Analytical methods for the chemical composition of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential

Abstract

The present invention discloses a method for analyzing the chemical components and pollution sources of atmospheric fine particulate matter and their contribution to oxidation potential, and relates to the technical field of prediction and analysis of atmospheric particulate pollutants. This method measures the concentration, chemical components, and oxidation potential of atmospheric fine particulate matter in the sample, uses a receptor model to obtain the contribution value of each pollution source to the atmospheric fine particulate matter sample, and uses a random forest algorithm, model interpretation, and structure mining methods to obtain the concentration of one or more chemical components, as well as the interactive effect of one or more pollution sources on the oxidation potential of atmospheric fine particulate matter. This method can quickly and accurately identify and quantify the key chemical components and pollution sources that affect the oxidation potential of atmospheric fine particulate matter, and quantify the single-factor and multi-factor effects of different pollution sources and chemical components on the oxidation potential; it has very high practical value and prospects for promotion and application, and provides a systematic technical guarantee for reducing the harm of air pollution to human health.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction and analysis of atmospheric particulate pollutants, and in particular to a method for analyzing the chemical components of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential. Background Art

[0002] Oxidative stress caused by reactive oxygen species (ROS) is a major cause of atmospheric fine particulate matter (PM) 2.5 The important potential mechanism of harm to human health. 2.5 Reactive oxygen species such as superoxide radicals (O2· - When the increase in hydroxyl radicals (HO·), peroxyhydroxyl radicals (HO2·), and hydrogen peroxide (H2O2) exceeds the body's antioxidant capacity, it will trigger oxidative stress in the body, leading to or exacerbating inflammation of the respiratory and cardiovascular systems. Oxidative potential (OP) refers to the ability of atmospheric fine particulate matter to directly or indirectly consume antioxidants in cells or produce oxidative substances; it is a key expression of oxidative stress in biological systems and is considered to be an effective way to evaluate PM2.5. 2.5 Indicators of exposure risk.

[0003] Atmospheric fine particulate matter with different pollution sources and chemical composition characteristics may produce different human health effects, and the potential toxicological mechanisms and pathological effects are not yet fully understood. For example, although crustal elements such as Si and Ca are present in PM 2.5 It accounts for a large proportion of the mass concentration, but its toxicity is low. There is an obvious intricate and nonlinear relationship between the chemical components, production sources and oxidation potential of different atmospheric fine particulate matter. There may be synergistic effects between some key chemical components or sources. In-depth exploration and understanding of the hidden interactions between different species and pollution sources can enable more targeted air pollution control and more effectively reduce the health risks of atmospheric fine particulate matter. Therefore, it is necessary to establish an accurate and effective assessment method to identify and quantify the key toxic components and sources that affect the oxidation potential of atmospheric fine particulate matter, so as to comprehensively assess the risks of atmospheric fine particulate matter to human health. Summary of the Invention

[0004] The present invention provides a method for analyzing the contributions of chemical components and pollution sources to the oxidation potential of atmospheric fine particulate matter. This method can identify and quantify the key toxic components and pollution sources that affect the oxidation potential of atmospheric fine particulate matter. It can also calculate the degree of influence of different chemical components and pollution sources on the oxidation potential of atmospheric fine particulate matter, and further quantify the influence of single factors on the oxidation potential of atmospheric fine particulate matter, as well as the synergistic influence of multiple factors. This is achieved specifically through the following technical methods.

[0005] A method for analyzing the chemical composition of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential comprises the following steps:

[0006] Obtain the concentration, chemical composition and oxidation potential of atmospheric fine particulate matter samples;

[0007] Constructing a first data set based on the concentration of the atmospheric fine particulate matter and the concentration of the chemical component, and combining it with regional pollution emission information, performing calculations using the receptor model; determining each pollution source based on the analysis results of the receptor model, analyzing the contribution value of each pollution source to the atmospheric fine particulate matter sample, and obtaining a source contribution result;

[0008] A first machine learning model and a second machine learning model are respectively constructed using a random forest algorithm; a second data set is constructed using the chemical component concentration as an independent variable and the oxidation potential of the atmospheric fine particulate matter as a dependent variable; and a third data set is constructed using the source contribution result as an independent variable and the oxidation potential of the atmospheric fine particulate matter as a dependent variable;

[0009] Using the second data set to train and verify the first machine learning model, and adjust and optimize the internal parameters of the first machine learning model; using the third data set to train and verify the second machine learning model, and adjust and optimize the internal parameters of the second machine learning model;

[0010] Based on the results of the first machine learning model, the average absolute value of the SHAP values ​​corresponding to the chemical components is calculated (the average absolute value of the SHAP values ​​is used to identify and quantify the effect of each chemical component on the oxidation potential of atmospheric fine particulate matter), and the effect of the concentration of a single chemical component on the oxidation potential is identified and quantified; a structure mining method is used to calculate the conditional minimum depth value (i.e., CMD value) corresponding to the concentration of the chemical component; and the interactive effect of the combination of the concentrations of at least two chemical components on the oxidation potential of atmospheric fine particulate matter is confirmed based on the conditional minimum depth value analysis;

[0011] Based on the results of the second machine learning model, the average absolute value of the SHAP value corresponding to the source contribution result is calculated to identify and quantify the impact of a single pollution source on the oxidation potential. The structural mining method is used to calculate the conditional minimum depth value corresponding to the source contribution result, and the interactive effect of the combination of at least two pollution sources on the oxidation potential of atmospheric fine particulate matter is analyzed and confirmed.

[0012] In the above analysis method of the present invention, when analyzing the source of atmospheric pollutants using the receptor model, it is possible to not rely on pollution sources and meteorological field information, but only through atmospheric fine particulate matter PM 2.5The chemical component concentrations in the data are quantitatively analyzed to derive the contribution of various pollution sources. This invention, based on a data-driven machine learning algorithm, learns the complex patterns and relationships in the data, achieving more flexible nonlinear fitting and efficiently and accurately mining nonlinear relationships between factors. This analytical method combines random forests (RF), model interpretation (SHAP), and structure mining analysis (SMA) to specifically analyze the corresponding relationships between individual features and dependent variables. This combination of chemical analysis, model simulation, and data-driven approaches provides new insights and directions for health risk assessment of atmospheric fine particulate matter.

[0013] It should be known to those skilled in the art that atmospheric fine particulate matter (PM 2.5 ) The atmospheric fine particulate matter concentration of the sample can be obtained by conventional measurement methods in this field.

[0014] In some embodiments of the present invention, the method for measuring the concentration of fine particulate matter in the atmosphere may be:

[0015] According to the Ambient Air Particulate Matter (PM 2.5 ) Manual Monitoring Method (Weight Method) Technical Specifications" by deploying PM 2.5 The sampling instrument collects fine particulate matter samples in the ambient air and obtains PM by weight method. 2.5 concentration.

[0016]

[0017] In the above formula (8), ρ is PM 2.5 Concentration (μg / m 3 ), w2 is the mass of the filter membrane after sampling (μg), w1 is the mass of the filter membrane before sampling (μg), V is the sampling volume under standard conditions (m 3 ).

[0018] Further, the chemical components include water-soluble ion components, elemental components and carbon components.

[0019] Furthermore, the water-soluble ion component includes F - 、Cl - 、NO3 - 、SO4 2- NH4 + 、Na + , K + , Ca 2+ Mg 2+ .

[0020] Furthermore, the elemental components include lithium, beryllium, aluminum, vanadium, chromium, manganese, iron, cobalt, nickel, copper, zinc, arsenic, rubidium, strontium, cadmium, tin, antimony, cesium, barium, thallium, lead and uranium.

[0021] Furthermore, the carbon components include organic carbon and elemental carbon, and the organic carbon includes water-soluble organic carbon. Furthermore, the regional pollution emission information is primarily used to understand historical data on atmospheric fine particulate matter emissions and pollution sources in a specific region, as well as data obtained from on-site field surveys. Therefore, regional pollution emission information includes, but is not limited to, regional economic development information, energy structure information, industrial layout information, pollutant emission information, and field research information.

[0022] Those skilled in the art should be aware that the concentrations of the above-mentioned chemical components in atmospheric fine particulate matter samples can be measured using instruments and detection methods commonly used in the art.

[0023] Furthermore, the receptor model is a positive matrix factorization model (PMF), a chemical mass balance model (CMB) or a principal factorization model (PCA).

[0024] Furthermore, the first machine learning model and the second machine learning model are constructed using the following method: using the root mean square error RMSE and the mean absolute error MAE as error indicators to respectively calculate the gap between the predicted values ​​and the true values ​​of the first machine learning model and the second machine learning model, and verify the accuracy of the prediction results of the first machine learning model and the second machine learning model.

[0025] Furthermore, the second data set is divided into a first training set and a first test set by using a ten-fold cross validation method; and the third data set is divided into a second training set and a second test set by using a ten-fold cross validation method.

[0026] Those skilled in the art should at least know that, in addition to being calculated using the DTT (dithiothreitol) detection method, the oxidative potential of atmospheric fine particulate matter can also be calculated using extracellular oxidative potential analysis methods such as the AA (ascorbic acid) detection method, the GSH (glutathione) detection method, or the EPR (electron paramagnetic resonance spectroscopy) detection method.

[0027] In some embodiments of the present invention, a method using dithiothreitol (DTT) as a strong reducing agent can be used to determine the oxidation potential in fine particulate matter. The experimental principle is as follows: when a DTT solution is added to an aqueous solution of atmospheric particles, the oxidative active substances in the atmospheric fine particulate matter catalyze the conversion of O2 into H2O2 and oxidize DTT into DTT-disulfide compounds; as the reaction continues, DTT is continuously consumed; DTNB is added at regular intervals, and the remaining DTT and DTNB generate yellow-brown TNB; by detecting the absorbance of the solution, the remaining DTT concentration in the reaction can be indirectly obtained, and the rate of DTT consumption during the reaction can be calculated.

[0028] The oxidative potential characteristic (OP) of fine particulate matter is characterized by the rate at which dithiothreitol is consumed per unit volume of atmospheric fine particulate matter, namely DTTv.

[0029]

[0030] In the above formula (9), V DTT represents the volume of DTT standard solution added to the centrifuge tube (μL), K m represents the DTT consumption rate (mM / min) determined by absorbance, V a Indicates the volume of gas passing through the sampling filter during sampling (m 3 ), S cut Indicates the area of ​​the cut sampling filter (m 2 ), S filter Indicates the total area actually sampled on the filter membrane (m 2 ).

[0031] Compared with existing technologies, the present invention offers significant advantages in that it provides a method for analyzing the chemical composition and pollution source contributions of atmospheric fine particulate matter to its oxidation potential. This method can rapidly and accurately identify and quantify the key toxic components and sources that influence the oxidation potential of atmospheric fine particulate matter, and quantify the single-factor and multi-factor effects of different pollution sources and chemical compositions on the oxidation potential. This method has significant practical value and promising prospects for widespread application, providing a systematic technical solution for reducing the harmful effects of air pollution on human health. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is an overall flow chart of the method for analyzing the chemical composition of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential provided in the examples.

[0033] Figure 2 The different pollution sources in the embodiment are the atmospheric fine particulate matter PM 2.5 contribution.

[0034] Figure 3 It represents the importance of the influence of different chemical component concentrations on the oxidation potential of atmospheric fine particulate matter in the examples.

[0035] Figure 4 In the examples, the concentrations of different chemical components are combined with each other to show the importance of these combinations in affecting the oxidation potential of atmospheric fine particulate matter.

[0036] Figure 5 for Figure 2 The different pollution sources analyzed in the paper have an impact on the atmospheric fine particulate matter PM 2.5 The oxidative potential of

[0037] Figure 6 For the general Figure 2The importance of the two-way combination of different pollution sources analyzed in the paper on the oxidation potential of atmospheric fine particulate matter. DETAILED DESCRIPTION

[0038] The technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0039] In some embodiments of the present invention, the oxidation potential of atmospheric fine particulate matter in the sample can be obtained by any one of the DTT (dithiothreitol) method, AA (ascorbic acid) detection method, GSH (glutathione) detection method or EPR (electron paramagnetic resonance spectroscopy) detection method.

[0040] In the following examples, the oxidation potential was determined using the DTT method. Therefore, the corresponding oxidation potential OP values ​​are all represented by , and in the following examples, the definitions of the two are the same. Example

[0041] The present embodiment provides a method for analyzing the chemical composition of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential. The overall process is as follows: Figure 1 As shown, the following steps are included.

[0042] 1. Sampling fine particulate matter and obtaining fine particulate matter samples

[0043] The atmospheric fine particulate matter samples selected in this example were specifically selected from Wuhan City, Hubei Province as the research area. The research period was from January to December 2018. According to the 2.5 ) Manual Monitoring Method (Weight Method) Technical Specifications" Deploy PM at monitoring points 2.5 The sampling instrument collected fine particulate matter samples in the ambient air. Finally, a total of 172 valid fine particulate matter samples were collected. The atmospheric fine particulate matter PM was calculated based on the weighing data of the sampling filter membrane. 2.5 mass concentration.

[0044] Specifically, the following formula (8) is used to obtain PM 2.5 mass concentration.

[0045]

[0046] In the above formula (8), ρ is PM 2.5 Concentration (μg / m 3 ), w2 is the mass of the filter membrane after sampling (μg), w1 is the mass of the filter membrane before sampling (μg), V is the sampling volume under standard conditions (m3 ).

[0047] 2. Test the water-soluble ion components, elemental components, carbon components and oxidation potential in fine particulate matter samples.

[0048] An atmospheric fine particulate matter sample is collected and the chemical component concentration and oxidation potential of the atmospheric fine particulate matter sample are measured. The chemical components include water-soluble ion components, elemental components, and carbon components.

[0049] (1) Determination of PM using ion chromatography 2.5 The water-soluble ion components in - 、Cl - 、NO3 - 、SO4 2- NH4 + 、Na + , K + , Ca 2+ Mg 2+ .

[0050] (2) Organic carbon (OC) and elemental carbon (EC) were determined using a carbon analyzer, and WSOC was determined using a total carbon analyzer.

[0051] (3) The elemental components were determined by an elemental analyzer, including lithium (Li), beryllium (Be), aluminum (Al), vanadium (V), chromium (Cr), manganese (Mn), iron (Fe), cobalt (Co), nickel (Ni), copper (Cu), zinc (Zn), arsenic (As), rubidium (Rb), strontium (Sr), cadmium (Cd), tin (Sn), antimony (Sb), cesium (Cs), barium (Ba), thallium (Tl), lead (Pb) and uranium (U).

[0052] (4) The oxidation potential of fine particulate matter was determined by using dithiothreitol (DTT) as a strong reducing agent. The oxidation potential characteristic (OP) of fine particulate matter was characterized by the rate of consumption of dithiothreitol per unit volume of atmospheric fine particulate matter (i.e., ).

[0053] The DTT method is a commonly used method in the field for measuring and calculating the oxidation potential of atmospheric fine particulate matter. In this embodiment, the method uses dithiothreitol (DTT) as a strong reducing agent to determine the oxidation potential of fine particulate matter. The experimental principle is as follows: When a DTT solution is added to an aqueous solution of atmospheric particulate matter, the oxidatively active substances in the atmospheric fine particulate matter catalyze the conversion of O2 into H2O2 and oxidize DTT into DTT-disulfide. As the reaction proceeds, DTT is continuously consumed. DTNB is added periodically, and the remaining DTT and DTNB form a yellow-brown TNB. By monitoring the absorbance of the solution, the concentration of residual DTT in the reaction can be indirectly determined, and the rate of DTT consumption during the reaction can be calculated.

[0054] The oxidative potential characteristic (OP) of fine particulate matter is characterized by the rate at which dithiothreitol is consumed per unit volume of atmospheric fine particulate matter.

[0055]

[0056] In the above formula (9), V DTT represents the volume of DTT standard solution added to the centrifuge tube (μL), K m represents the DTT consumption rate (mM / min) determined by absorbance, V a Indicates the volume of gas passing through the sampling filter during sampling (m 3 ), S cut Indicates the area of ​​the cut sampling filter (m 2 ), S filter Indicates the total area actually sampled on the filter membrane (m 2 ).

[0057] The method for measuring the oxidation potential of atmospheric fine particulate matter in this embodiment is specifically as follows:

[0058] ① Use a clean cutter to remove the sample filter membrane, place it in a clean brown glass bottle, add 5 mL of ultrapure water and sonicate for 30 minutes;

[0059] ② Using a disposable syringe filter to filter the mixed solution after ultrasonication to obtain a filtrate;

[0060] ③ Add 1000 μL of phosphate buffer and 50 μL of sample filtrate to five brown centrifuge tubes in sequence; add 250 μL of DTT working solution and quickly place in a constant temperature shaking incubator; start timing at 37°C and a speed of 120 rpm. The reaction times for the five brown centrifuge tubes are 0 min, 5 min, 10 min, 15 min, and 20 min, respectively.

[0061] ④ After the reaction is completed, quickly add 1000 μL of DTNB working solution to the centrifuge tube and allow to react for 5 minutes. Pour the solution into a cuvette and measure the absorbance of the solution at 412 nm using a UV-visible spectrophotometer.

[0062] Before the measurement of each sample, the absorbance intensity of the blank group must be measured, and 50 μL ultrapure water is used to replace the sample filtrate in the blank group.

[0063] 3. Use receptor models to analyze the sources and contributions of atmospheric fine particulate matter.

[0064] (1) The concentration of atmospheric fine particulate matter and the concentration of chemical components are used to construct a first data set as input data to the receptor model. The receptor model can use the positive definite matrix factorization method (PMF model), the chemical mass balance method (CMB model), the principal factor analysis method (PCA model), etc. In this example, the PMF model is selected for pollution source analysis.

[0065] (2) Based on the collected information on economic development, energy structure, industrial layout, pollutant emissions, etc. in the research area, as well as the results of actual investigations and surveys, the calculation parameters of the receptor model are set, including the number of pollution sources, uncertainty, number of runs, etc.; based on the analytical results of the receptor model, several different pollution sources of fine particulate matter are determined, and the contribution value of each pollution source to the atmospheric fine particulate matter sample is analyzed to obtain the source contribution results.

[0066] This example uses the PMF model to analyze pollution sources. The concentration of atmospheric fine particulate matter and chemical components are input into the PMF model, which contains 34 chemical components, including PM 2.5 Concentration was designated as the "total variable." Model parameters were selected for factors ranging from 3 to 7, and error estimates were performed for each case. Ultimately, a solution with 6 factors was found to be optimal. The final model parameters were: 172 rows, 35 columns, and 6 factors, and the calculation was repeated 20 times.

[0067] Identify pollution sources based on characteristic components of source spectrum (coal-burning source Cl - 、SO4 2- , EC; biomass combustion source K + , OC, EC; industrial sources As, Cd, Cu, Pb, Zn; motor vehicle sources Zn, EC, Ni, Cu; dust source Mg 2+ , Al, Ca 2+ ; Secondary generation source NO3 - 、SO4 2- NH4 + ); select the most reasonable result.

[0068] The contribution of different pollution sources is as follows Figure 2 As shown in the figure, the positive matrix factorization model (PMF model) was used to analyze six pollution sources, which are secondary sources (SA), coal burning sources (CC), dust sources (Dust), biomass combustion sources (BB), motor vehicle sources (VE), and industrial sources (IS). The source analysis results are consistent with the collected data of the study area and the actual survey situation, and are consistent with the real atmospheric characteristics. It identifies and quantifies the contribution of different pollution sources to PM 2.5 contribution.

[0069] 4. Calculate the effect of different chemical component concentrations on PM 2.5 The degree of influence of oxidation potential.

[0070] (1) The measured atmospheric fine particulate matter PM 2.5 The concentration of chemical components in the atmosphere was used as the independent variable, and the oxidation potential of atmospheric fine particulate matter was used as the dependent variable to construct a second data set. The second data set was divided into a first training set and a first test set through ten-fold cross-validation.

[0071] (2) Build a first machine learning model using a random forest algorithm, and train and verify the first machine learning model using the first training set and the first test set; continuously adjust and optimize the internal parameters of the machine learning based on the size of the model evaluation index to achieve model optimization.

[0072] Random forest is a machine learning algorithm based on decision trees. It improves the accuracy and stability of the model by constructing multiple decision trees and combining the results of the decision trees. It has strong nonlinear fitting capabilities.

[0073] The root mean square error (RMSE) and mean absolute error (MAE) are used as error indicators to determine and calculate the difference between the predicted value and the true value of the first machine learning model, and to verify the accuracy of the prediction results of the first machine learning model. The calculation formula is as follows:

[0074]

[0075] Formulas (1) and (2), DTTv i is the predicted value calculated by the first machine learning model or the second machine learning model, DTTv i * is the true value, and n is the number of samples.

[0076] (3) Based on the results of the first machine learning model, the average absolute value of the SHAP value corresponding to the concentration of the chemical component is calculated. The average absolute value of the SHAP value can be used to identify and quantify the effect of different chemical component concentrations on the oxidation potential of atmospheric fine particulate matter.

[0077] SHAP is an interpretable method for machine learning models based on Sharpley's idea in game theory. It helps explain the model's complex predictions and calculates the physical significance of each characteristic factor in the prediction results. This example combines random forests and SHAP to verify the accuracy of variable importance. A larger average absolute value of the SHAP value indicates a greater impact on DTTv.

[0078] The calculation formula for the average absolute value of the SHAP value is as follows:

[0079]

[0080] In formulas (3) to (6): DTTv (base) DTTv i The expected value of shap(x i,j ) is the effect of characteristic factor j on DTTv in sample i i Contribution value; DTTv M is the simulated value of the random forest algorithm under the condition of feature subset M; feature subset M is the feature subset that does not include feature factor j; |M| is the number of types of feature variables included in feature subset M; S is the number of types of feature variables in the total set of feature variables; β M is the weight of feature subset M.

[0081] like Figure 3 As shown, EC, Cl - 、NO3 - , K + These are the top four chemical components that affect DTTv.

[0082] (4) A structural mining method was used to calculate the CMD value (conditional minimum depth) to explore the importance of paired chemical component concentration characteristics on the dependent variable, so as to analyze and confirm the interactive effects between the concentrations of each chemical component. The smaller the CMD value, the greater the contribution to the DTTv.

[0083] In the random forest model, the minimum depth at which a feature affecting oxidative potential first appears under the condition that one or more other features have been selected as split points is the conditional minimum depth.

[0084] The method for calculating the conditional minimum depth value is: under the given characteristic factors n, p, ..., q affecting the oxidation potential, the conditional minimum depth calculation formula of the characteristic factor m affecting the oxidation potential is:

[0085]

[0086] In formula (7), t is a “tree” set by the “random forest model”, and T is the number of “trees” set by the “random forest model”.

[0087] When CMD m-(n,p,…q) The smaller it is, the closer the relationship between characteristic factor m and other characteristic factors such as n, p,..., q is, and the greater the effect of the combination of characteristic factors on oxidation potential.

[0088] For this embodiment, when only the interaction between two characteristic factors is considered, formula (7) is simplified to the following formula (10).

[0089]

[0090] CMD m-nThe smaller the value, the closer the relationship between the two characteristic factors m and n is, and the greater the effect of the combination of the two characteristic factors on the oxidation potential. Figure 4 shown.

[0091] Combine Figure 3 and Figure 4 It was found that the interaction between variables was significant. For example, the effect of As as a single chemical component on DTTv ranked 9th among all chemical components, while in the presence and influence of EC, the effect of EC-As on DTTv ranked 6th among all combinations. In other words, the combination of EC and As makes the contribution of As to DTTv significant. Figure 3 and 4 It can also be seen that other metal elements such as Co, Rb, V, Tl, etc., in combination with other elements, contribute significantly to the oxidation potential.

[0092] By arbitrarily selecting the interaction of three characteristic factors (such as the concentration of chemical components), we studied and analyzed the impact of the combination of the three on the oxidation potential. We also found that the ranking of the contribution of the combination of different chemical components to the oxidation potential changed significantly compared with the ranking of the contribution of the individual chemical components to the oxidation potential. The contribution of certain chemical components to the oxidation potential increased or decreased significantly in the presence of other chemical components, and the mutual influence between chemical components was obvious.

[0093] 5. Calculate the concentration of each pollution source and chemical component, and calculate the PM 2.5 The degree of influence of oxidation potential.

[0094] (1) The source contribution results obtained from the receptor model analysis were used as the independent variable, and the atmospheric fine particulate matter oxidation potential was used as the dependent variable to form a third data set. The third data set was divided into a second training set and a second test set through ten-fold cross-validation.

[0095] (2) Building a second machine learning model using a random forest algorithm, and training and verifying the second machine learning model using the second training set and the second test set.

[0096] (3) According to the size of the evaluation index of the second machine learning model, the internal parameters of the second machine learning model are continuously adjusted and optimized to achieve model optimization.

[0097] (4) Combine the average absolute value of SHAP value and CMD value to identify the factors affecting PM 2.5 Important pollution sources of oxidation potential, quantify the single factor and multi-factor effects of pollution sources on oxidation potential.

[0098] The results are as follows Figure 5 、 6 As shown. It can be seen that although according to Figure 2PM 2.5 Pollution source analysis results, Dust source to PM 2.5 The contribution is 22.26%, which is the PM 2.5 The largest contributing source; however Figure 5 The results show that the contribution of dust sources to DTTv is only the fourth among all pollution sources. This indicates that dust sources are not the main contributors to PM2.5. 2.5 The sources of pollution that have the greatest impact on health risks.

[0099] You can also see that Figure 5 Among them, BB source is the second largest source affecting DTTv. Figure 6 The BB-SA combination in the study became the most important combination for DTTv. The ranking of other pollution sources in terms of their contribution to oxidation potential changed significantly when combined with SA compared to their individual rankings.

[0100] By arbitrarily selecting the interaction of three characteristic factors (such as different pollution sources), we studied and analyzed the impact of the combination of the three on the oxidation potential. We also found that the ranking of the contribution of the combination of different pollution sources to the oxidation potential was significantly different from the ranking of the contribution of a single pollution source to the oxidation potential. The contribution of some pollution sources to the oxidation potential increased or decreased significantly in the presence of other pollution sources, and the mutual influence between different pollution sources was obvious.

[0101] This shows that the interactions between different pollution sources cannot be ignored.

[0102] In the above analysis method of the present invention, when analyzing the source of atmospheric pollutants using the receptor model, it is possible to not rely on pollution sources and meteorological field information, but only through atmospheric fine particulate matter PM 2.5 The contribution of various pollution sources can be quantitatively analyzed using component information. This invention, based on a data-driven machine learning algorithm, learns complex patterns and relationships in the data, achieving more flexible nonlinear fitting and efficiently and accurately mining nonlinear relationships between factors. Combining Random Forest (RF), the Model Interpretation Method (SHAP), and Structural Mining Analysis (SMA) methods, it can specifically analyze the corresponding relationships between individual features and dependent variables.

[0103] The above specific embodiments describe the implementation of the present invention in detail, but the present invention is not limited to the specific details of the above embodiments. Within the scope of the claims and technical concept of the present invention, various simple modifications and changes can be made to the technical solution of the present invention, and these simple modifications all fall within the scope of protection of the present invention.

Claims

1. A method for analyzing the chemical composition of atmospheric fine particulate matter and the contribution of pollution sources to oxidation potential, characterized in that: The following steps are involved: Obtain the concentration, chemical composition and oxidation potential of atmospheric fine particulate matter samples; Constructing a first data set based on the atmospheric fine particulate matter concentration and the chemical component concentration, and performing calculations using a receptor model in combination with regional pollution emission information; determining each pollution source based on the analytical results of the receptor model, analyzing the contribution value of each pollution source to the atmospheric fine particulate matter sample, and obtaining a source contribution result; A first machine learning model and a second machine learning model are respectively constructed using a random forest algorithm; a second data set is constructed using the concentration of the chemical component as an independent variable and the oxidation potential of the atmospheric fine particulate matter as a dependent variable; A third data set is constructed using the source contribution results as the independent variable and the oxidation potential of the atmospheric fine particulate matter as the dependent variable; Using the second data set to train and verify the first machine learning model, and adjust and optimize the internal parameters of the first machine learning model; using the third data set to train and verify the second machine learning model, and adjust and optimize the internal parameters of the second machine learning model; Calculating the average absolute value of the SHAP value corresponding to the chemical component concentration based on the first machine learning model results to identify and quantify the impact of the concentration of a single chemical component on the oxidative potential; The structure mining method is used to calculate the conditional minimum depth value corresponding to the concentration of the chemical components, and the interactive effect of the combination of at least two chemical component concentrations on the oxidation potential of atmospheric fine particulate matter is analyzed and confirmed; Based on the results of the second machine learning model, the average absolute value of the SHAP value corresponding to the source contribution result is calculated to identify and quantify the impact of a single pollution source on the oxidation potential. The structural mining method is used to calculate the conditional minimum depth value corresponding to the source contribution result, and the interactive effect of the combination of at least two pollution sources on the oxidation potential of atmospheric fine particulate matter is analyzed and confirmed.

2. The analysis method according to claim 1, characterized in that Chemical components include water-soluble ion components, elemental components and carbon components; the water-soluble ion components include F - 、Cl - 、NO3 - 、SO4 2- NH4 + 、Na + , K + , Ca 2+ Mg 2+ The elemental components include lithium, beryllium, aluminum, vanadium, chromium, manganese, iron, cobalt, nickel, copper, zinc, arsenic, rubidium, strontium, cadmium, tin, antimony, cesium, barium, thallium, lead and uranium, and the carbon component includes organic carbon and elemental carbon, and the organic carbon includes water-soluble organic carbon.

3. The analysis method according to claim 1, characterized in that The regional pollution emission information includes regional economic development information, energy structure information, industrial layout information and pollutant emission information, as well as field survey information.

4. The analysis method according to claim 1, characterized in that The receptor model is a positive definite matrix factorization model, a chemical mass balance model or a principal factorization model.

5. The analysis method according to claim 1, characterized in that The first machine learning model and the second machine learning model are constructed using the following method: using the root mean square error RMSE and the mean absolute error MAE as error indicators to calculate the gap between the predicted values ​​and the true values ​​of the first machine learning model and the second machine learning model respectively, and verify the accuracy of the prediction results of the first machine learning model and the second machine learning model.

6. The analysis method according to claim 1, characterized in that The second data set is divided into a first training set and a first test set by using a ten-fold cross validation method; the third data set is divided into a second training set and a second test set by using a ten-fold cross validation method.

7. The analysis method according to claim 1, characterized in that The oxidation potential of the atmospheric fine particulate matter in the sample is determined by any one of the DTT detection method, the AA detection method, the GSH detection method, and the EPR detection method.

Images