PM2.5 heavy metal emission reduction health benefit evaluation system and method based on combination of machine learning and receptor model
Through machine learning and receptor models, heavy metal monitoring data are integrated and meteorological impact is stripped away, and the contribution rate of pollution sources is quantified, and the accuracy of air quality model to evaluate the benefits of heavy metal emission reduction is solved, and timely and accurate health benefits assessment is achieved, supporting coordinated assessment of multiple pollutants and policy optimization.
Patent Information
- Application Number
- CN202510603380.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-12
- Publication Date
- 2025-08-26
AI Technical Summary
The existing air quality models and receptor models cannot accurately and timely evaluate the health benefits of toxic and harmful heavy metal emission reduction, and it is difficult to achieve long-term effect evaluation due to computing power limitations, and cannot meet the deep emission reduction needs for toxic and harmful heavy metals.
Using the method of combining machine learning with receptor models, the data integration module, meteorological normalization module, dynamic source analysis module and health benefit mapping module are integrated online heavy metal monitoring instrument data, stripped off meteorological impact, quantified the contribution rate of pollution sources, and evaluated health benefits through the BenMAP model.
It has achieved a timely and accurate assessment of toxic and harmful heavy metal emission reduction, reduced the evaluation error to ±10%, supported the coordinated assessment of multiple pollutants, and improved the scientificity and economic benefits of policy formulation.
Smart Images
Figure CN120543347A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of environmental monitoring and public health management, and specifically relates to a PM based on machine learning and receptor model. 2.5 Heavy metal emission reduction health benefit assessment system and method. Background Art
[0002] Atmospheric fine particulate matter PM 2.5 The harm to public health has attracted widespread attention; toxic and harmful heavy metals (Al, V, Cr, Mn, Ni, Cu, Zn, As, Cd, Ba, Pb, Sb, etc.) are PM 2.5 As, Cr, Ni, Zn, Cu and Cd are important components in the human body and have significant biological toxicity, which can cause damage to the normal physiological metabolism and important organs of the human body. For example, As, Cr, Ni, Zn, Cu and Cd are carcinogenic, and Pb can cause fetal malformations in pregnant women.
[0003] With the control of air pollution, PM 2.5 The concentration has improved significantly, but the toxic and harmful heavy metals and PM 2.5 The concentration decrease shows a nonlinear relationship. Taking a provincial capital city as an example, compared with 2010, the concentration decrease rate of toxic and harmful heavy metals in 2018 was between As (9%) and Cd (87%), while the concentration of Al increased by nearly 2 times; compared with 2017, the concentrations of Cr, Ni, Zn, Cd and Pb increased by 8%-91% in 2018. It can be seen that toxic and harmful heavy metals still pose significant health risks; therefore, it is imperative to strengthen the deep reduction of toxic and harmful heavy metals.
[0004] The social costs of air pollution are one of the most important obstacles to economic development. Accurately assessing the health benefits of reducing toxic and harmful heavy metal emissions is crucial to formulating and implementing effective pollution control policies. Numerical air quality models are a common method for analyzing air pollution sources and evaluating emission reduction effects. However, due to the lack of heavy metal emission inventories and source spectra, air quality model simulation results are poor and have serious lags.
[0005] In addition, due to the limitation of computing power, air quality models are difficult to achieve long-term effect evaluation; with the gradual popularization of online heavy metal monitoring instruments, a large amount of heavy metal observation data has been accumulated; receptor source analysis models such as the chemical mass balance method (CMB) and the positive factor matrix method (PMF) can be used for source analysis of heavy metal observation data; however, receptor models cannot analyze the influence of meteorological, transport and chemical processes, so it is difficult to accurately and quantitatively evaluate the implementation effect of emission reduction measures; and conventional methods such as air quality models and receptor models cannot meet the needs of timely, accurate and effective evaluation of the health benefits of toxic and harmful heavy metal emission reduction. Summary of the Invention
[0006] In order to solve the above technical problems, the present invention provides a PM based on machine learning and receptor model. 2.5 Heavy metal emission reduction health benefit assessment system and method.
[0007] Based on the above objectives, the present invention is achieved through the following technical solutions:
[0008] The present invention provides a PM based on machine learning and receptor model 2.5 The heavy metal emission reduction health benefit assessment system includes a data integration module for generating a time series data set, a meteorological normalization module for generating a concentration component dominated by anthropogenic emissions, a dynamic source resolution module for quantifying the dynamic impact of specific measures on source contribution rates, and a health benefit mapping module for generating source-specific risk maps; the data integration module is sequentially connected to the meteorological normalization module, the dynamic source resolution module, and the health benefit mapping module.
[0009] Preferably, the data integration module is used to integrate the PM data obtained by the online heavy metal monitoring instrument. 2.5 Chemical composition, meteorological parameters and pollution source activity data are collected, and a time series data set is generated through data cleaning and standardization; the meteorological normalization module adopts a random forest model, with heavy metal concentration, meteorological factors and time variables as input, to train a meteorological impact stripping model to generate a heavy metal concentration component dominated by anthropogenic emissions.
[0010] Preferably, the dynamic source analysis module inputs normalized data into the PMF receptor model to analyze the contribution rate of pollution sources such as industrial processes, coal burning, motor vehicles, and dust to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rate of heavy metal sources; the health benefit mapping module converts the reduction in heavy metal concentrations from different sources into avoidable premature deaths, reduced respiratory diseases, and medical cost savings through the BenMAP model, and generates a source-specific risk map.
[0011] Preferably, the data integration module acquires hourly concentration data of more than 30 heavy metals such as As, Cr, and Pb monitored by online heavy metal monitors in real time; the meteorological normalization module integrates multiple machine learning models to ensure that the influence of meteorology on the observed metal concentration is removed.
[0012] Preferably, the dynamic source apportionment module ensures that PM 2.5 The health benefit mapping module supports source-specific risk assessment and generates a contribution map of industrial, transportation and other sources to the number of premature deaths, reductions in respiratory diseases and medical cost savings.
[0013] A PM based on the combination of machine learning and receptor model2.5 The method of the health benefit assessment system of heavy metal emission reduction includes the following steps:
[0014] S1. The data integration module obtains the observation data of the online heavy metal monitoring instrument.
[0015] S2, meteorological normalization module uses random forest model to analyze PM 2.5 The concentrations of heavy metals were normalized to meteorological values to separate the anthropogenic emission components.
[0016] S3, the dynamic source analysis module uses the PMF receptor model to analyze the contribution rates of various pollution sources to identify effective emission reduction measures.
[0017] S4. The health benefit mapping module uses the BenMAP model to quantify the health benefits corresponding to the reduction in heavy metal concentrations of different pollution sources and generate source-specific risk maps.
[0018] Preferably, in step S1, the data integration module uses an online heavy metal monitoring instrument to obtain PM 2.5 Chemical composition, meteorological parameters and pollution source activity data; online heavy metal monitoring instrument data including PM 2.5 concentration, PM 2.5 The concentrations of water-soluble ions, metals (including toxic and harmful heavy metals) and carbon components, meteorological data, etc.
[0019] Preferably, in step S2, the meteorological normalization module adopts a random forest model, with heavy metal concentration, meteorological factors and time variables as input, and the output variable is PM 2.5 and the anthropogenic emission constraint concentrations of each heavy metal component, a meteorological impact stripping model was constructed to generate the heavy metal concentration components dominated by anthropogenic emissions.
[0020] Preferably, in step S3, the PMF receptor model analyzes the contribution rate of various pollution sources (industrial processes, coal burning, motor vehicles, dust, etc.) to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rate of heavy metal sources.
[0021] Preferably, in step S4, the BenMAP model converts the reduction in heavy metal concentrations from different sources into avoidable premature deaths, reduced respiratory diseases and medical cost savings, and generates a source-specific risk map.
[0022] Compared with the prior art, the present invention has the following beneficial effects:
[0023] 1. This invention combines the random forest model with the PMF receptor model for the first time, and eliminates the interference of meteorological conditions such as wind speed and humidity on source analysis through meteorological normalization, so that the error of heavy metal pollution contribution assessment is reduced to ±10%, which can meet the needs of timely, accurate and effective assessment of the health benefits of toxic and harmful heavy metal emission reduction; This invention adopts the "emission reduction measures-metal source contribution-PM 2.5 The "reduction-health benefit" dynamic correlation model breaks through the static limitations of the traditional threshold method, supports the coordinated assessment of multiple pollutants such as heavy metals, polycyclic aromatic hydrocarbons, and VOCs, and can be expanded to be applied to the optimization of emission reduction paths for joint prevention and control of pollution in urban agglomerations.
[0024] 2. This paper addresses the core issues in current heavy metal pollution control, such as the reliance of air quality models on emission inventories, the inability of receptor models to separate meteorological interference, and the disconnection between health benefit assessment and policy. It proposes a data-driven dynamic quantification method that integrates observational data, machine learning, source apportionment, and health risk models to achieve accurate assessment of the effects of heavy metal emission reduction measures and their health benefits, especially for PM2.5. 2.5 Pollution source tracing and policy optimization for toxic components such as medium and heavy metals and polycyclic aromatic hydrocarbons.
[0025] 3. This invention is not only conducive to promoting the formulation of toxic and harmful heavy metal prevention and control policies, but also has significant social and economic benefits. 2.5 Health benefit assessment of species such as polycyclic aromatic hydrocarbons and toxic and harmful VOCs in the atmosphere, effectively improving the analytical capabilities of researchers and the utilization of data BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 This is a schematic diagram of the technical solution route of the present invention;
[0027] Figure 2 Schematic diagram of the system framework of the present invention;
[0028] Figure 3 It is a schematic diagram of the application of the present invention;
[0029] Figure 4 For PM 2.5 Schematic diagram of the PMF factor spectrum of the elements in;
[0030] Figure 5 This is a time series diagram of the concentration of metals contributed by process sources after meteorological standardization;
[0031] Figure 6 Standardized health risk diagrams for observations and meteorology. DETAILED DESCRIPTION
[0032] The present invention is further described in detail below through specific examples, but the scope of the present invention is not limited thereto.
[0033] Example 1
[0034] A PM based on the combination of machine learning and receptor model 2.5 Heavy metal emission reduction health benefit assessment system, such as Figure 1 As shown, it includes a data integration module for generating time series data sets, a meteorological normalization module for generating concentration components dominated by anthropogenic emissions, a dynamic source resolution module for quantifying the dynamic impact of specific measures on source contribution rates, and a health benefit mapping module for generating source-specific risk maps; the data integration module is sequentially connected to the meteorological normalization module, the dynamic source resolution module, and the health benefit mapping module.
[0035] The data integration module is used to integrate the PM data obtained by the online heavy metal monitoring instrument. 2.5 Chemical composition, meteorological parameters and pollution source activity data are collected, and a time series data set is generated through data cleaning and standardization; the meteorological normalization module adopts a random forest model, with heavy metal concentration, meteorological factors and time variables as input, to train a meteorological impact stripping model to generate a heavy metal concentration component dominated by anthropogenic emissions.
[0036] The dynamic source analysis module inputs normalized data into the PMF receptor model to analyze the contribution rate of pollution sources such as industrial processes, coal burning, motor vehicles, and dust to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rate of heavy metal sources; the health benefit mapping module uses the BenMAP model to convert the reduction in heavy metal concentrations from different sources into the number of avoidable premature deaths, reductions in respiratory diseases, and medical cost savings, and generates a source-specific risk map.
[0037] The data integration module acquires hourly concentration data of more than 30 heavy metals such as As, Cr, and Pb monitored by online heavy metal monitors in real time; the meteorological normalization module integrates multiple machine learning models to ensure that the influence of meteorology on the observed metal concentrations is removed.
[0038] The dynamic source apportionment module constrains the factor spectrum by the relative contribution rate of heavy metals to ensure PM 2.5 The health benefit mapping module supports source-specific risk assessment and generates a contribution map of industrial, transportation and other sources to the number of premature deaths, reductions in respiratory diseases and medical cost savings.
[0039] A PM based on the combination of machine learning and receptor model 2.5 Methods for evaluating the health benefits of heavy metal emission reduction systems, such as Figure 2 As shown, the following steps are included:
[0040] S1, the data integration module obtains the observation data of the online heavy metal monitoring instrument. 2.5 Chemical composition, meteorological parameters and pollution source activity data; online heavy metal monitoring instrument data including PM 2.5 concentration, PM 2.5 The concentrations of water-soluble ions, metals (including toxic and harmful heavy metals) and carbon components, meteorological data, etc.
[0041] Specifically, the data integration module performs data acquisition and preprocessing (data input layer); through the online equipment network communication protocol (such as Xact 625 heavy metal online monitor, Thermo 1400a PM 2.5 Analyzer) real-time collection of PM 2.5 The system collects concentration and component data (more than 30 heavy metals such as Al, As, Cr, water-soluble ions, and carbon components), and accesses real-time data from meteorological stations (wind speed, wind direction, temperature, humidity, and air pressure) and historical meteorological databases. It also integrates external data, such as the public emission inventory (MEIC) and pollution source activity data (industrial electricity consumption and traffic flow). The collected data are cleaned and standardized, and the time series interpolation method is used to fill missing values. The box plot method is used to remove outliers, and the Z-score standardization method is used to unify the dimensions of each variable to ensure that the data quality meets the needs of subsequent analysis.
[0042] S2, meteorological normalization module uses random forest model to analyze PM 2.5 The concentrations of heavy metals were normalized to meteorological values to separate the anthropogenic emission components.
[0043] The meteorological normalization module adopts a random forest model, with heavy metal concentration, meteorological factors and time variables as input, and the output variable is PM 2.5 and the anthropogenic emission constraint concentrations of each heavy metal component, a meteorological impact stripping model was constructed to generate the heavy metal concentration components dominated by anthropogenic emissions.
[0044] Specifically, the meteorological normalization module (machine learning engine) uses the random forest model (rmweather package in R language), whose input variables include meteorological parameters (wind speed, temperature, humidity), time variables (year, month, day, week, hour) and holiday labels, and the output variable is PM 2.5 and the anthropogenic emission constraint concentrations of each heavy metal component; the model's hyperparameters were set to number of trees = 500, maximum depth = 10, and the contribution of meteorological factors was analyzed using SHAP values; K = 5 cross-validation was used to evaluate the model accuracy, and finally the meteorologically normalized heavy metal concentration time series was output.
[0045] At the same time, the meteorological normalization module adopts the RF model (random forest model) based on decision tree with high accuracy and ability to calculate variable importance as a specific modeling method; compared with a single decision tree, the RF model introduces randomness and uses error evaluation to reduce bias, which can ensure more stable analysis results and avoid overfitting; PM 2.5 The mass concentration of metals in specific sources of metals was used as the dependent variable, and meteorological parameters and time variables were used as independent variables; the original data set was randomly classified, including a training data set (80% of the original data set) used to develop the RF model, and the remaining original data set was used as a test data set to evaluate the model performance; after the RF model was established, the randomly selected meteorological conditions were used as input for predicting meteorologically standardized species; the prediction process was repeated 800 times at each node and then averaged to determine the meteorologically standardized concentration of each element in each source at a specific time point predicted using the meteorological standardization method, represented by MN-; the difference between the observed pollutant concentration and the pollutant concentration after meteorological standardization was regarded as the concentration of meteorological contribution.
[0046] S3, the dynamic source analysis module uses the PMF receptor model to analyze the contribution rates of various pollution sources to identify effective emission reduction measures.
[0047] The PMF receptor model analyzes the contribution rate of various pollution sources (industrial processes, coal burning, motor vehicles, dust, etc.) to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rate of heavy metal sources.
[0048] Specifically, the dynamic source resolution module (PMF calculation engine) converts the normalized PM 2.5 The component data were input into the PMF receptor model, the number of pollution source categories was set (6-8 categories), and the source characteristic spectrum constraints were set according to the literature (such as coal burning sources were marked with As and Se, and traffic sources were marked with Ba and Sb); the uncertainty of the analysis was evaluated by the Bootstrap method (1000 times), and the contribution rate of each pollution source was output (such as the proportion of industrial sources in PM 2.5 The model supports a dynamic update mechanism and automatically retrains every quarter to adapt to changes in emission sources (such as new factories and traffic restriction policies) to ensure the timeliness of source analysis results.
[0049] Heavy metal source contribution constraint analysis (combined iterative module); PM 2.5 The contribution rate of heavy metals in the factor spectrum is used as a constraint condition, and the PMF receptor model is re-run to perform heavy metal-specific analysis; only heavy metal concentration data is input to generate a heavy metal-specific source spectrum, and the hourly contribution of each pollution source to heavy metals is output; it can strengthen the relationship between heavy metals and PM 2.5The correlation of emission sources improves the accuracy of source analysis.
[0050] At the same time, the dynamic source analysis module analyzes the contribution rate and time series of pollution sources such as industry, transportation, and coal burning to heavy metals by inputting normalized data into the PMF receptor model; the PMF receptor model is a receptor-based source allocation model, and its principle is to first calculate the errors of each chemical component in the particulate matter with the help of weights, and then calculate the pollution source and contribution rate of the particulate matter with the help of the least squares method.
[0051] Assume an n×m matrix X, where the number of samples is n and the number of chemical components is m (such as various ions, ECs, and metals). Then X=GF+E is the decomposition of the matrix X, where the n×p matrix is defined as G, the p×m matrix is F, the number of major pollution sources is represented by p, and the residual matrix is E, which can be expressed as:
[0052]
[0053] Among them, X ij is the concentration of substance j in sample i; g ik is the contribution of source k to the receptor in sample i; f kj is the factor distribution of species j in source k; e ij is the residual of the jth chemical component in the i-th sample; X ij The uncertainty is s ij , represents the uncertainty of the jth chemical component in the ith sample; Q is used to characterize the size of the overall fitting residual after uncertainty conversion; the error evaluation analysis of the factor results was performed using the Bootstrap (BS), Displacement (DISP) and BS-DISP methods.
[0054] In addition, by fully investigating the policy implementation nodes, analyzing the control measures, identifying effective emission reduction measures and calculating the reduction in heavy metal concentrations during the same period.
[0055] S4. The Health Benefit Mapping module uses the BenMAP model to quantify the health benefits associated with reductions in heavy metal concentrations at different pollution sources, generating source-specific risk maps. The BenMAP model converts reductions in heavy metal concentrations from different sources into avoided premature deaths, reduced respiratory illnesses, and healthcare cost savings, generating source-specific risk maps.
[0056] Specifically, the health benefit mapping module (BenMAP interactive layer) calculates non-carcinogenic risks and carcinogenic risks based on US EPA and Chinese population parameters; sets the baseline scenario (no emission reduction) and control scenario in BenMAP, outputs health benefit data, including the number of premature deaths avoided, medical cost savings, and generates a spatial distribution map; the system provides a "policy simulator" function that supports users to customize emission reduction intensity and predict health benefit changes in real time, and finally outputs a standardized report, including a source analysis map, a risk heat map, and a cost-benefit analysis table.
[0057] At the same time, the health benefit mapping module calculates the carcinogenic risk (CR) and non-carcinogenic risk (HI) based on the exposure response model recommended by the US EPA and combined with Chinese population parameters (such as respiratory rate and life expectancy); uses the BenMAP model to quantify the health benefits corresponding to emission reduction measures based on the exposure response parameters of the Chinese population; input parameters include effective emission reduction measures identified based on S3, extraction of PM2.5 levels from the same source during the same period, and the use of the BenMAP model to quantify the health benefits corresponding to emission reduction measures. 2.5 Concentration reduction and health parameters, such as population data, exposure-response relationship, economic value, etc.; calculate health benefits and obtain health outcomes and economic value conversion.
[0058] Example 2
[0059] The method of Example 1 was used to analyze the PM 2.5 Analyze the long-term evolution of metal element sources and health risks, such as Figure 3-6 The specific process is as shown in Figure 3 As shown, including:
[0060] S1. Collection and preprocessing of multi-source data.
[0061] Collect PM data from Zhengzhou City from 2018 to 2021 2.5 Component data (including more than 30 heavy metals such as Al, Cr, and As), meteorological data (wind speed, temperature, humidity, and precipitation), and pollution source activity data (industrial energy consumption and traffic flow); before data analysis, outliers (such as data during sandstorms) were first removed, missing values were filled using time series interpolation, and the dimensions of each variable were standardized.
[0062] S2, random forest model for meteorological normalization.
[0063] The random forest model was constructed using the R language rmweather package. The input variables of the model included time variables and meteorological variables: time variables included year, month, week, and hour; meteorological variables included wind speed, temperature, humidity, precipitation, and air pressure; the output of the random forest model was a meteorologically normalized heavy metal concentration sequence (MN concentration); 80% of the data was divided into a training set and 20% into a test set to verify the model. 2It is 0.6-0.8, and the result is credible.
[0064] S3. Origin analysis of the PMF receptor model.
[0065] The MN concentration data were input into the PMF receptor model to analyze the pollution sources of heavy metals. Through residual analysis (Q / Qexp=0.95) and source spectrum matching, four types of pollution sources were identified, such as Figure 4 As shown in the figure, the metal elements in Zhengzhou City come from process sources, with the factor characteristics of Pb (87%), Cu (72%), Mn (47%) and Zn (37%); motor vehicle non-exhaust sources, with the factor characteristics of Cr (75%), Ni (59%), Zn (39%), Mn (36%) and Fe (26%); coal + biomass combustion sources, with the factor characteristics of As (67%), Se (60%) and K (34%); dust sources, with the factor characteristics of Si (82%), V (77%), Ca (69%), Al (43%), Ba (43%) and Fe (38%).
[0066] S4. Mapping of emission reduction policies.
[0067] like Figure 5 The figure shows the actual benefits of process source emission reduction policies in reducing metal pollution source concentrations, evaluated in combination with pollution reduction policies. Since 2018, the implementation of a series of industrial source control policies in Zhengzhou, such as end-of-pipe disposal, ultra-low emissions, and capacity adjustment, has led to the annual average concentration of metal elements from process sources decreasing from 1.47 μg / m in 2018. 3 Down to 0.69 μg / m2 in 2021 3 , a decrease of 53%.
[0068] S5. Quantification of health risks.
[0069] like Figure 6 The figure shows a comparison of the metal health risks of specific pollution sources before and after meteorological normalization; the health risk assessment of pollution sources after meteorological standardization shows that the non-carcinogenic risks mainly come from process sources (adults: HI = 0.5; children: 0.7) and motor vehicle non-exhaust sources (adults: 0.4; children: 0.6), and the carcinogenic risks mainly come from coal + biomass combustion sources (1.36×10-5); from 2018 to 2020, the reduction in health risks from process sources and coal + biomass combustion sources dominated the decline in health risks of metal pollution sources after meteorological standardization, while the non-carcinogenic risks rebounded in 2021, mainly due to a significant increase in the non-carcinogenic risks from motor vehicle non-exhaust sources (adults: 0.11; children: 0.15); in addition, the contribution of motor vehicle non-exhaust sources to non-carcinogenic risks showed an increasing trend, increasing by 6%; therefore, in the future, attention should be paid to the health risks brought by motor vehicle non-exhaust emissions.
[0070] The above embodiments are specific implementation methods of the present invention, but the implementation methods of the present invention are not limited to the above embodiments. Any other combination, change, modification, substitution, and simplification that does not exceed the design concept of the present invention shall fall within the scope of protection of the present invention.
Claims
1. A PM based on the combination of machine learning and receptor model 2.5 The heavy metal emission reduction health benefit assessment system is characterized by: It includes a data integration module for generating time series data sets, a meteorological normalization module for generating concentration components dominated by anthropogenic emissions, a dynamic source resolution module for quantifying the dynamic impact of specific measures on source contribution rates, and a health benefit mapping module for generating source-specific risk maps; the data integration module is sequentially connected to the meteorological normalization module, the dynamic source resolution module, and the health benefit mapping module.
2. The PM based on machine learning combined with receptor model according to claim 1 2.5 The heavy metal emission reduction health benefit assessment system is characterized by: The data integration module is used to integrate the PM data obtained by the online heavy metal monitoring instrument. 2.5 Chemical composition, meteorological parameters and pollution source activity data are collected, and a time series data set is generated through data cleaning and standardization; the meteorological normalization module adopts a random forest model, with heavy metal concentration, meteorological factors and time variables as input, to train a meteorological impact stripping model to generate a heavy metal concentration component dominated by anthropogenic emissions.
3. The PM based on machine learning combined with receptor model according to claim 1 2.5 The heavy metal emission reduction health benefit assessment system is characterized by: The dynamic source analysis module inputs normalized data into the PMF receptor model to analyze the contribution rate of pollution sources such as industrial processes, coal burning, motor vehicles, and dust to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rate of heavy metal sources; the health benefit mapping module uses the BenMAP model to convert the reduction in heavy metal concentrations from different sources into the number of avoidable premature deaths, reductions in respiratory diseases, and medical cost savings, and generates a source-specific risk map.
4. The PM based on machine learning combined with receptor model according to claim 1 2.5 The heavy metal emission reduction health benefit assessment system is characterized by: The data integration module acquires hourly concentration data of various heavy metals monitored by the online heavy metal monitor in real time; the meteorological normalization module integrates multiple machine learning models to ensure that the influence of meteorology on the observed metal concentration is removed.
5. The PM based on machine learning combined with receptor model according to claim 1 2.5 The heavy metal emission reduction health benefit assessment system is characterized by: The dynamic source apportionment module constrains the factor spectrum by the relative contribution rate of heavy metals to ensure PM 2.5 The health benefit mapping module supports source-specific risk assessment and generates a contribution map of industrial, transportation and other sources to the number of premature deaths, reductions in respiratory diseases and medical cost savings.
6. The PM based on machine learning combined with receptor model according to any one of claims 1 to 5 2.5 The method of the heavy metal emission reduction health benefit assessment system is characterized in that: The following steps are involved: S1, data integration module obtains observation data of online heavy metal monitoring instruments; S2, meteorological normalization module uses random forest model to analyze PM 2.5 and heavy metal concentrations were normalized to meteorological values to separate anthropogenic emissions; S3, the dynamic source apportionment module uses the PMF receptor model to analyze the contribution rates of various pollution sources to identify effective emission reduction measures; S4. The health benefit mapping module uses the BenMAP model to quantify the health benefits corresponding to the reduction in heavy metal concentrations of different pollution sources and generate source-specific risk maps.
7. The PM based on machine learning combined with receptor model according to claim 6 2.5 The method of the heavy metal emission reduction health benefit assessment system is characterized in that: In step S1, the data integration module uses an online heavy metal monitoring instrument to obtain PM 2.5 Chemical composition, meteorological parameters and pollution source activity data.
8. The PM based on machine learning combined with receptor model according to claim 6 2.5 The method of the heavy metal emission reduction health benefit assessment system is characterized in that: In step S2, the meteorological normalization module uses a random forest model with heavy metal concentration, meteorological factors and time variables as input, and the output variable is PM 2.5 and the anthropogenic emission constraint concentrations of each heavy metal component, a meteorological impact stripping model was constructed to generate the heavy metal concentration components dominated by anthropogenic emissions.
9. The PM based on machine learning combined with receptor model according to claim 6 2.5 The method of the heavy metal emission reduction health benefit assessment system is characterized in that: In step S3, the PMF receptor model analyzes the contribution rates of various pollution sources to heavy metals, and introduces a time constraint algorithm to associate policy implementation nodes to quantify the dynamic impact of specific measures on the contribution rates of heavy metal sources.
10. The PM based on machine learning combined with receptor model according to claim 6 2.5 The method of the heavy metal emission reduction health benefit assessment system is characterized in that: In step S4, the BenMAP model converts the reduction in heavy metal concentrations from different sources into the number of avoidable premature deaths, reduction in respiratory diseases, and medical cost savings, and generates a source-specific risk map.