Method and system for analyzing pollution source of rainwater pipe network

Abstract

The invention discloses a method and a system for analyzing a pollution source of a rainwater pipe network. The method comprises the following steps of: 1, obtaining a first key fluorescence characteristic parameter combination according to basic physicochemical indexes by adopting a fluorescence characteristic parameter prediction model; 2, three-dimensional fluorescence characteristic parameters of the pollution source are obtained, and a pollution source three-dimensional fluorescence characteristic data set is obtained; screening by adopting a multivariate statistical method to obtain a second key fluorescence characteristic parameter combination; 3, determining a fluorescence characteristic parameter combination according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination; 4, inputting the fluorescence characteristic parameter combination into the Bayesian mixture model to obtain the contribution rate of the pollution source to the receptor water quality, and completing the analysis of the pollution source of the rainwater pipe network; according to the method, the DOM fluorescent fingerprint information is predicted by adopting basic data; and based on the three-dimensional fluorescence characteristic parameters and in combination with a Bayesian mixture model, precise quantitative analysis of the pipe network pollution source is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of source apportionment, and particularly relates to a method and system for source apportionment of pollution sources in a rainwater pipe network. Background Art

[0002] With the continuous expansion of the scale of municipal pipe networks, including pipe network systems such as water supply, drainage, and gas. However, the pipe networks are prone to pollution during operation, especially the pollution problems in the drainage pipe networks are becoming increasingly prominent, and it is difficult to solve problems such as mismatched urban pipe network construction, combined sewer overflows, rain-sewage mixing, and pipe infiltration. Traditional methods for monitoring pollution in rainwater pipe networks mainly rely on the layout of limited online monitoring points and manual inspection and sampling, and some conventional water quality parameters can be obtained. However, due to the complex structure and wide range of the pipe networks, traditional methods are difficult to identify pollution sources in real time, comprehensively, and accurately, resulting in slow pollution emergency responses and large diffusion ranges, which exacerbate the environmental pollution risk. Therefore, the accurate identification of pollution sources in pipe networks is the key to achieving precise pollution control and scientific pollution control. Dissolved organic matter (DOM) is ubiquitous in urban rivers and pipe networks, and fluorescence properties are important optical properties. The DOM compositions of different pollution sources have huge fluorescence characteristics differences. Three-dimensional fluorescence spectroscopy (Excitation—Emission—Matrix Spectra, EEM) technology can characterize the fluorescence properties of DOM in water bodies, and pollution sources can be quickly and accurately identified by comparing fluorescence spectra, but the concentration and contribution rate of pollution sources cannot be quantitatively calculated. In addition, due to the non-portability of large fluorescence spectrometers and the complexity of experimental operations, it is difficult to obtain the fluorescence fingerprint characteristics of DOM in pipe networks in real time through three-dimensional fluorescence. Summary of the Invention

[0003] The present invention provides a method and system for source apportionment of pollution sources in a rainwater pipe network for the problems of the prior art.

[0004] The technical solution adopted by the present invention is: A method for source apportionment of pollution sources in a rainwater pipe network, comprising the following steps:

[0005] Step 1: Obtain the basic physical and chemical indexes and three-dimensional fluorescence characteristic parameters of the water samples in the rainwater pipe network, train a multi-objective machine learning model to obtain a fluorescence characteristic parameter prediction model; use the fluorescence characteristic parameter prediction model to obtain a key fluorescence characteristic parameter combination according to the basic physical and chemical indexes;

[0006] Step 2: Obtain the three-dimensional fluorescence characteristic parameters of the pollution sources to obtain a three-dimensional fluorescence characteristic data set of the pollution sources; use a multivariate statistical method to screen and obtain a second key fluorescence characteristic parameter combination;

[0007] Step 3: Determine the fluorescence characteristic parameter combination according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination;

[0008] Step 4: Input the fluorescence feature parameter combination obtained in Step 3 into the Bayesian mixture model to obtain the contribution rate of the pollution source to the water quality of the receptor, and complete the analysis of the pollution sources in the rainwater pipe network.

[0009] Furthermore, the multi-objective machine learning model in Step 1 is one of random forest, XGBoost, and support vector machine.

[0010] Furthermore, it also includes identifying the peak positions of the three-dimensional fluorescence feature parameters to obtain the peak intervals or peak center coordinates, and inputting the recognition results and basic physicochemical indexes into the fluorescence feature parameter prediction model.

[0011] Furthermore, the pollution source three-dimensional fluorescence feature data set in Step 2 includes three-dimensional fluorescence spectra;

[0012] Based on the three-dimensional fluorescence spectra, it is determined whether the pollution sources can be distinguished; the distinguishing process is as follows:

[0013] Take two parameters in the parameter combination as the X-axis and Y-axis, and project the characteristic data of the pollution source onto the scatter plot;

[0014] Judge the resolvability of the fluorescence characteristics based on whether there is overlap in the characteristic regions;

[0015] If the fluorescence characteristics are resolvable, the pollution sources can be distinguished; otherwise, they cannot be distinguished.

[0016] Furthermore, the multivariate statistical methods in Step 2 include principal component analysis, redundancy analysis, correlation analysis, and cluster analysis.

[0017] Furthermore, in Step 1, MSE, MAE, RMSE, and R 2 are used to evaluate the fluorescence feature parameter prediction model.

[0018] Furthermore, the input end of the Bayesian mixture model includes a grouping module for grouping the input data according to different time periods before and after rainfall.

[0019] Furthermore, an adaptive update mechanism is introduced during the training process of the Bayesian mixture model.

[0020] Furthermore, the fluorescence feature parameter prediction model in Step 2 is combined with SHAP analysis to obtain the first key fluorescence feature parameter combination.

[0021] A system for a method of analyzing pollution sources in a rainwater pipe network includes: a DOM feature parameter prediction module and a rainwater pipe network pollution source analysis module;

[0022] The DOM characteristic parameter prediction module is used to predict the fluorescence characteristic parameters based on the basic physical and chemical indexes of water quality, and obtain the first key fluorescence characteristic parameter combination;

[0023] The rainwater pipe network pollution source analysis module is used to obtain the three-dimensional fluorescence characteristic data set of pollution sources, and screen to obtain the second key fluorescence characteristic parameter combination; according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination, the Bayesian mixture model is used to obtain the contribution rate of pollution sources to the receptor water quality, and the rainwater pipe network pollution source analysis is completed.

[0025] The beneficial effects of the present invention are:

[0026] (1) Through the existing machine learning method, the present invention first establishes the relationship between the basic physical and chemical indexes of the pipe network water body and the three-dimensional fluorescence characteristics, and obtains the key fluorescence characteristic parameter combination; uses the basic data to predict the DOM fluorescence fingerprint information;

[0027] (2) Based on the three-dimensional fluorescence characteristics and combined with the Bayesian mixture model, the present invention realizes the accurate quantitative analysis of the pipe network pollution sources. Description of the Drawings

[0028] Figure 1 It is a schematic flow chart of the method of the present invention.

[0029] Figure 2 It is a schematic diagram of the training result of the XGBoost model in the embodiment of the present invention.

[0030] Figure 3 It is a schematic diagram of the SHAP analysis result of the machine learning model in the embodiment of the present invention.

[0031] Figure 4 It is a schematic diagram of the screening process of the key parameters for tracing in the embodiment of the present invention.

[0032] Figure 5 It is a schematic diagram of the pollution source DOM fingerprint information database in the embodiment of the present invention.

[0033] Figure 6 It is a schematic diagram of the calculation result of the pollution source analysis in the embodiment of the present invention. Detailed Embodiments

[0034] The following further describes the present invention in conjunction with the drawings and specific embodiments.

[0035] As Figure 1 shown, a method for analyzing pollution sources in a rainwater pipe network includes the following steps:

[0036] Step 1: Obtain the basic physical and chemical indexes and three-dimensional fluorescence characteristic parameters of the rainwater pipe network water samples, train a multi-objective machine learning model to obtain a fluorescence characteristic parameter prediction model; use the fluorescence characteristic parameter prediction model to obtain the first key fluorescence characteristic parameter combination according to the basic physical and chemical indexes;

[0037] The multi-objective machine learning model is one of random forest, XGBoost, and support vector machine. It also includes identifying the peak positions of the three-dimensional fluorescence characteristics to obtain the peak intervals or peak center coordinates, inputting the recognition results and the basic physical and chemical indexes into the multi-objective machine learning model, and combining SHAP analysis to obtain the first key fluorescence characteristic parameter combination; it also includes evaluating the multiple regression model.

[0038] Step 2: Obtain the three-dimensional fluorescence characteristics of the pollution sources to obtain a pollution source three-dimensional fluorescence characteristic data set; use multivariate statistical methods to screen and obtain the second key fluorescence characteristic parameter combination; the pollution source three-dimensional fluorescence characteristic data set includes three-dimensional fluorescence spectra;

[0039] Determine whether the pollution sources are distinguishable according to the three-dimensional fluorescence spectra. The process of determining whether the pollution sources are distinguishable according to the three-dimensional fluorescence spectra is as follows:

[0040] Take two parameters in the parameter combination as the X-axis and Y-axis, and project the characteristic data of the pollution sources onto a scatter plot;

[0041] Judge the resolvability of the fluorescence characteristics based on whether there is overlap in the characteristic regions;

[0042] If the fluorescence characteristics are resolvable, the pollution sources are distinguishable; otherwise, they are not.

[0043] The multivariate statistical methods include principal component analysis, redundancy analysis, correlation analysis, and cluster analysis.

[0044] Step 3: Determine the fluorescence characteristic parameter combination according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination; the input end of the Bayesian mixture model includes a grouping module for grouping the input data according to different time periods before and after rainfall. The Bayesian mixture model introduces an adaptive update mechanism during the training process.

[0045] Step 4: Input the fluorescence characteristic parameter combination obtained in Step 3 into the Bayesian mixture model to obtain the contribution rate of the pollution sources to the receptor water quality, and complete the analysis of the pollution sources in the rainwater pipe network.

[0046] A system for analyzing pollution sources in a rainwater pipe network includes: a DOM characteristic parameter prediction module and a rainwater pipe network pollution source analysis module;

[0047] The DOM characteristic parameter prediction module is used to predict the fluorescence characteristic parameters according to the basic physical and chemical indexes of the water quality to obtain the first key fluorescence characteristic parameter combination;

[0048] The pollution source analysis module for the rainwater pipe network is used to obtain a three-dimensional fluorescence characteristic data set of pollution sources, and screen to obtain a second key fluorescence characteristic parameter combination; according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination, a Bayesian mixture model is used to obtain the contribution rate of pollution sources to the quality of the receptor water, and the pollution source analysis of the rainwater pipe network is completed.

[0049] Embodiment

[0050] A method for analyzing pollution sources in a rainwater pipe network includes the following steps:

[0051] Step 1: Obtain the basic physical and chemical indexes and three-dimensional fluorescence characteristic parameters of the rainwater pipe network water samples, train a multi-objective machine learning model to obtain a fluorescence characteristic parameter prediction model; use the fluorescence characteristic parameter prediction model to obtain a first key fluorescence characteristic parameter combination according to the basic physical and chemical indexes.

[0052] First, collect water samples from key pollution sections of the rainwater pipe network. The measured basic physical and chemical indexes include redox potential, pH value, conductivity, total dissolved solids (TDS), chemical oxygen demand (COD), total nitrogen (TN), ammonia nitrogen (NH4 + -N) and total phosphorus (TP), etc. At the same time, use EEM spectroscopy technology to measure the three-dimensional fluorescence characteristics of DOM in the water body, including fluorescence peak intensities (Peak A, Peak C, and Peak T), biological index (BIX), humification index (HIX), and fluorescence index (FI).

[0053] Using the above basic physical and chemical indexes and EEM fluorescence characteristic parameters, a multi-objective machine learning model, such as random forest, XGBoost, support vector machine, etc., is used to mine the non-linear relationship between the two under a multi-output regression framework. Train the machine model, and input the actually measured basic physical and chemical indexes into the trained machine model to obtain the first key fluorescence characteristic parameter combination, such as any combination of Peak A, Peak C, Peak T, BIX, HIX, FI, etc.

[0054] During the training process, considering the influence of the water environment or the nature of pollution sources on the fluorescence peak position, the peak position recognition is used as a pre-classification step to perform a partitioned peak extraction process. Input the classification results (identified peak intervals or peak center coordinates) together with the basic physical and chemical indexes into the regression model to correct the prediction results of parameters such as fluorescence peak intensity, BIX, HIX, FI, etc. Enhance the robustness of the model to peak drift, overlap, and noise. Further, through SHAP analysis, identify the parameter combinations that play a key role in model prediction and improve the interpretability of the machine learning model.

[0055] It is also possible to use indicators such as MSE, MAE, and RMSE to evaluate the prediction performance of each machine learning algorithm, determine the optimal model with the best training effect (such as XGBoost), and its optimal parameter combinations (such as C / T and HIX, C / A and HIX, A / T and HIX, etc.). In the present invention, DOM characteristic data equivalent to the actual three-dimensional fluorescence measurement (Peak A, Peak C, Peak T, BIX, HIX, FI, etc.) can be obtained using basic physical and chemical indicators, breaking through the bottleneck that it is difficult to obtain DOM fluorescence fingerprint information in real time and on a large scale. The purpose of predicting advanced data (such as fluorescence peak intensity, biological index, humification index, and fluorescence index, etc.) with basic physical and chemical indicator data (such as pH, conductivity, chemical oxygen demand, ammonia nitrogen concentration, etc.) is achieved. The problem that it is difficult to obtain a large amount of DOM fluorescence fingerprint information of water samples in real time is overcome.

[0056] Step 2: Obtain the three-dimensional fluorescence characteristics of the pollution sources to obtain a three-dimensional fluorescence characteristic dataset of the pollution sources; use multivariate statistical methods to screen and obtain the second key fluorescence characteristic parameter combination.

[0057] By taking a large number of samples of typical urban pollution sources such as catering wastewater, medical wastewater, domestic sewage, balcony wastewater, and industrial sewage, and combining with parallel factor analysis (PARAFAC) technology, the three-dimensional fluorescence characteristics of DOM of each source (three-dimensional fluorescence spectrum, Peak A, Peak C, Peak T, BIX, HIX, FI, etc.) are obtained, and based on this, a three-dimensional fluorescence characteristic database of pollution sources within the research area is constructed to form a three-dimensional fluorescence characteristic dataset of the pollution sources.

[0058] In the three-dimensional fluorescence spectrum of the pollution source, different parameter combinations (such as A / T and HIX, C / A and HIX) are used as the X-axis and Y-axis respectively, and the characteristic data of the pollution source are projected onto a scatter plot. The resolvability of the fluorescence characteristics is judged by whether there is overlap in the characteristic area. Among them, for the parameter combination of C / A and HIX, only domestic sewage and catering wastewater cannot be distinguished and are all regarded as domestic sewage, and other pollution sources such as medical wastewater, industrial sewage, and balcony wastewater can be distinguished. Analyze each three-dimensional fluorescence characteristic parameter of the water samples in the pipe network. Determine whether the sample distribution is dispersed or significantly clustered through principal component analysis, whether the absolute value of the load of different parameters on the sample is high through redundancy analysis, and determine the significant relationship between the parameters through correlation analysis. If two parameters are highly positively or negatively correlated, there is redundancy and the contribution rate calculation effect is limited.

[0059] Through the above analysis, parameter combinations that meet the set threshold requirements are obtained, thereby obtaining key fluorescence characteristic parameter combinations suitable for tracing, such as A / T and HIX, C / A and HIX, etc.

[0060] Step 3: Determine the fluorescence characteristic parameter combination according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination;

[0061] The parameter combinations obtained in Step 1, such as C / T and HIX, C / A and HIX, A / T and HIX, etc., and the parameter combinations obtained in Step 2, such as A / T and HIX, C / A and HIX, etc.; introduce expert knowledge constraints (i.e., pre-set threshold and other judgment criteria). For example, A / T is the fluorescence intensity ratio of fulvic acid and protein-like substances. Protein-like substances are easily degraded in the environment, so this parameter is not stable. While C / A is the fluorescence peak intensity ratio of humic acid and fulvic acid. These two types of substances are relatively stable and difficult to degrade in the environment, making this parameter relatively constant in the environment. Finally, it is determined that the parameter combination of C / A and HIX can best distinguish different pollution sources in terms of environment and statistical significance.

[0062] Step 4: Input the fluorescence characteristic parameter combination obtained in Step 3 into the Bayesian mixture model to obtain the contribution rate of pollution sources to the water quality of the receptor, and complete the pollution source analysis of the rainwater pipe network. The Bayesian mixture model quantitatively calculates the contribution rate of each pollution source to the water quality of the receptor through probability inference, solving the problem that most existing source analysis models can only be qualitative or difficult to accurately quantify.

[0063] Since the pollutant composition in the rainwater pipe network is greatly affected by rainfall, traditional end-member mixture models often assume that all observations come from the same fixed prior distribution and cannot distinguish the differences in end-member characteristics during different time periods or different rainfall processes. The input end of the Bayesian mixture model includes a grouping module (hierarchical structure, Dirichlet hierarchical structure) for grouping the input data according to different time periods before and after rainfall. And a "dynamic correction term" is introduced for each time period, enabling its end-member characteristics to be locally adjusted on the basis of the global prior. Finally, the end-member characteristics within the same time period have a certain correlation, but there are independent prior corrections in different time periods. Dynamically track the change characteristics of pipe network pollutants affected by rainfall, and further improve the model's ability to perceive and quantitatively analyze pollution during rainstorm events.

[0064] An uncertainty evaluation and adaptive update mechanism is introduced into the Bayesian end-member mixture model. By statistically analyzing the distribution of historical monitoring data, the prior distribution of the parameters is set to follow a distribution with interval or segmented characteristics, and the prior of the model is dynamically updated according to the observation results of real-time new data. By introducing the above adaptive mechanism, the model can still maintain a high-precision tracing ability in the ever-changing rainwater pipe network environment, and can be dynamically adjusted according to the dynamic changes of data, forming a closed loop of "monitoring while correcting".

[0065] The above method and system can be used to prepare virtual sensors. In the following embodiments, a device or structure containing the method or system of the present invention is called an on-line monitoring sensor.

[0066] The method of the present invention can be introduced into the online monitoring system of municipal pipe networks to obtain real-time basic pipe network data. These data include basic physical and chemical indexes such as redox potential, pH value, conductivity, total dissolved solids, chemical oxygen demand, total nitrogen, ammonia nitrogen, and total phosphorus. All sensor data will be continuously uploaded to the background system as the basic data source for real-time monitoring. The system will continuously collect the basic physical and chemical index data of each key node of the pipe network and store these data. Long-term data monitoring can update the training set in real time, improving the accuracy and prediction ability of the model.

[0067] When abnormal water quality changes or signs of pollution sources (such as a sudden increase in the concentration of a certain pollutant) are detected, it will immediately judge whether a pollution source appears, and then issue an instruction to trigger the process of pollution source analysis of the present invention. Once it is determined that pollution has occurred, the pollution source analysis process will be immediately started based on the collected real-time data. The system uses the existing basic physical and chemical index data to quickly make predictions to obtain corresponding DOM characteristic data. The above-mentioned key tracer index data obtained by prediction are input into the Bayesian end-member mixing model. Based on the previously trained pollution source analysis data, this model can accurately calculate the types and concentrations of pollution sources. The Bayesian model combines historical data and real-time data to analyze the relative contribution degrees of each potential pollution source and determine the size and influence range of the pollution source. The system will output the pollution source analysis results, including information such as the location of the pollution source, pollutant concentration, and relative contribution degree of the pollution source. All results will be presented to the pipe network management personnel through a graphical interface to help them comprehensively understand the specific situation of the pollution source.

[0068] Embodiment

[0069] 15 water quality index online monitoring sensors are arranged in the rainwater pipe network of a certain city. Each sensor collects the basic water quality data in the rainwater pipe network at intervals for a period of time, including: redox potential (ORP), acidity and alkalinity (pH), conductivity (EC), total dissolved solids (TDS), chemical oxygen demand (COD), total nitrogen (TN), ammonia nitrogen (NH 4+ -N), total phosphorus (TP), and the water quality parameter statistical table is shown in Table 1.

[0070] Table 1. Water quality parameter statistical table

[0071]

[0072]

[0073] While the sensors are collecting data, water samples are taken during the same time period as the data collected by the sensors, and a total of 165 water samples are taken. The collected water samples are quickly transported to the laboratory for three-dimensional fluorescence spectroscopy measurement in a low-temperature and light-shielded environment, and the DOM fluorescence spectral characteristics of the water samples are extracted. The fluorescence spectral characteristic statistical table is shown in Table 2.

[0074] Table 2. Statistical Table of Fluorescence Spectral Characteristics of DOM in Water Samples

[0075]

[0076] The complex non - linear relationship between the basic water quality physical and chemical indexes and DOM fluorescence characteristics collected in the above steps was extracted using random forest, XGBoost, and support vector machine algorithms, and parameters such as MSE, MAE, and RMSE were used to evaluate the performance of the models used, screening the optimal model and the best source tracing parameter combination. The results are shown in Table 3 and Figure 2 as follows.

[0077] Table 3. XGBoost Model

[0078]

[0079]

[0080] From Table 3 and Figure 2 it can be seen that under the XGBoost algorithm, the prediction effects of C / T and HIX are the best, and the R 2 values of the dual - target training set both reach 0.9999, and the R 2 values of the test set reach 0.851 and 0.968 respectively. In addition, parameter combinations such as A / T and HIX, C / A and HIX also achieved good prediction effects. Considering that the C / A in the water body is more stable than C / T, the parameter combination of C / A and HIX was finally selected. At the same time, SHAP analysis was used to identify that TN is the basic parameter with the greatest impact on the model, as Figure 3 shown.

[0081] By establishing a pollution source characteristic database of the rainwater pipe network as Figure 5 shown, combined with correlation analysis, principal component analysis, and redundancy analysis, parameter combinations suitable for source analysis were obtained, including A / T and HIX, C / A and HIX, etc. Combining with the parameter combinations with good prediction performance in machine learning, the parameter combination suitable for pollution source analysis of the rainwater pipe network was finally determined to be C / A and HIX, as Figure 4 shown. Finally, the MixSIAR (Bayesian mixing model) of the end - member mixing model was used to calculate the contribution rate of each pollution source to the water body at each sampling point of the pipe network, as shown in Table 4 and Figure 6 shown.

[0082] Table 4. Statistical Table of Contribution Rate Calculation of MixSIAR (Bayesian Mixing Model)

[0083]

[0084] The present invention first establishes the relationship between the basic physical and chemical indicators of the pipe network water body and the three-dimensional fluorescence characteristics through machine learning methods (e.g., random forest, XGboost, neural network model). At the same time, parameters such as MSE, MAE, and RMSE are used to evaluate the performance of the models used. Then, the trained network model is used to predict advanced data (e.g., biological index, humification index, and fluorescence index) with basic data (e.g., pH value, conductivity, chemical oxygen demand, ammonia nitrogen concentration, etc.). It overcomes the problem that it is difficult to obtain a large amount of DOM fluorescence fingerprint information of water samples in real time. The Bayesian mixture model is used to model the polluted water samples in the rainwater pipe network and various pollution sources. Combining the previous data with the real-time monitoring data, the relative contribution rates of each pollution source are accurately calculated through probability inference, so as to obtain the specific contributions of different pollution sources to the pipe network pollution. This model can accurately calculate the contribution rates of each pollution source, making up for the defect that most models can only perform qualitative analysis but not quantitative calculation. The peak position classification and discrimination are additionally introduced to solve the problem that the peak position in the EEM spectrum may shift slightly with changes in water quality, pollution source type, etc.

[0085] Through the "source characteristic database" constructed by parallel factor analysis (PARAFAC) or previous multivariate statistics, machine learning is introduced to optimize the combination of key parameters (such as C / A, HIX, etc.), and a fast discrimination or clustering algorithm is built into the model, which can efficiently complete the "source type identification - mixing ratio analysis" process even at a large data scale. If the original Bayesian endmember mixture model adopts a single prior distribution, the inference speed will slow down under a large number of source samples or complex working conditions. Through hierarchical, time series and other structures, the contribution rates of multi-source pollutants can be converged and solved more quickly. By introducing an uncertainty evaluation and adaptive update mechanism into the Bayesian mixture model, the present invention can dynamically correct the prior and quantify the confidence interval of the prediction results when dealing with measurement errors, missing data or sudden environmental changes, ensuring the robustness of the model in a changing situation. By introducing a time factor or a state space model, it is allowed that the characteristics of the endmembers (fluorescence peak intensity distribution or physical and chemical index range) evolve with time, so that the model can maintain the ability to track the changes of pollution sources in a long time series. In the face of complex dynamic situations such as continuous changes in the composition of pollution sources, pipe network renovation or new sewage outlets, the model can be quickly iterated to maintain high precision and high adaptability with minimal manual intervention, truly realizing real-time intelligent management and control.

[0086] The present invention utilizes the online monitoring data of the basic physical and chemical indicators in the pipe network to realize the real-time monitoring and analysis of the pipe network pollution sources. This method can output the pollution concentration information in real time after the occurrence of water body pollution in the pipe network, significantly improving the speed and efficiency of emergency response. The rapid identification of pollution sources provides effective decision-making support for pipe network managers, helps to take corresponding measures in time, reduce the scope of pollution diffusion and effectively control the impact of pollution incidents.

Claims

1. A method for analyzing pollution sources in a rainwater pipe network, characterized in that, It includes the following steps: Step 1: Obtain the basic physicochemical indexes and three-dimensional fluorescence characteristic parameters of the rainwater pipe network water samples, train a multi-objective machine learning model to obtain a fluorescence characteristic parameter prediction model; use the fluorescence characteristic parameter prediction model to obtain the first key fluorescence characteristic parameter combination according to the basic physicochemical indexes; Step 2: Obtain the three-dimensional fluorescence characteristic parameters of the pollution sources to obtain a three-dimensional fluorescence characteristic data set of the pollution sources; use multivariate statistical methods to screen and obtain the second key fluorescence characteristic parameter combination; Step 3: Determine the fluorescence characteristic parameter combination according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination; Step 4: Input the fluorescence characteristic parameter combination obtained in Step 3 into the Bayesian mixture model to obtain the contribution rate of the pollution sources to the receptor water quality, and complete the analysis of the pollution sources in the rainwater pipe network.

2. The method for analyzing pollution sources of a rainwater pipe network according to claim 1, wherein In Step 1, the multi-objective machine learning model is one of random forest, XGBoost, and support vector machine.

3. The method for analyzing pollution sources of a rainwater pipe network according to claim 2, wherein, It also includes identifying the peak positions of the three-dimensional fluorescence characteristic parameters to obtain the peak intervals or peak center coordinates, and inputting the identification results and the basic physicochemical indexes into the fluorescence characteristic parameter prediction model.

4. A method for analyzing pollution sources in a rainwater pipe network according to claim 1, characterized in that, In Step 2, the three-dimensional fluorescence characteristic data set of the pollution sources includes three-dimensional fluorescence spectra; Whether the pollution sources can be distinguished is obtained according to the three-dimensional fluorescence spectra; the distinguishing process is as follows: Take two parameters in the parameter combination as the X-axis and Y-axis, and project the characteristic data of the pollution sources onto a scatter plot; Judge the resolvability of the fluorescence characteristics based on whether the characteristic regions overlap; If the fluorescence characteristics are resolvable, the pollution sources can be distinguished; otherwise, they cannot be distinguished.

5. The method for analyzing pollution sources of a rainwater pipe network according to claim 4, wherein, In Step 2, the multivariate statistical methods include principal component analysis, redundancy analysis, correlation analysis, and cluster analysis.

6. The method for analyzing pollution sources of a rainwater pipe network according to claim 1, characterized in that, In step 1, MSE, MAE, RMSE, and R are used 2 to evaluate the fluorescence feature parameter prediction model 7. A method for analyzing pollution sources in a rainwater pipe network according to claim 1, characterized in that, The input end of the Bayesian mixture model includes a grouping module for grouping the input data according to different time periods before and after rainfall.

8. A method for analyzing pollution sources of a rainwater pipe network according to claim 1, characterized in that, An adaptive update mechanism is introduced during the training process of the Bayesian mixture model.

9. A method for analyzing pollution sources of a rainwater pipe network according to claim 1, characterized in that, In Step 2, the fluorescence characteristic parameter prediction model combines SHAP analysis to obtain the first key fluorescence characteristic parameter combination.

10. A system adopting any one of the methods for analyzing pollution sources in rainwater pipe networks according to claims 1 to 9, characterized in that, It includes: A DOM characteristic parameter prediction module and a rainwater pipe network pollution source analysis module; The DOM characteristic parameter prediction module is used to predict the fluorescence characteristic parameters according to the basic physicochemical indexes of the water quality to obtain the first key fluorescence characteristic parameter combination; The rainwater pipe network pollution source analysis module is used to obtain a three-dimensional fluorescence characteristic data set of the pollution sources, screen and obtain the second key fluorescence characteristic parameter combination; use the Bayesian mixture model according to the first key fluorescence characteristic parameter combination and the second key fluorescence characteristic parameter combination to obtain the contribution rate of the pollution sources to the receptor water quality, and complete the analysis of the pollution sources in the rainwater pipe network.

Images