Reservoir multi-medium pollution source identification method and device based on machine learning

Abstract

The invention belongs to the related technical field of water environment monitoring, and discloses a reservoir multi-medium pollution source identification method and equipment based on machine learning, and the method comprises the steps: (1) combining principal component analysis with multiple linear regression to construct an absolute principal component-multiple linear regression model; (2) verifying the absolute principal component-multiple linear regression model by utilizing pollution source measured data of runoff input, atmospheric sedimentation and water-soil interface release so as to determine pollution source parts corresponding to different principal components of the model; (3) based on reservoir water quality monitoring data, calculating and simulating contribution degrees of runoff input, atmospheric settlement and water-soil interface input pollution sources in different time and space of the reservoir by using the model; and (4) calculating atmospheric settlement and a water-soil interface input pollution source by using the obtained contribution degree, the runoff input pollution source and the model, so as to identify and obtain the spatial-temporal change process of the multi-medium pollution source. The method improves the recognition speed.

Description

Technical Field

[0001] The present invention belongs to the technical field related to water environment monitoring, and more specifically, relates to a method and device for identifying multi-media pollution sources in a reservoir based on machine learning. Background Art

[0002] Accurately grasping the water quality status of the water source and the spatio-temporal variation of pollution sources is an important task for the management of reservoir-type drinking water sources. Effective pollution source identification can enable relevant agencies and managers to master the change patterns of water quality parameters in the reservoir area and identify potential adverse impacts on the water body. Therefore, the development and application of reliable spatio-temporal variation identification models for pollution sources are crucial for effective water resource management and environmental protection.

[0003] Pollution source analysis methods are divided into forward and reverse tracing. The traditional pollution source inventory method, as an effective method for forward tracing, determines the pollutant input coefficient into the river and then determines the pollutant emission process. However, it is difficult for this method to adapt to the emission status of various pollution sources with spatio-temporal heterogeneity in the study area, and the uncertainty of the emission coefficient also makes it uncontrollable in tracing applications. Another forward tracing method is the mechanism model. Although this method can calibrate the input coefficient into the river, it requires rich indoor test data and complex mathematical model support, and it is difficult to apply to research areas with large reservoir areas and numerous pollution sources.

[0004] Reverse tracing methods include principal component analysis, chemical mass balance method (CMB), etc. These methods are widely used in the analysis of pollution sources in the atmosphere, soil, and groundwater. Principal component analysis, as a commonly used reverse tracing method, although it can comprehensively analyze the pollution sources, is limited to qualitative determination and cannot quantify the contribution of pollution sources. Although the chemical mass balance method (CMB) can quantitatively analyze the contribution of pollution sources, it requires prior information such as clear pollution source information and pollution source component spectra, which limits the application of the CMB model to water bodies.

[0005] The absolute principal component-multiple linear regression (APCS-MLR) model, as a new type of reverse tracing method, calculates the absolute factor scores of principal component analysis and combines them with the multiple linear regression model to calculate the contribution rate of each principal component to water body indicators. With the in-depth study of the absolute principal component-multiple linear regression (APCS-MLR) model, the research results of the absolute principal component-multiple linear regression (APCS-MLR) model mainly include:

[0006] "Environmental Geochemistry and Health", Volume 42, Pages 3795 - 3810 in 2020, "Source apportionment of water pollutants in the upstream of yangtze river using APCS–MLR" analyzed the pollution sources of the main stream of the Jinsha River and its three main tributaries, and considered that minerals, meteorology, domestic wastewater and agricultural non-point sources are the main pollution sources based on experience.

[0007] "China Rural Water and Hydropower", Issue 12, Pages 113 - 118 in 2023, "Identification and contribution rate analysis of water pollution sources in reservoir-type drinking water sources based on APCS-MLR" took the Shanxi Reservoir in Wenzhou City as the research object, and used the method of combining absolute principal component evaluation and multiple linear regression to quantitatively evaluate the water quality status of the reservoir, identify the types of pollution sources, and analyze the distribution and contribution rate of pollution sources.

[0008] However, the existing research on identifying pollution sources using the above-mentioned absolute principal component - multiple linear regression (APCS-MLR) model still has the following problems and defects:

[0009] (1) The contributions of pollution sources vary with time and space, and the existing research has not analyzed the temporal and spatial variations of the contributions of pollution sources in detail;

[0010] (2) The input amounts of pollution sources such as atmospheric deposition and release at the water-soil interface cannot be simulated and analyzed quickly and effectively. Summary of the Invention

[0011] In view of the above-mentioned defects or improvement requirements of the existing technology, the present invention provides a method and device for identifying multi-media pollution sources in reservoirs based on machine learning, aiming to solve the problem that the existing methods cannot simulate and analyze the input amounts of pollution sources such as atmospheric deposition and release at the water-soil interface quickly and effectively.

[0012] To achieve the above object, according to one aspect of the present invention, there is provided a method for identifying multi-media pollution sources in reservoirs based on machine learning, the method comprising the following steps:

[0013] (1) Based on the reservoir water quality monitoring data, combine principal component analysis and multiple linear regression to construct an absolute principal component - multiple linear regression model, and the mathematical expression of the absolute principal component - multiple linear regression model is:

[0014]

[0015] In the formula, C k is the measured concentration of the k water chemical factors, mg / L; a jkis the regression coefficient of pollution source j on water chemical factor k; APCS jk is the absolute principal component score value of all samples of water chemical factor k; b k is the constant term;

[0016] (2) Use the measured data of pollution sources from runoff input, atmospheric deposition, and water-soil interface release to verify the absolute principal component-multiple linear regression model to determine the pollution source parts corresponding to different principal components of the absolute principal component-multiple linear regression model;

[0017] (3) Based on the reservoir water quality monitoring data, use the absolute principal component-multiple linear regression model to calculate and simulate the contribution degrees of runoff input, atmospheric deposition, and water-soil interface input pollution sources at different times and spaces in the reservoir;

[0018] (4) Use the obtained contribution degrees, runoff input pollution sources, and the absolute principal component-multiple linear regression model to calculate the atmospheric deposition and water-soil interface input pollution sources, so as to identify the spatio-temporal variation process of multi-media pollution sources.

[0019] Furthermore, construct a water quality data set using water quality monitoring data, evaluate the applicability of the water quality data set through the KMO sampling adequacy test and Bartlett sphericity test, and screen the water quality indicators of reservoir pollution to construct a training data set.

[0020] Furthermore, use the Z-Score method to standardize the water quality training data, and the corresponding calculation formula is:

[0021]

[0022] where z is the standardized value, x is the original value, is the sample mean, is the sample mean square deviation.

[0023] Furthermore, calculate the covariance matrix between water quality indicators, obtain multiple groups of eigenvectors and eigenvalues, and select the components with the first three eigenvalues greater than or equal to 1 as the principal components; calculate the scores of each principal component according to the eigenvectors of the principal components and the standardized values of the water quality indicators.

[0024] Furthermore, construct a covariance matrix through covariance calculation between indicators, arrange the obtained eigenvalues from large to small and denote them as λ 1 ≥λ 2 ≥…≥λ p ≥0, and the corresponding eigenvectors are e 1 , e 2 ,…, e p , and the contribution rate of each eigenvalue is The cumulative contribution rate is

[0025] Furthermore, the principal component score is a linear combination of the principal component eigenvector and the standardized values of each original variable, and the comprehensive score is the weighted mean of the principal component scores. The calculation formula is as follows:

[0026]

[0027] In the formula, A j is the score of the j-th principal component; Z k is the standardized value of the k-th water quality metric; w jk is the loading coefficient of the j-th principal component on the k-th water quality index;

[0028] The principal component score needs to be converted into a non-standard absolute principal component score for the contribution analysis of pollution indicators. The absolute principal component score can reflect the contribution of pollution sources to water quality indicators. The calculation formula of the absolute principal component score is as follows:

[0029] APCS jk =(A j ) k -(A 0 ) j

[0030]

[0031]

[0032] In the formula: APCS jk is the absolute principal component score value of the j-th principal component; (A j ) k is the score value of the j-th principal component; (A 0 ) j is the score value of the j-th principal component at 0 value; S jk is the factor score coefficient of the j-th principal component on the water quality factor k, where i is the serial number of the hydrochemical factor; (Z 0 ) k is the standardized value when the pollutant concentration at the observation point is set to 0; is the arithmetic mean of the k-th water quality factor concentration, mg / L; σ k is the standard deviation of the k water quality factor concentration, mg / L.

[0033] Furthermore, in the absolute principal component - multiple linear regression model, a jk ×APCS jk is the contribution of pollution source m to the water quality index concentration. The calculation formula for the pollution contribution degree PC my of each principal component is as follows:

[0034]

[0035] The calculation formula for the contribution of other pollution sources is as follows:

[0036]

[0037] Furthermore, based on the APCS scores and the measured water quality concentrations, a multiple linear regression equation for the pollution source contribution is established using the multiple linear regression method to obtain the absolute principal component - multiple linear regression model.

[0038] The present invention also provides a reservoir multi - medium pollution source identification system based on machine learning. The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the above - mentioned reservoir multi - medium pollution source identification method based on machine learning.

[0039] The present invention also provides a computer - readable storage medium. The computer - readable storage medium stores machine - executable instructions. When the machine - executable instructions are called and executed by the processor, the machine - executable instructions prompt the processor to implement the above - mentioned reservoir multi - medium pollution source identification method based on machine learning.

[0040] Generally speaking, compared with the prior art through the above - conceived technical solution of the present invention, the reservoir multi - medium pollution source identification method and device provided by the present invention mainly have the following

[0041] Beneficial effects:

[0042] 1. The identification method combines principal component analysis and multiple linear regression to construct an absolute principal component - multiple linear regression model, and uses the obtained contribution degree, runoff - input pollution sources and the absolute principal component - multiple linear regression model to calculate the atmospheric deposition and soil - water interface input pollution sources, thereby identifying the spatio - temporal variation process of multi - medium pollution sources, and then realizing the rapid and effective simulation of the pollution source input amount.

[0043] 2. Use the measured data of the pollution sources of runoff input, atmospheric deposition and soil - water interface release to verify the absolute principal component - multiple linear regression model to determine the pollution source parts corresponding to different principal components of the absolute principal component - multiple linear regression model, and then use APCS - MLR to simulate the spatio - temporal variation of pollution sources.

[0044] 3. The identification method can identify the contributions of reservoir pollution sources with different temporal and spatial variations, and use the measurable data to simulate the pollution sources that are not easy to monitor, so as to comprehensively grasp the change trend of pollution sources in the reservoir area.

[0045] 4. The present invention combines principal component analysis and multiple linear regression models, which can not only identify the contributions of reservoir pollution sources with different temporal and spatial variations, but also quantify the specific impacts of pollution sources on water quality indicators through the transformation of absolute principal component scores, thereby providing a scientific basis for the precise tracing and zonal treatment of pollution sources and enhancing the pertinence and efficiency of pollution control. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flow chart of a method for identifying multi-media pollution sources in a reservoir based on machine learning provided by the present invention;

[0047] Figure 2 is a schematic diagram of the reservoir area and water quality monitoring stations;

[0048] Figure 3 In (a), (b), (c), and (d) of

[0049] Figure 4 are schematic diagrams of the contributions of different pollution sources;

[0050] Figure 5 is a schematic diagram of the temporal variation trend of the input amounts of pollution sources simulated by the model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0052] The present invention provides a method for identifying multi-media pollution sources in a reservoir based on machine learning. The identification method constructs an absolute principal component - multiple linear regression model (APCS-MLR) to identify three parts of pollution sources, namely runoff input, atmospheric deposition, and water-soil interface release in the reservoir, and calculates the contribution degrees of different pollution sources in the reservoir in terms of time or space. The contribution degrees of the model to each pollution source are verified through measured runoff input, atmospheric deposition, and water-soil interface input data, and then the APCS-MLR is used to simulate the temporal and spatial variations of pollution sources.

[0053] Please refer to Figure 1 , the identification method mainly includes the following steps:

[0054] Step 1: Based on the reservoir water quality monitoring data, combine principal component analysis and multiple linear regression to construct an absolute principal component - multiple linear regression model. The mathematical expression of the absolute principal component - multiple linear regression model is:

[0055]

[0056] Wherein, C k is the measured concentration of the k water chemical factor, mg / L; a jk is the regression coefficient of pollution source j on water chemical factor k; APCS jk is the absolute principal component score value of all samples of water chemical factor k; b k is the constant term.

[0057] The construction of the absolute principal component-multiple linear regression model mainly includes the following sub-steps:

[0058] (1) Data preprocessing and sample selection: Use water quality monitoring data to construct a water quality data set, evaluate the applicability of the water quality data set through the KMO sampling adequacy test (KMO value greater than 0.6) and the Bartlett sphericity test (test value less than 0.05), screen the main water quality indicators of reservoir pollution, and construct a training data set. Among them, when KMO≥0.6, the PCA sample size requirement is met; further test the mutual independence between variables. If the sig. of the Bartlett sphericity ≤0.05, the PCA applicability is met.

[0059] (2) Standardize the water quality training data using the Z-Score method, and the corresponding calculation formula is:

[0060]

[0061] Where z is the standardized value, x is the original value, is the sample mean, is the sample mean square deviation.

[0062] (3) Calculate the covariance matrix between water quality indicators, obtain multiple groups of eigenvectors and eigenvalues, and select the first three components with eigenvalues greater than or equal to 1 as the principal components. Calculate the scores of each principal component according to the eigenvectors of the principal components and the standardized values of the water quality indicators. Assuming there are p measurement indicators and the sample size is n, the covariance calculation formula between any two indicators is:

[0063]

[0064] Where z ik is the standardized value of water quality indicator i on sample k; is the mean of water quality indicator i; z jk is the standardized value of water quality indicator j on sample k; is the mean of water quality indicator j.

[0065] Construct a covariance matrix by calculating the covariance between indicators, and arrange the obtained eigenvalues from largest to smallest as λ 1 ≥ λ 2 ≥ … ≥ λ p ≥ 0, and the corresponding eigenvectors are e 1 , e 2 , …, e p , and the contribution rate of each eigenvalue is The cumulative contribution rate is

[0066] (4) Calculate the loading coefficient of the principal component, and judge the interpretability of the principal component for the original variables. The score of a certain principal component is the linear combination of the eigenvector of the principal component and the standardized values of each original variable, and the comprehensive score is the weighted average of the scores of each principal component. The calculation formula is:

[0067]

[0068] In the formula, A j is the score of the j-th principal component; Z k is the standardized value of the k-th water quality measurement index; w jk is the loading coefficient of the j-th principal component on the k-th water quality index.

[0069] (5) Calculate the absolute principal component score: The principal component score needs to be converted into a non-standard absolute principal component score (APCS) to be used for the contribution analysis of PCs to pollution indicators. The absolute principal component score can reflect the contribution of pollution sources to water quality indicators. The calculation formula of the absolute principal component score is:

[0070] APCS jk =(A j ) k -(A 0 ) j

[0071]

[0072]

[0073] In the formula: APCS jk is the absolute principal component score value of the j-th principal component; (A j ) k is the score value of the j-th principal component; (A 0 ) j is the score value of the j-th principal component under the value of 0; S jk is the factor score coefficient of the j-th principal component on the water quality factor k, where i is the serial number of the hydrochemical factor; (Z 0 ) k is the standardized value when the pollutant concentration at the observation point is set to 0; is the arithmetic mean of the concentration of the k-th water quality factor, mg / L; σ k is the standard deviation of the concentration of the k water quality factor, mg / L.

[0074] (6) Based on the APCS scores and the measured water quality concentrations, use the multiple linear regression method to establish a multiple linear regression equation for the contribution degree of pollution sources.

[0075] On the basis of absolute principal component analysis, taking the measured water quality concentration C as the dependent variable and APCS as the independent variable, establish a multiple linear regression equation:

[0076]

[0077] In the formula, C k is the measured concentration of the k water chemical factor, mg / L; a jk is the regression coefficient of pollution source j to water chemical factor k; APCS jk is the absolute principal component score value of all samples of water chemical factor k; b k is the constant term.

[0078] Step 2: Use the measured data of pollution sources of runoff input, atmospheric deposition and water-soil interface release to verify the absolute principal component-multiple linear regression model to determine the pollution source parts corresponding to different principal components of the absolute principal component-multiple linear regression model.

[0079] By ranking the contribution of the measured data of runoff input, atmospheric deposition and water-soil interface to the reservoir water quality, and comparing it with the ranking of the contribution of the three extracted principal components, determine the pollution sources corresponding to different principal components.

[0080] Analyze the contribution degree of pollution sources. In the APCS-MLR model, a jk ×APCS jk is the contribution of pollution source m to the concentration of water quality indicators. The pollution contribution degree PC of each principal component my The calculation formula is:

[0081]

[0082] The calculation formula for the contribution of other pollution sources is:

[0083]

[0084] Step 3: Based on the reservoir water quality monitoring data, use the absolute principal component-multiple linear regression model to calculate and simulate the contribution degrees of pollution sources of runoff input, atmospheric deposition and water-soil interface input at different times and spaces in the reservoir.

[0085] Step 4: Calculate the atmospheric deposition and the pollution sources input at the water-soil interface by using the obtained contribution degree, runoff input pollution sources and the absolute principal component-multivariate linear regression model, so as to identify the spatio-temporal variation process of the multi-media pollution sources.

[0086] Classify the pollution sources of the reservoir into runoff input, atmospheric deposition, and pollution sources transported at the water-soil interface. The output of runoff input pollutants is easy to measure and calculate, and it is used as a known pollution source. The input amount of the runoff input pollution source is:

[0087]

[0088] In the formula, C i is the pollutant concentration monitored in the i-th time period, mg / L; Q i is the average flow rate in the i-th time period, m 3 / s; Δt is the time period length represented by the i-th monitoring, s.

[0089] In the simulation of the pollution source change process, the known pollution source is the runoff input pollution source. Input it into the APCS-MLR model, and then simulate and predict the spatio-temporal variation amounts of atmospheric deposition, water-soil interface input and other pollution sources. The calculation formula for the pollution input to be solved is:

[0090]

[0091] Among them, C w is the input amount of atmospheric deposition or water-soil interface input or other pollution sources, PC w is the contribution of atmospheric deposition or water-soil interface input or other pollution sources, PC y is the contribution of the runoff input pollution source, C y is the input amount of the known pollution source.

[0092] The following further elaborates on the present invention in conjunction with specific embodiments.

[0093] The present invention takes the water quality index data and total nitrogen pollution source data of the Danjiangkou Reservoir in the Hanjiang River Basin of China as the research object, and the monitoring points cover the scope of the Danjiangkou Reservoir ( Figure 2 ). According to the nitrogen cycle process and multi-media pollution source input process in the reservoir, select water temperature (WT), pH value (pH), dissolved oxygen (DO), permanganate index (COD Mn ), chemical oxygen demand (COD), biochemical oxygen demand (BOD), ammonia nitrogen (NH 4+ -N), total phosphorus (TP), total nitrogen (TN), sulfate (SO 4 2- ), nitrate (NO 3--N), transparency (SD), and chlorophyll a (Chl-a) as the principal component analysis indicators.

[0094] (1) Model construction

[0095] Principal component analysis was performed on the annual average values of water quality indicators at each station. The results of KMO (0.65) and Bartlett's test (p < 0.001) confirmed that the dataset was suitable for PCA analysis. Components with eigenvalues greater than 1 were selected to determine the principal components (PCs). After rotating the factor loading matrix using the varimax method, PC1, PC2, and PC3 explained 64.08%, 15.92%, and 8% of the total variance, respectively. The cumulative variance percentage of the three PCs was 82.90%, indicating that these PCs could sufficiently explain most of the information in the original dataset.

[0096] Based on the principal component analysis, an absolute principal component-multiple linear regression model was constructed. The regression equation for the total nitrogen contribution in the entire reservoir area was:

[0097] C TN = 0.029×APCS1 - 0.0052×APCS2 + 0.0082×APCS2 + 1.6 (P < 0.05)

[0098] The regression coefficient R of the equation 2 = 0.865, indicating that the model had a good fit.

[0099] (2) Model verification

[0100] Calculating the contributions of each pollution source, it was found that the APCS1 source contributed 16.83% of the total nitrogen input, the APCS2 source contributed 0.24% of the total nitrogen input, the APCS3 source contributed only 0.03% of the total nitrogen input, and the remaining (R) source contributed 82.9% of the total nitrogen input.

[0101] Refer to Figure 3 , using the measured data of existing runoff input, atmospheric deposition, and water-soil interface input pollution sources, the estimated average annual runoff input was 48678.99 t, the atmospheric deposition was about 326.92 t, and the water-soil interface flux was about 10110 t. By comparing the calculated contribution rates of pollution sources with those calculated by the APCS-MLR model, it could be inferred that APCS1 represented the water-soil interface input pollution source, APCS2 represented the atmospheric deposition pollution source, APCS3 represented other pollution sources, and R represented the runoff input.

[0102] (3) Model simulation and analysis

[0103] The PCA-PACS-MLR model was constructed for the station-by-station and month-by-month water quality data to analyze the spatio-temporal evolution of nitrogen pollution sources in the reservoir.

[0104] The station-by-station PCA-PACS-MLR model calculates the APCS scores of different pollution sources at each station, and uses Arcgis to perform spline interpolation of the scores of each station in the reservoir area ( Figure 4 ). For APCS1, the scores around the reservoir are higher than those in the middle of the reservoir, and the scores in the bays are more significant. For APCS2, the Danjiang Reservoir is higher than the Hanjiang Reservoir in the whole reservoir area. For APCS3, there is little difference in the whole reservoir area, and the scores in the middle of the reservoir are slightly higher than those around the reservoir. The R score shows that the Hanjiang Reservoir is higher than the Danjiang Reservoir.

[0105] Please refer to Figure 5 , the monthly APCS-MLR model, to analyze the monthly changes in the total nitrogen contribution. From the monthly changes in the total nitrogen contribution rate, runoff input accounts for the largest proportion of the overall input for most of the time, followed by the input at the water-soil interface. Atmospheric deposition is not currently the main pollution source of the Danjiangkou Reservoir.

[0106] When one of the pollution sources is known, the constructed APCS-MLR model can be used to deduce the situation of other pollution sources. Taking the pollution source situation of the water-soil interface input as an example, using the data of the runoff input pollution source, the pollution source quantity of the water-soil interface input from 2018 to 2021 was simulated and calculated.

[0107] The present invention also provides a reservoir multi-media pollution source identification system based on machine learning. The system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it executes the above-mentioned reservoir multi-media pollution source identification method based on machine learning.

[0108] The present invention also provides a computer-readable storage medium. The computer-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by the processor, the machine-executable instructions cause the processor to implement the above-mentioned reservoir multi-media pollution source identification method based on machine learning.

[0109] It is easy for those skilled in the art to understand that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A reservoir multi-media pollution source identification method based on machine learning, characterized in that: The method comprises the following steps: (1) Based on the reservoir water quality monitoring data, the principal component analysis is combined with the multivariate linear regression to construct an absolute principal component-multivariate linear regression model. The mathematical expression of the absolute principal component-multivariate linear regression model is: In the formula, C k is the measured concentration of water chemical factor k, mg / L; a jk is the regression coefficient of pollution source j on water chemical factor k; APCS jk is the absolute principal component score of all samples of water chemical factor k; b k is a constant term; (2) The absolute principal component-multivariate linear regression model was validated using measured data on pollution sources of runoff input, atmospheric deposition, and release at the water-soil interface to determine the pollution source parts corresponding to the different principal components of the absolute principal component-multivariate linear regression model; (3) Based on the reservoir water quality monitoring data, the absolute principal component-multivariate linear regression model is used to calculate and simulate the contribution of runoff input, atmospheric deposition and water-soil interface input pollution sources at different times and spaces in the reservoir; (4) The obtained contribution, runoff input pollution sources and absolute principal component-multivariate linear regression model are used to calculate atmospheric deposition and water-soil interface input pollution sources, thereby identifying the spatiotemporal variation process of multi-media pollution sources.

2. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 1, characterized in that: The water quality monitoring data were used to construct a water quality dataset. The KMO sampling suitability test and Bartlett sphericity test were used to evaluate the applicability of the water quality dataset. The water quality indicators of reservoir pollution were screened to construct a training dataset.

3. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 2, characterized in that: The Z-Score method is used to standardize the water quality training data, and the corresponding calculation formula is: Where z is the standardized value and x is the original value. is the sample mean, is the sample mean square error.

4. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 3, characterized in that: The covariance matrix between water quality indicators was calculated to obtain multiple groups of eigenvectors and eigenvalues, and the first three components with eigenvalues ​​greater than or equal to 1 were selected as principal components. The scores of each principal component were calculated based on the eigenvectors of the principal components and the standardized values ​​of the water quality indicators.

5. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 4, characterized in that: The covariance matrix is ​​constructed by calculating the covariance between indicators, and the obtained eigenvalues ​​are arranged from large to small as λ1≥λ2≥…≥λ p ≥0, the corresponding eigenvectors are e1, e2, …, e p The contribution rate of each eigenvalue is The cumulative contribution rate is 6. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 5, characterized in that: The principal component score is a linear combination of the principal component eigenvector and the standardized values ​​of each original variable, while the comprehensive score is the weighted mean of the principal component scores, calculated as: In the formula, A j is the score of the jth principal component; Z k is the standardized value of the kth water quality indicator; w jk is the loading coefficient of the jth principal component on the kth water quality index; The principal component scores must be converted into non-standard absolute principal component scores to be used for contribution analysis of pollution indicators. The absolute principal component scores can reflect the contribution of pollution sources to water quality indicators. The calculation formula for the absolute principal component scores is: APCS jk =(A j ) k -(A0) j Where: APCS jk is the absolute principal component score of the jth principal component; (A j ) k is the score of the jth principal component; (A0) j is the jth principal component score under the value of 0; S jk is the factor score coefficient of the jth principal component on the water quality factor k, where i is the serial number of the water chemistry factor; (Z0) k It is the standardized value when the pollutant concentration at the observation point is set to 0; is the arithmetic mean of the concentration of the kth water quality factor, mg / L; σ k is the standard deviation of the concentration of water quality factor k, mg / L.

7. The method for identifying reservoir multi-media pollution sources based on machine learning according to any one of claims 1 to 6, characterized in that: Absolute principal component-multiple linear regression model a jk ×APCS jk is the contribution of pollution source m to the concentration of water quality indicators, and the pollution contribution of each principal component PC my The calculation formula is: The calculation formula for the contribution of other pollution sources is:

8. The method for identifying reservoir multi-media pollution sources based on machine learning as claimed in claim 6, characterized in that: The input amount of runoff input pollution source is: In the formula, C i is the pollutant concentration monitored in the i-th time period, mg / L; Q i is the average flow rate in the i-th time period, m 3 / s; Δt is the length of the time period represented by the i-th monitoring, s; In the simulation of the pollution source change process, the known pollution source is the runoff input pollution source, which is input into the model to simulate and predict the temporal and spatial changes of atmospheric deposition, water-soil interface input and other pollution sources. The calculation formula of the pollution input to be solved is: Among them, C w is atmospheric deposition or water-soil interface input or other pollution source input, PC w For contributions from atmospheric deposition or soil-water interface input or other pollution sources, PC y Contribution of runoff input pollution sources, C y Enter quantities for known pollution sources.

9. A reservoir multi-media pollution source identification system based on machine learning, characterized by: The system includes a memory and a processor, the memory stores a computer program, and the processor executes the method for identifying reservoir multi-media pollution sources based on machine learning as described in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by the processor, the machine-executable instructions prompt the processor to implement the method for identifying multi-media pollution sources of reservoirs based on machine learning as described in any one of claims 1-8.

Images