Therefore, there is an urgent need for a system and method that can achieve high-precision location and type identification of dynamic pollution sources. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a water environment pollution source tracing system and method. By integrating flow field reconstruction and Bayesian inversion technology, it can achieve high-precision identification and parameter estimation of dynamic pollution sources in water bodies.

[0006] The specific technical solution is as follows:

[0007] In a first aspect, the present invention provides a water environment pollution source tracing system, the system comprising, in sequence, a multi-level monitoring module, a flow field reconstruction module, an inversion solution module, an intelligent identification module, and a decision-making module.

[0008] The multi-level monitoring module uses a sensor network consisting of N water quality monitoring nodes and M flow velocity and direction monitoring nodes deployed in the target water body to collect pollutant concentration time series and three-dimensional flow field parameter time series in real time, and performs spatial interpolation and temporal filtering on the flow field data.

[0009] The flow field reconstruction module reconstructs the pollutant transport trajectory network in the water body based on the processed three-dimensional flow field data and uses Lagrange particle tracking technology. It also establishes a dynamic source-receptor response matrix based on the transport trajectory network and the location relationship of monitoring points.

[0010] The inversion solving module includes a pollution source parameter estimation model, which solves the position coordinates, release intensity and release time of the pollution source according to a probability inference method, and calculates the posterior probability distribution of the parameters.

[0011] The intelligent identification module is used for extracting the space-time distribution characteristics and concentration evolution characteristics of the pollution source, and identifying the type of the pollution source through a pre-trained classification model.

[0012] The decision determination module is used for comprehensively determining the effective pollution source according to the posterior probability distribution confidence and the space-time evolution characteristics.

[0013] In a second aspect, the present application provides a water environment pollution tracing method based on the system of the first aspect, comprising the following steps:

[0014] Step S1, deploying a multi-level monitoring sensor network to collect pollutant concentration time series and three-dimensional flow field parameter time series in real time; performing data preprocessing on the three-dimensional flow field parameters to obtain continuous flow field data.

[0015] Step S2, reconstructing a pollutant transport trajectory network in the water body based on the continuous flow field data; establishing a dynamic source-receptor response matrix according to the transport trajectory network, and introducing a decay correction to obtain a corrected response matrix.

[0016] Step S3, establishing a Bayesian estimation model of the pollution source parameters, and solving the posterior probability distribution of the pollution source position, intensity and release time by using a probability inference algorithm.

[0017] Step S4, extracting the space-time distribution characteristics and concentration evolution characteristics of the pollution source, and inputting a pre-trained classification model to obtain the type of the pollution source; when the posterior probability confidence exceeds a preset threshold and the space-time characteristics meet the dynamic release mode, the pollution source is determined as an effective pollution source.

[0018] Further, when the sensor nodes in the multi-level monitoring module are deployed, the characteristic diffusion distance of the pollutant is calculated according to the hydrological and geographical parameters of the target water body, and the optimal spatial density of the monitoring nodes is determined ; the deployment positions of N water quality monitoring nodes are determined by using a spatial optimization algorithm according to the optimal density and the water body boundary constraint, and M flow velocity and direction monitoring nodes are deployed along the main axis direction of the water flow.

[0019] The deployment scheme is subjected to coverage blind area detection, and if there is a monitoring blind area with an area greater than 20% of the total area, the node positions are iteratively optimized until the coverage requirement is met.

[0020] Further, the trajectory network reconstruction method comprises: generating uniformly distributed virtual tracer particles in the space of the target water body, and performing forward and backward trajectory integration on each particle according to the continuous three-dimensional flow field data.

[0021] The time step of trajectory integration is adaptively adjusted according to the flow field, and the calculation formula is:

[0022] ; wherein, is the basic time step, is the Courant number, is the minimum grid size, is the local maximum flow velocity; a directed trajectory graph is constructed according to the particle trajectory sequence, the connectivity and path weight of the graph are analyzed, and a pollutant transport trajectory network is obtained.

[0023] Further, the method for establishing the dynamic source-receptor response matrix comprises: identifying candidate source locations in the trajectory network, and preferentially selecting node locations with high connectivity and high centrality; calculating the optimal transmission path and transmission time from each candidate source location to each monitoring point, and calculating the path transmission efficiency coefficient according to the flow velocity distribution on the path .

[0024] Establishing a response matrix of candidate sources and monitoring points , the matrix element The calculation formula is:

[0025] ; wherein, is the pollutant decay coefficient, is the transmission time from the source to the receptor , and is the effective diffusion coefficient.

[0026] Further, the method for establishing the Bayesian estimation model comprises: defining a pollutant source parameter vector , wherein, is the coordinate of the source location, is the release intensity, and is the release start time; establishing a likelihood function of observation data, assuming that the observation error obeys a Gaussian distribution:

[0027] ; wherein, is the observation concentration of the monitoring point at time , and is the predicted concentration under the given parameters; setting the prior probability distribution of the parameters, and then using a probability inference algorithm to obtain the posterior probability distribution of the parameters.

[0028] Further, the probability inference algorithm uses a variational inference method to establish a variational approximation distribution family , using a parameterized distribution form, wherein is a variational parameter.

[0029] The evidence lower bound ELBO is defined as the optimization objective function: ; wherein, represents the expectation under the distribution , represents the divergence, is the prior distribution; the gradient optimization algorithm is adopted to maximize the ELBO to obtain the optimal approximate posterior distribution.

[0030] Further, the uncertainty ellipse of the pollution source position is calculated according to the posterior probability distribution, and the ratio of the long axis to the short axis is taken as the position dispersion degree feature ; the variation coefficient of the release intensity is calculated as the intensity stability feature ; the time concentration degree feature is calculated according to the release time posterior distribution ; the Pearson correlation coefficient of the predicted concentration and the observed concentration is calculated as the goodness-of-fit feature ; the above features are combined into a comprehensive feature vector for pollution source type identification.

[0031] Further, the construction method of the pre-trained classification model comprises: constructing a training data set containing typical pollution source types of point source, line source, surface source and mobile source; a deep neural network is used to establish a classifier, including an input layer, multiple hidden layers and an output layer; a cross-entropy loss function and an adaptive optimizer are used to train the classifier, and a regularization technique is used to prevent overfitting; the comprehensive feature vector is input, and the probability distribution of each pollution source type is output.

[0032] Further, the decision condition of the decision module comprises: the posterior probability distribution confidence is measured by calculating the ratio of the 95% confidence interval width to the mean value of the parameter estimate, and when the ratio is less than a preset threshold 0.3, the parameter estimate is considered reliable; the pollution source space-time evolution feature conforms to the dynamic release mode, including: the position dispersion degree feature , the intensity stability feature , and the time feature ; when the above conditions are met at the same time, it is determined as an effective pollution source.

[0033] Compared with the prior art, the present application has the following advantages:

[0034] The present application acquires accurate flow field information through a multi-level sensor network, reconstructs a refined transmission trajectory by using Lagrangian particle tracking, realizes parameter uncertainty quantification based on Bayesian theory, and realizes intelligent identification of pollution source types by combining deep learning; high-precision identification and parameter estimation of dynamic pollution sources in water bodies are realized, and the present application has a broad application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a water environment pollution source tracing system composition schematic diagram of the present application;

[0036] Figure 2 A water environment pollution tracing method flow chart of the present application. DETAILED DESCRIPTION

[0037] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application are described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0038] Embodiment 1

[0039] As shown in Fig. 1, it is a water environment pollution tracing system composition schematic diagram of the present application, which comprises the following modules connected in sequence: a multi-level monitoring module, a flow field reconstruction module, an inversion solving module, an intelligent identification module and a decision-making module. Figure 1

[0040] The multi-level monitoring module is composed of a sensor network of N water quality monitoring nodes and M flow velocity and flow direction monitoring nodes deployed in the target water body, and is used for collecting pollutant concentration time series and three-dimensional flow field parameter time series in real time, and performing spatial interpolation and time filtering processing on the flow field data.

[0041] The flow field reconstruction module is used for reconstructing a pollutant transmission trajectory network in the water body based on the processed three-dimensional flow field data by using a Lagrangian particle tracking technology, and establishing a dynamic source-receptor response matrix according to the transmission trajectory network and the position relationship of the monitoring nodes.

[0042] The inversion solving module comprises a pollution source parameter estimation model, and is used for solving the position coordinates, release intensity and release time of the pollution source according to a probability inference method, and calculating the posterior probability distribution of the parameters.

[0043] The intelligent identification module is used for extracting the spatio-temporal distribution characteristics and concentration evolution characteristics of the pollution source, and identifying the type of the pollution source by using a pre-trained classification model.

[0044] The decision-making module is used for comprehensively determining the effective pollution source according to the confidence of the posterior probability distribution and the spatio-temporal evolution characteristics.

[0045] Embodiment 2

[0046] As shown in Fig. 2, it is a water environment pollution tracing method flow chart of the present application, which comprises the following steps:

[0047] ​Step S1, deploy a multi-level monitoring sensor network, collect pollutant concentration time series and three-dimensional flow field parameter time series in real time; perform data preprocessing on the three-dimensional flow field parameters to obtain continuous flow field data.

[0048] When the sensor nodes in the multi-level monitoring module are deployed, the characteristic diffusion distance of the pollutant is calculated according to the hydrological and geographical parameters of the target water body, and the optimal spatial density of the monitoring nodes is determined ; the spatial optimization algorithm is used to determine the deployment positions of N water quality monitoring nodes according to the optimal density and the boundary constraints of the water body, and M flow velocity and direction monitoring nodes are deployed along the main axis direction of the water flow.

[0049] The coverage blind area detection is performed on the deployment scheme, and if the monitoring blind area area is greater than 20% of the total area, the node position is iteratively optimized until the coverage requirement is met.

[0050] Step S2, based on the continuous flow field data, reconstruct the pollutant transport trajectory network in the water body; establish a dynamic source-receptor response matrix according to the transport trajectory network, and introduce a decay correction to obtain a corrected response matrix.

[0051] Uniformly distributed virtual tracer particles are generated in the space of the target water body, and forward and backward trajectory integrations are performed on each particle according to the continuous three-dimensional flow field data;

[0052] The time step of the trajectory integration is adaptively adjusted according to the flow field change, and the calculation formula is:

[0053] ; wherein, is the basic time step, is the Courant number, is the minimum grid size, is the local maximum flow velocity;

[0054] According to the particle trajectory sequence, a directed trajectory graph is constructed, the connectivity and path weight of the graph are analyzed, and a pollutant transport trajectory network is obtained.

[0055] The method for establishing the dynamic source-receptor response matrix comprises:

[0056] Identify the candidate positions of the pollution sources in the trajectory network, and preferentially select the node positions with high connectivity and high centrality; calculate the optimal transmission path and transmission time from each candidate source position to each monitoring point, and calculate the path transmission efficiency coefficient according to the flow velocity distribution on the path ;

[0057] Establish the response matrix of the candidate source and the monitoring point , the matrix element The calculation formula is:

[0058] ; wherein is the pollutant decay coefficient, is the transmission time from source to receptor , is the effective diffusion coefficient.

[0059] Step S3, a Bayesian estimation model of the pollution source parameters is established, and a probability inference algorithm is used to solve the posterior probability distribution of the pollution source position, intensity and release time.

[0060] The method for establishing the Bayesian estimation model comprises:

[0061] A pollution source parameter vector is defined, wherein is the pollution source position coordinate, is the release intensity, is the release start time;

[0062] A likelihood function of the observation data is established, and it is assumed that the observation error obeys a Gaussian distribution:

[0063] ; wherein is the observation concentration of the monitoring point at time , is the predicted concentration under the given parameters; by setting the prior probability distribution of the parameters, the posterior probability distribution of the parameters can be obtained by using the probability inference algorithm.

[0064] The probability inference algorithm uses a variational inference method to establish a variational approximation distribution family , and uses a parameterized distribution form, wherein is a variational parameter; the evidence lower bound ELBO is defined as an optimization objective function:

[0065] ; wherein represents an expectation under the distribution , represents a divergence, is the prior distribution; a gradient optimization algorithm is used to maximize the ELBO to obtain an optimal approximate posterior distribution.

[0066] Step S4, the spatiotemporal distribution characteristics and concentration evolution characteristics of the pollution source are extracted, and a pre-trained classification model is input to obtain the pollution source type; when the posterior probability confidence exceeds a preset threshold and the spatiotemporal characteristics meet the dynamic release mode, the pollution source is determined to be effective.

[0067] According to the posterior probability distribution, the uncertainty ellipse of the pollution source position is calculated, and the ratio of the long axis to the short axis is taken as the position dispersion degree feature ; the coefficient of variation of the release intensity is calculated as the intensity stability feature ; Calculate the temporal concentration characteristics based on the posterior distribution at the release time. The Pearson correlation coefficient between predicted and observed concentrations was calculated as a goodness-of-fit feature. The above features are combined into a comprehensive feature vector for pollution source type identification.

[0068] The method for constructing the pre-trained classification model includes: constructing a training dataset containing typical pollution source types such as point sources, line sources, area sources, and moving sources; using a deep neural network to build a classifier, including an input layer, multiple hidden layers, and an output layer; training the classifier using a cross-entropy loss function and an adaptive optimizer, and using regularization techniques to prevent overfitting; inputting a comprehensive feature vector and outputting the probability distribution of each pollution source type.

[0069] The decision-making criteria of the decision-making module include: the posterior probability distribution reliability is measured by the ratio of the 95% confidence interval width of the parameter estimate to the mean; when this ratio is less than a preset threshold of 0.3, the parameter estimate is considered reliable; the spatiotemporal evolution characteristics of the pollution source conform to a dynamic release pattern, including: location dispersion characteristics. Strength and stability characteristics Time characteristics A source is considered a valid pollution source if all of the above conditions are met.

[0070] Taking a sudden pollution incident in a city river as an example, the specific application process of the system and method of the present invention is explained in detail; the monitoring deployment area of ​​the river is about 15 kilometers long, with an average width of 120 meters, an average water depth of 3.5 meters, and the main flow direction is from west to east.

[0071] On a certain morning in August 2024 at 9:30, the No. 3 water quality monitoring station located downstream of the river detected an abnormal increase in the concentration of chemical oxygen demand (COD), which rapidly rose from the normal 15 mg / L to 31 mg / L. Subsequently, the No. 4 and No. 5 monitoring stations also detected an increase in pollutant concentration.

[0072] A total of N=12 water quality monitoring nodes and M=8 flow velocity and direction monitoring nodes were deployed along this river section. The water quality monitoring nodes are equipped with multi-parameter water quality sensors, capable of real-time monitoring of indicators such as COD, ammonia nitrogen, and total phosphorus, with a sampling frequency of 5 minutes per node. The flow velocity and direction monitoring nodes employ acoustic Doppler current meters (ADCP), deployed every 2 kilometers along the main river axis, to collect real-time three-dimensional flow field data, including flow velocity (accuracy ±0.01 m / s), flow direction (accuracy ±2°), and water depth information.

[0073] Based on the river's hydrological parameters (average flow velocity 0.3 m / s, longitudinal dispersion coefficient 150 m² / s, lateral dispersion coefficient 8 m² / s), the system calculates the characteristic diffusion distance of pollutants to be approximately 2.8 km, determining the optimal spatial density of monitoring nodes ρo = 0.8 nodes / km². Using a spatial optimization algorithm, under the constraint of a 20% coverage blind zone, the system optimizes the deployment coordinates of 12 water quality monitoring nodes. The system performs cubic spline spatial interpolation and 5-point moving average time filtering on the collected flow field data to obtain spatiotemporally continuous flow field data. 20,000 virtual tracer particles are generated within the river space, uniformly distributed in a three-dimensional space of 15 km long × 120 m wide × 3.5 m deep. Lagrange trajectory integration is performed on each particle based on real-time flow field data. In this case, the base time step is... Set to 60 seconds, Courant number CCFL to 0.5, minimum grid size. It is 10 meters.

[0074] During the period of the pollution incident (9:00-12:00), the local maximum flow velocity Reaching 0.45 m / s, the adaptive time step is adjusted to... ≈11 seconds.

[0075] Through particle trajectory analysis, the system constructed a directed trajectory network containing 538 nodes and 1247 edges, identifying 23 highly connected candidate pollution source locations. For each candidate source, the optimal transmission path to 12 monitoring points was calculated. Taking monitoring station No. 3 as an example, the optimal path length from candidate source S7, located 5.2 kilometers upstream, to the monitoring station is 5350 meters, and the transmission time is... Minutes, path transmission efficiency coefficient Given a pollutant attenuation coefficient λ = 0.15 / day and an effective diffusion coefficient Deff = 45 m² / s, the response matrix element A7,3(t) = 0.0234 is calculated.

[0076] Define pollution source parameter vector Where (x, y, z) are three-dimensional coordinates (unit: meters), and Q is the release intensity (unit: kg / h). The release start time is given in minutes. A likelihood function was established based on concentration observation data from 36 time points at 12 monitoring stations between 9:00 and 12:00.

[0077] Given a prior distribution of parameters and a uniform distribution of location coordinates within the river space, a 5-dimensional normal distribution family is established as a variational approximation distribution. , where variational parameters The total number of parameters is 20, including the mean vector and the covariance matrix. The Adam optimizer is used to maximize the ELBO objective function with a learning rate of 0.01. After 500 iterations, the ELBO value converges to -2847.3; the posterior probability distribution of the pollution source is obtained by inversion: position meters (95% confidence interval), meters, meters; release intensity ; release time minutes; the long axis of the uncertainty ellipse of the posterior distribution is 360 meters, the short axis is 50 meters, the ratio of the long axis to the short axis is 7.2, and the position dispersion characteristic is calculated. The coefficient of variation of the release intensity is 0.18, and the intensity stability characteristic is obtained. The concentration of the posterior distribution of the release time . The Pearson correlation coefficient between the predicted concentration and the observed concentration .

[0078] The comprehensive feature vector is input into the pre-trained deep neural network classifier. The classifier includes an input layer (4 neurons), 3 hidden layers (32, 16, and 8 neurons respectively, with ReLU activation function), and an output layer (4 neurons corresponding to point source, line source, surface source, and mobile source); the output results show that the probability of point source is 0.89, the probability of line source is 0.06, the probability of surface source is 0.03, and the probability of mobile source is 0.02, determining the type of pollution source as point source. The decision-making module calculates the ratio of the 95% confidence interval width to the mean value of the parameter estimates, the position is , and the intensity Q is .

[0079] Since the ratio of the intensity is 0.36, which exceeds the threshold value of 0.3, the position dispersion , the intensity stability , and the time concentration , the spatiotemporal evolution characteristics meet the determination criteria of the dynamic release mode; after comprehensive evaluation, the system determines this pollution source as an effective pollution source.

[0080] The final tracing result points to a chemical enterprise sewage outlet located 6.85 kilometers upstream of the river, 45 meters from the left bank. On-site verification found that the enterprise directly discharged high-concentration wastewater due to equipment failure at around 8:45, lasting about 40 minutes, with an average discharge intensity of about 150 kg / h, which is highly consistent with the system inversion results.

[0081] The above detailed description of the specific embodiments of the present application has been given to understand the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A water pollution source tracing system, characterized in that, The system comprises, in sequence, a multi-level monitoring module, a flow field reconstruction module, an inversion solution module, an intelligent identification module, and a decision-making module; The multi-level monitoring module uses a sensor network consisting of N water quality monitoring nodes and M flow velocity and direction monitoring nodes deployed in the target water body to collect pollutant concentration time series and three-dimensional flow field parameter time series in real time, and performs spatial interpolation and temporal filtering on the flow field data. The flow field reconstruction module reconstructs the pollutant transport trajectory network in the water body based on the processed three-dimensional flow field data and uses Lagrange particle tracking technology. It also establishes a dynamic source-receptor response matrix based on the transport trajectory network and the location relationship of monitoring points. The method for establishing the dynamic source-receptor response matrix includes: Identify candidate pollution source locations in the trajectory network, calculate the optimal transport path and transport time from each candidate source location to each monitoring point, and calculate the path transport efficiency coefficient based on the velocity distribution along the path. ; Establish response matrix of candidate sources and monitoring points Matrix elements The calculation formula is: ;in, The pollutant attenuation coefficient, To start from the source to receptor Transmission time, The effective diffusion coefficient; The inversion solution module includes a pollution source parameter estimation model, which solves for the location coordinates, release intensity and release time of the pollution source according to the probabilistic inference method, and calculates the posterior probability distribution of the parameters; The intelligent identification module is used to extract the spatiotemporal distribution characteristics and concentration evolution characteristics of pollution sources, and to identify the type of pollution source through a pre-trained classification model; The uncertainty ellipse of the pollution source location is calculated based on the posterior probability distribution, and the ratio of its major and minor axes is used as the feature of location dispersion. ; Calculate the coefficient of variation of release intensity as a characteristic of intensity stability. ; Calculate the temporal concentration characteristics based on the posterior distribution at the release time. ; The Pearson correlation coefficient between predicted and observed concentrations was calculated as a goodness-of-fit feature. ; The above features are combined into a comprehensive feature vector for pollution source type identification; The decision-making module is used to comprehensively determine the effective pollution source based on the posterior probability distribution confidence, spatiotemporal distribution characteristics, and concentration evolution characteristics.

2. A method for tracing the source of water pollution, implemented based on the water pollution tracing system described in claim 1, characterized in that, The method includes the following steps: Step S1: Deploy a multi-level monitoring sensor network to collect pollutant concentration time series and three-dimensional flow field parameter time series in real time; perform data preprocessing on the three-dimensional flow field parameters to obtain continuous flow field data; Step S2: Based on continuous flow field data, reconstruct the pollutant transport trajectory network in the water body; establish a dynamic source-receptor response matrix according to the transport trajectory network, and introduce attenuation correction to obtain a corrected response matrix; Step S3: Establish a Bayesian estimation model for pollution source parameters, and use a probabilistic inference algorithm to solve the posterior probability distribution of pollution source location, intensity, and release time. Step S4: Extract the spatiotemporal distribution features and concentration evolution features of the pollution source, input them into the pre-trained classification model to obtain the pollution source type; when the posterior probability confidence exceeds the preset threshold and the spatiotemporal features conform to the dynamic release pattern, it is determined to be a valid pollution source.

3. The water pollution source tracing method according to claim 2, characterized in that, When deploying sensor nodes in a multi-level monitoring module, the characteristic diffusion distance of pollutants is calculated based on the hydrological and geographical parameters of the target water body to determine the optimal spatial density of the monitoring nodes. The spatial optimization algorithm is used to determine the deployment locations of N water quality monitoring nodes based on the optimal density and water body boundary constraints, and M flow velocity and direction monitoring nodes are deployed along the main axis of water flow. The deployment plan is tested for coverage blind spots. If the area of ​​a monitoring blind spot is greater than 20% of the total area, the node positions are iteratively optimized until the coverage requirements are met.

4. The water pollution source tracing method according to claim 3, characterized in that, Trajectory network reconstruction methods include: A uniformly distributed virtual tracer particle is generated within the target water body space, and the forward and backward trajectories of each particle are integrated based on continuous three-dimensional flow field data. The time step of the trajectory integral is adaptively adjusted according to the changes in the flow field, and the calculation formula is as follows: ;in, Based on the time step, For Courant number, For the smallest grid scale, The local maximum flow rate; A directed trajectory graph is constructed based on the particle trajectory sequence. The connectivity and path weights of the graph are analyzed to obtain the pollutant transport trajectory network.

5. The water pollution source tracing method according to claim 4, characterized in that, Methods for establishing Bayesian estimation models include: Define pollution source parameter vector ,in, The coordinates of the pollution source location To release strength, The start time of release; Establish the likelihood function for the observed data, assuming that the observation error follows a Gaussian distribution: ;in, For monitoring points At any moment The observed concentration, Given the predicted concentration with specific parameters, and by setting the prior probability distribution of the parameters, the posterior probability distribution of the parameters can be obtained using a probabilistic inference algorithm.

6. The water pollution source tracing method according to claim 5, characterized in that, The probabilistic inference algorithm employs a variational inference method to establish a variational approximate distribution family. It adopts a parameterized distribution form, in which For variational parameters; Define the evidence lower bound ELBO as the optimization objective function: ;in, Indicates distribution The expectations below Denotes divergence, The prior distribution is used; the optimal approximate posterior distribution is obtained by maximizing ELBO using a gradient optimization algorithm.

7. The water pollution source tracing method according to claim 6, characterized in that, The method for constructing the pre-trained classification model includes: constructing a training dataset containing typical pollution source types such as point sources, line sources, area sources, and moving sources; and using a deep neural network to build a classifier, including an input layer, multiple hidden layers, and an output layer. The classifier is trained using the cross-entropy loss function and an adaptive optimizer, and regularization techniques are employed to prevent overfitting. Input a comprehensive feature vector and output the probability distribution of each pollution source type.

8. The water pollution source tracing method according to claim 7, characterized in that, In step S4, the criteria for determining a valid pollution source include: The reliability of the posterior probability distribution is measured by the ratio of the width of the 95% confidence interval of the parameter estimate to the mean. When this ratio is less than the preset threshold of 0.3, the parameter estimate is considered reliable. The spatiotemporal evolution characteristics of pollution sources conform to a dynamic release pattern, including: location dispersion characteristics. Strength and stability characteristics Time characteristics ; A source is considered a valid pollution source if all of the above conditions are met.

Images