Water quality pollution tracing method based on Bayesian reasoning and MCMC sampling

Through Bayesian inference and MCMC sampling methods, a convection diffusion model of water quality pollutants was designed, and the functional distribution of pollutant discharge source parameters was established, which solved the problem of uncertainty in the current technology of pollutant traceability results, and achieved efficient and accurate pollution sources. Trace the origin.

CN118917857APending Publication Date: 2024-11-08BEIJING UNIV OF TECH +1

Patent Information

Application Number
CN202410897864.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art is difficult to achieve accurate traceability of water pollutants, especially under limited monitoring data conditions, the location of the pollution source cannot be effectively inverted, resulting in high inverse uncertainty in the reverse traceability results, which is complex and time-consuming.

Method used

The water pollution traceability method based on Bayesian reasoning and Markov chain Monte Carlo algorithm (MCMC) sampling is adopted. By designing the convection diffusion model of water quality pollutants, the functional distribution of pollutant discharge source parameters is established, and the MCMC probability distribution sampling method is used. Get a numerical summary of the parameters to be inverted.

Benefits of technology

It effectively solves the problem of high-dimensional complex probability distribution sampling in pollution source traceability, realizes relatively accurate pollution source traceability, reduces calculation complexity and time-consuming, and improves traceability accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917857A_ABST
    Figure CN118917857A_ABST
Patent Text Reader

Abstract

The invention relates to a water quality pollution source tracing method, in particular to a water body pollutant source tracing method based on a mechanism model, Bayesian reasoning and a Markov Chain Monte Carlo (MCMC) algorithm (Markov Chain Monte Carlo, MCMC). Firstly, a convective diffusion model based on water pollutants is designed; secondly, combining the convective diffusion model of the pollutants with a Bayesian reasoning framework to obtain posterior distribution of to-be-inverted parameters of the pollution discharge source based on experimental data; and finally, aiming at the problems of high posterior distribution probability function parameter dimension, poor continuity and differentiability and extremely complex calculation in Bayesian reasoning, realizing simplification of high-dimensional complex probability distribution sampling by using an MCMC probability distribution sampling method. According to the method, the value range of the pollution discharge parameters can be greatly narrowed in the absence of priori knowledge, and finally a parameter summary of the pollution discharge parameters is given.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for tracing the source of water pollution, and in particular to a method for tracing the source of water pollution based on Bayesian reasoning and MCMC sampling. Background Art

[0002] With the rapid progress of society, the people's demand for ecological and environmental safety is constantly increasing, and the situation of improving water environment quality is grim. It is difficult to completely solve the structural pollution in my country in the short term, and the industrial structure and layout still need to be optimized. The black and smelly problem of rivers and ditches in some cities is prominent, which has attracted widespread attention and strong response from the public. The high-efficiency utilization of water resources continues to exist, the space of water ecosystems is significantly compressed, the ecological water demand is scarce, and the eutrophication problem of local lakes is significant. These factors have led to the long-term occupation of ecological water use by domestic and industrial water. At the same time, the pollution emissions in the basin continue to exceed the self-purification capacity of the water body, and the pollution problems of non-traditional pollutants such as ammonia nitrogen, total phosphorus and total nitrogen have become more serious. The complex pollution problem of the water environment urgently needs to be solved. Despite this, the complexity, urgency and sustainability of regional water pollution control have not changed fundamentally, and the task of water pollution prevention and control is still very arduous. Therefore, in order to further improve the ecological environment scientifically and effectively and improve the quality of water ecological environment, a new generation of information technology is used to analyze prominent ecological and environmental problems, so that decision-making, management and governance in the field of ecological environment are oriented to intelligence.

[0003] In recent years, the problem of inversion of river pollution source terms has begun to attract attention. In actual situations, it is very difficult to accurately trace the source of pollutants. Traditional identification models require a large amount of prior information, supported by the leakage history of pollutants, chemical products derived from pollutants, and accurate accident environmental monitoring data. However, in water environment monitoring, it is extremely unrealistic to achieve real-time monitoring of the spatial location of all types of pollutants with full coverage. How to use limited monitoring data to invert key information such as the location of pollution sources requires a set of effective methods for tracing pollution sources. Chinese scholars have done a lot of work and proposed various solutions. For example, analytical methods, optimization methods based on genetic algorithms, differential evolution algorithms, geostatistical methods, reverse position probability density function methods, adjoint equation methods, etc., are all based on the same pollutant transport, dilution and transformation mechanism to obtain the unique value of the source term parameters. The reverse uncertainty of the inversion result needs to be expressed by multiple calculations such as perturbation methods. The direct method solves the tracing problem by timely inverting the control equation and reconstructing the historical distribution of observed pollutants. The probability statistics method estimates the probability of an event and continuously improves the identification value based on the likelihood function to identify the parameters of the pollution source. The main methods include Bayesian reasoning, minimum relative entropy, etc. After 2015, some people began to focus on the analysis of inversion factors from a practical perspective, such as the analysis of the impact of the inversion accuracy of instantaneous emission source terms under the generalization of wide and shallow rivers and two-dimensional rivers. Summary of the invention

[0004] In view of the deficiencies of the above prior art, the present invention provides a method for tracing the source of water pollutants based on a mechanism model, Bayesian reasoning and Markov Chain Monte Carlo (MCMC) probability distribution sampling method. It includes: a design scheme based on a convection diffusion model of water pollutants; a function distribution scheme for establishing pollution source parameters based on Bayesian reasoning; and a numerical total of the parameters to be inverted indirectly based on the MCMC probability distribution sampling method. The purpose of the present invention is achieved through the following technical solutions.

[0005] A water pollution source tracing method based on Bayesian reasoning and MCMC sampling, the method comprising the following steps:

[0006] 1) Obtain tracer experimental data from the river monitored over a period of time;

[0007] 2) Processing the data to obtain sampling concentration data C, and studying the relationship between the sampling concentration data C and the pollution source emission location parameter s, the pollution source emission mass m, and the pollutant emission time t;

[0008] 3) Based on 2), design a convection-diffusion model based on water pollutants;

[0009] 4) Substitute the convection-diffusion model obtained in 3) into the Bayesian framework, establish a likelihood function under the assumption that the error is uncorrelated with the mean square error, and obtain the posterior distribution of the parameters to be inverted of the pollution source based on the experimental data through formula calculation;

[0010] 5) Perform MCMC sampling on the complex posterior distribution outputted in 4) to obtain a numerical summary of the parameters to be inverted. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A schematic diagram of the composition of a water pollution source tracing method based on Bayesian reasoning and MCMC sampling;

[0012] Figure 2 Flowchart of Bayesian inference and MCMC sampling. DETAILED DESCRIPTION

[0013] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. The following description covers many specific details in order to provide a comprehensive understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a clearer understanding of the present invention by illustrating examples of the present invention. The present invention is by no means limited to any specific configuration and algorithm proposed below, but covers any modification, replacement and improvement of related elements, parts and algorithms without departing from the spirit of the present invention.

[0014] The following will refer to the attached Figure 1 The specific steps of describing a method for tracing the source of water pollution based on Bayesian reasoning and MCMC sampling according to an embodiment of the present invention are as follows:

[0015] The first step is to obtain tracer experimental data monitored in the river over a period of time.

[0016] Rhodamine WT dye was selected as the tracer due to its inertness to adsorb to aquatic plants, suspended clay or contact with riverbed and riverbank. Compared with other dyes, it has lower attenuation during collection and will not be lost in large quantities during the process. The tracer was discharged at different locations in the river at different doses, and samples were recorded downstream at intervals ranging from 2 to 10 minutes. The sampling intervals were the same in each experiment, but varied between experiments.

[0017] The second step is to study the relationship between the sampling concentration data C and the pollution source emission location parameter s, the pollution source emission mass m, and the pollutant emission time t.

[0018] In the scenario of pollution source tracing, when pollutants exceeding the standard are detected in a certain section, the pollutant concentration change indicators are collected to obtain the sampling concentration data C at n times. The data is combined with the water environment pollutant diffusion mechanism model to obtain the following formula.

[0019] C = g(s, m, t|Θ) + e meas +e model

[0020] Among them, g is the function that maps the river mechanism model to the data environment, e model It represents the error between the mathematical model and the real physical environment, which is an n-dimensional vector. meas An n-dimensional vector representing random measurement errors. Among them, s is the emission location parameter of the pollution source, m is the emission mass of the pollution source, t is the pollutant emission time, Θ is the model parameter, and the goal of tracing the source is to monitor the sampling data C to estimate the pollution source emission information s, m and t.

[0021] The third step is to design a convection-diffusion model based on water pollutants.

[0022] To simplify the problem, a convection-diffusion model based on the inference model is constructed, and the following assumptions are temporarily made: the model parameters are set as constants, and the pollutants are completely mixed in the river in the horizontal and vertical directions. The analytical solution of the pollutant in the water quality model under different conditions can be obtained by methods such as characteristic lines. s , t s ) is discharged quickly into the mass M s The information of the pollution source can be obtained by the Delta function s = M s δ(xx s )δ(xx t ), the pollutant concentration formula at the detection zone (x, t) is as follows:

[0023]

[0024] Where A is the cross-sectional area of ​​the river, in m 2 , D x is the longitudinal average diffusion coefficient, i.e. the direction of river flow, in m 2 / min, U is the average flow velocity of the river, in m, K is the attenuation coefficient, in min -1 , t is the time after the detection starts, in min, and x is the distance of the monitoring point, in m. Since the location of the pollution source in most pollution source tracing scenarios is unknown, the coordinate origin of time and location is set at the start of the detection and the monitoring point.

[0025] The fourth step is to substitute the convection-diffusion model into the Bayesian framework, establish the likelihood function under the assumption that the error is uncorrelated with the mean square error, and obtain the posterior distribution of the parameters to be inverted of the pollution source based on the experimental data through formula calculation.

[0026] In the Bayesian statistical method, all unknown parameters are regarded as random variables that conform to a certain distribution, and their distribution can be determined by known information. Therefore, the Bayesian statistical method is an effective and reliable method in uncertainty analysis, and can provide key theoretical data support for management decisions. The inference formula based on Bayesian theory is as follows:

[0027]

[0028] Among them, p(X|Y) is the posterior probability distribution function of event X, which indicates the distribution law of X after the observed value Y is determined. p(X) is the prior probability distribution function, which indicates the distribution law of event X obtained from prior experience or data. p(Y|X) is the likelihood function, which represents the gap between the parameters in the model and the observed values, that is, the degree of fit. Substituting the pollutant convection diffusion model determined above into the Bayesian framework, the formula can be obtained:

[0029]

[0030] p(C|I)=∫ all s p(s)l(C|I,s)ds

[0031] Among them, p(s|C, I) is the posterior distribution of pollution source related parameters, p(s|l) is the prior distribution of pollution source parameters, l(C|I, s) is the likelihood function indicating the degree of fitting, p(C|I) is the evidence of pollution occurrence, C is the distribution of pollutant concentration change data obtained by monitoring, and I is environmental information, which is used to infer the background distribution of pollution source emission behavior parameters.

[0032] Determining the prior distribution of parameters is a crucial part in Bayesian theory. In the absence of a prior distribution, Bayesian reasoning cannot be performed. Prior distributions are mostly determined by objective experience. Especially in the field of hydrology, based on a large amount of historical water quality condition change monitoring data, common distributions include Gamma distribution, normal distribution and uniform distribution. Usually, under the constraint of limited knowledge, uniform distribution is usually selected as the prior distribution of parameters to express our attitude towards possible values. In real scenarios, large-scale risk pollution sources are often traceable, such as sewage treatment plants, basic industry factories or livestock farms, etc., and more appropriate prior functions can be set at that time. In the present invention, uniform distribution is still selected as the prior distribution of pollution source parameters. The prior probability density function of uniform distribution is as follows:

[0033] p(s|I) = constant, s∈R

[0034] The likelihood function gives the degree of fit between the modeling parameters and the observed values, plays an important role in the parameter estimation method, and indicates the probability of different parameter values ​​in the Bayesian statistical inference process. For different environments, due to different factors, the construction of the likelihood function is often different. A good likelihood function design can obtain a posterior probability distribution with a better fit. In the experimental environment of the present invention, the concentration detection data Ci of the i-th sampling point is compared with the actual data C true,i The difference is caused by the measurement error e meas and model construction error e model Here we assume that both errors follow a standard normal distribution and Using R i It represents the relationship between the pollution source emission diffusion distribution and the traceability model, that is, the g(s, m, t|Θ) part in the above formula, which is the same as the true value C true,i There is also an error between According to the law of error propagation, the error formula between the model prediction parameter value and the monitoring true value is as follows:

[0035]

[0036] Assuming that the two errors mentioned are independent of each other and have their own distribution, the error can be transformed into the formula:

[0037]

[0038] To simplify the calculation, we can further assume that the two errors are independent and identically distributed, that is, and In this way, the likelihood function can be derived as follows:

[0039]

[0040] Among them, there are If σ all Knowing that, we can obtain l(C|s, I) by calculating g(s;m).

[0041] The prior probability density function and likelihood function have been modeled above. Combining the above formula, the posterior probability density function of the pollution point parameters can be obtained. The formula is as follows:

[0042]

[0043] In the fifth step, MCMC sampling is performed on the output complex posterior distribution to obtain the numerical summary of the parameters to be inverted.

[0044] Even after layers of simplification, the posterior distribution of pollution source parameters is still a complex and uncommon distribution. Direct modeling and calculation will bring great trouble and huge computational overhead. This is why the Bayesian reasoning method has encountered great troubles and limitations when applied in practice for a period of time.

[0045] The MCMC algorithm is a probability distribution sampling algorithm based on constructing a Markov chain. The constructed Markov chain has an ideal stationary distribution, that is, its distribution does not change over time. The basic idea of ​​the algorithm is to use a suitable algorithm to sample from a probability distribution p that is complex or has no clear mathematical modeling expression, and obtain a sample sequence that obeys the p distribution and satisfies the properties of the Markov chain. An important property of the Markov chain is that the generation of each new state depends only on the current state and a simple probability transition function. Thanks to this, we do not need to directly analyze the complex probability distribution p, but only need to analyze the sampled sample sequence to obtain statistical results close to the analysis of the original distribution.

[0046] In summary, the tracing problem of pollution sources is transformed into the sampling problem of the posterior probability density function of pollution source parameters. The posterior probability density function of the parameters to be inverted is constructed based on Bayesian reasoning, and the MCMC method is used for alignment sampling. The present invention adopts the Metroplis-Hastings algorithm with strong applicability in the MCMC method to sample the posterior probability distribution. This process can obtain a large number of sample values ​​of the parameters to be inverted, and then summarize the posterior laws of the parameters based on sample statistics. The basic idea and calculation steps are as follows:

[0047] 1) Using the prior distribution of the parameter to be inverted and the likelihood function, Bayesian inference is used to obtain the posterior distribution and generate the initial state of the parameter

[0048] 2) Obtain a new parameter state from the proposed distribution (the present invention uses Gaussian distribution π(·)) Calculate the probability of acceptance

[0049] 3) Pick a random number u from the uniform distribution U(0,1), if u≤α(x i , x * ), then accept the new parameters generated in step 2 as the next state, otherwise reject.

[0050] 4) Keep or discard the generated new parameters according to the decision made in step 3).

[0051] 5) The number of iterations i increases, and steps 2)-4) are repeated.

[0052] Technical contribution of the present invention

[0053] Water pollution source tracing is to trace the source of the entire pollutant through the concentration changes of water pollutants monitored. It is the key and prerequisite for effectively obtaining the pollution situation of any water body. In actual situations, it is very difficult to accurately trace the source of pollutants. How to use limited monitoring data to invert key information such as the location of the pollution source requires a set of effective methods for tracing the source of pollution. The present invention designs a water pollution source tracing method based on Bayesian reasoning and MCMC sampling, which effectively solves these problems. Compared with existing work, the main contributions of the present invention are in the following aspects:

[0054] (1) Design a unique convection-diffusion model and establish the functional distribution of pollution source parameters based on Bayesian reasoning.

[0055] (2) The more applicable MCMC probability distribution sampling method is used to simplify the sampling of high-dimensional complex probability distributions, effectively solving the problems of high dimension of the parameters of the posterior distribution probability function, poor continuity and differentiability, and extremely complex calculation in Bayesian reasoning.

[0056] (3) The method of sampling and extracting parameter numerical summaries is used to solve the difficulty of obtaining parameters from complex and high-dimensional function distributions, and a more accurate pollution source tracing method is achieved in the given scenario.

[0057] The present invention provides a method for tracing the source of water pollution based on Bayesian reasoning and MCMC sampling proposed above. It should be understood that the detailed description of the technical solution of the present invention with the help of the preferred embodiment is illustrative rather than restrictive. A person of ordinary skill in the art may modify the technical solutions described in the various embodiments based on reading the specification of the present invention, or replace some of the technical features therein with equivalents, however, these modifications or replacements do not deviate the essence of the corresponding technical solution from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A water pollution source tracing method based on Bayesian reasoning and MCMC sampling, characterized in that: The method comprises the following steps: 1) Obtain tracer experimental data from the river monitored over a period of time; 2) Processing the data to obtain sampling concentration data C, and studying the relationship between the sampling concentration data C and the pollution source emission location parameter s, the pollution source emission mass m, and the pollutant emission time t; 3) Based on 2), design a convection-diffusion model based on water pollutants; 4) Substitute the convection-diffusion model obtained in 3) into the Bayesian framework, establish a likelihood function under the assumption that the error is uncorrelated with the mean square error, and obtain the posterior distribution of the parameters to be inverted of the pollution source based on the experimental data through formula calculation; 5) Perform MCMC sampling on the complex posterior distribution outputted in 4) to obtain a numerical summary of the parameters to be inverted.

2. The method according to claim 1, characterized in that: Pollutant source tracing based on water pollution source tracing methods, including: Based on the numerical summary of the parameters to be inverted, possible sources of water pollution are determined; related water pollution incidents are tracked and traced based on the determined sources; and traceability results and related pollution source information are provided.

3. The method according to claim 1, characterized in that Sampling concentration data is obtained based on historical water quality monitoring tracer data, including: Study the relationship between the sampling concentration data C and the pollution source emission location parameter s, the pollution source emission mass m, and the pollutant emission time t; apply data processing techniques, such as interpolation, regression or correlation analysis, to determine the mathematical relationship between the sampling concentration data and the pollution source parameters.

4. The method according to claim 1, characterized in that: Design a unique convection-diffusion model and establish the functional distribution of pollution source parameters based on Bayesian reasoning, including: A mathematical model is established to describe the transmission, diffusion and attenuation of water pollutants in rivers, taking into account factors such as river velocity, channel shape, and characteristics of pollutants. The convection-diffusion model is combined with the Bayesian statistical method, taking into account measurement errors and prior knowledge, to derive the likelihood function and calculate the posterior distribution.

5. The method according to claim 1, characterized in that The complex posterior distribution of the output is sampled by MCMC to obtain a numerical summary of the parameters to be inverted, including: Through Bayesian reasoning, the initial state of the parameter is obtained by combining the prior distribution and likelihood function of the parameter to be inverted; a new parameter state is generated multiple times according to the proposed distribution (usually Gaussian distribution is selected), and the probability of accepting the state is calculated; the posterior distribution of the parameter to be inverted is approximated by generating random samples, and the statistical summary of the parameter is calculated, such as the mean, variance or probability density function.

Citation Information

Patent Citations

  • A Bayesian statistical traceability method for discharging industrial waste water exceeding the standard of sewage pipe network

    CN108897964A

  • Water pollution rapid tracing method and system

    CN111986064A

Cited By

  • Deep learning-based drainage basin non-point source pollution accurate tracing method

    CN120541444A

  • Bayesian reasoning-based water body pollution accident source identification method, equipment and medium

    CN120849843A

  • Water pollution value tracing method based on information fusion

    CN121009434A

  • Layered sampling method and system for soil remediation and storage medium

    CN122287360A

  • A pollution traceability causal inference method and system based on a spatial grid and mathematical morphology

    CN122864696A