A Machine Learning-Based Method for Source Analysis of PM2.5 Particulate Matter

Abstract

The present application relates to PM 2.5 The technical field of pollution, specifically relates to a kind of PM 2.5 Particle source analysis method based on machine learning.The present application provides a kind of PM 2.5 Particle source analysis method based on machine learning, based on PM 2.5 Chemical component mass concentration data acquisition difficulty, establish the relationship between chemical component mass concentration data, meteorological data, atmospheric pollutant emission data and source contribution rate data, build PMF source analysis simulation model to PM 2.5 Particle source analysis.The PMF source analysis simulation model provided by the present application can realize the rapid and effective source analysis analysis of PM 2.5 ;And it is not related to the use of short equipment, applicable to the PM 2.5 Supervision and monitoring of most areas.

Description

Technical Field

[0001] This invention relates to PM 2.5 In the field of pollution technology, specifically involving a PM based on machine learning. 2.5 Methods for analyzing particulate matter sources. Background Technology

[0002] In the field of atmospheric environment, in order to quantify fine particulate matter (PM2.5) in the urban atmosphere... 2.5 The source contribution of PM2.5 in the atmosphere is usually determined by using field observation equipment. 2.5 The mass concentration of PM2.5 and its chemical components is observed in real time, and then the observed high-resolution PM2.5 is used to... 2.5 The mass concentration data of the pollutants and their chemical components are used as input data for a positive definite matrix factorization (PMF) model. The receptor model is then used to calculate and analyze the impact of pollution sources on urban atmospheric PM2.5 levels. 2.5 The contribution rate.

[0003] Currently, environmental protection departments typically need to implement timely and immediate measures to control urban air quality. However, the aforementioned source apportionment methods for atmospheric particulate matter require the accumulation of high-resolution chemical component mass concentration data and model calculations, which is not only time-consuming but also affects the timeliness of recommendations for air quality control measures. Meanwhile, monitoring PM2.5... 2.5 There are few devices available for measuring the mass concentration of chemical components, which cannot meet the PM2.5 requirements in areas with equipment shortages. 2.5 Regulatory needs. Summary of the Invention

[0004] This invention provides a machine learning-based PM 2.5 Methods, systems, apparatus, and storage media for particulate matter source analysis. This method can achieve analysis of PM2.5. 2.5 Rapid and effective source analysis can effectively solve current equipment-dependent problems, benefiting PMs in many areas with equipment shortages. 2.5 Regulatory oversight and monitoring.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a machine learning-based PM 2.5 Methods for analyzing particulate matter sources include:

[0007] S1. Data collection unit time PM 2.5 Given the mass concentration data of the chemical components, calculate the uncertainty data corresponding to the mass concentration data of each chemical component.

[0008] S2. The PM per unit time collected by S1 2.5 The chemical component mass concentration data and the calculated uncertainty data are used to obtain PM. 2.5The source contribution rate;

[0009] S3. In terms of unit time PM 2.5 The data collection point for the chemical component mass concentration is center A, at PM 2.5 During the data collection period for the mass concentration of chemical components, the collection points B1, B2, and B3 near the collection center A were included. n (n = 2, 3, 4, 5, ..., a; a is a positive integer > 2) represent the meteorological data and atmospheric pollutant emission data per unit time, respectively, PM2.5. 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same time unit; then, meteorological samples and air pollutant emission samples were obtained by sampling the meteorological data and air pollutant emission data from different collection points.

[0010] PM collected by S4.S1 2.5 The chemical component mass concentration data is used as the chemical component sample, the source contribution rate calculated by S2 is used as the source contribution rate sample, and the meteorological sample and atmospheric pollutant emission sample obtained by S3 are used as the source contribution rate sample. A total of four samples are used to form a multi-source dataset. The multi-source dataset is used for simulation training to obtain the PMF source apportionment simulation model.

[0011] S5. Collection point B near collection center A m (m is a positive integer ≥ 1) Meteorological data and atmospheric pollutant emission data per unit time during a certain period, and PM2.5 emissions per unit time for center A during that period. 2.5 Chemical component mass concentration data, PM 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same unit time; the collected PM2.5 data were then processed. 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data were substituted into the PMF source apportionment simulation model of S4 to obtain the data from collection point B. m PM during that period 2.5 The source contribution rate.

[0012] Known chemical component mass concentration data and PM 2.5 The source of the relationship exists: when PM 2.5 Characteristic species in chemical components that are associated with emission sources are considered source indicator species. When the content of a certain source characteristic species is high, it can be attributed to the contribution of that source. For example, when PM2.5 is high... 2.5 When the chemical composition contains high levels of organic carbon, sulfate, chloride ions, arsenic, and sodium ions, the source can be identified as coal combustion; when PM2.5 levels are high... 2.5 When the chemical composition contains high levels of elements such as carbon, nickel, lead, zinc, and copper, the source can be identified as traffic contamination; when PM2.5 levels are high...2.5 When the content of calcium ions, magnesium ions, silicon, and aluminum elements in the chemical composition is high, the source can be identified as a dust source; when PM... 2.5 When the content of ammonium salts, nitrates, and sulfates in the chemical composition is high, the source can be identified as secondary aerosols; when PM... 2.5 When the chemical composition contains high levels of elemental carbon, organic carbon, potassium, and chloride ions, the source can be identified as biomass combustion; when PM2.5 content is high... 2.5 When the chemical composition contains high levels of metallic elements such as manganese, titanium, chromium, magnesium, iron, zinc, and arsenic, the source can be identified as industrial. Therefore, the applicant first uses a positive definite matrix factorization model to determine the PM2.5 content. 2.5 Chemical component mass concentration converted to corresponding PM 2.5 The source contribution rate, here meaning the contribution of pollution sources to urban atmospheric PM2.5 levels. 2.5 By obtaining the source contribution rate, the atmospheric PM2.5 concentration can be analyzed. 2.5 The types of pollution sources and the proportion of each type can comprehensively and effectively analyze PM2.5. 2.5 The purpose of pollution sources.

[0013] Furthermore, based on the actual PM situation, the applicant... 2.5 To address the difficulty of obtaining chemical component mass concentration data, this paper establishes relationships between chemical component mass concentration data, meteorological data, air pollutant emission data, and source contribution rate data, and constructs a PMF source apportionment simulation model. The PMF source apportionment simulation model inputs easily monitored meteorological and air pollutant emission data, sharing chemical component mass concentration data at a common central point. The PMF source apportionment simulation model can then directly obtain the PM2.5 concentration data for the area near the central point. 2.5 The source contribution rate.

[0014] Preferably, in step S1, the PM data is collected per unit time. 2.5 After obtaining the chemical component mass concentration data, the data is first preprocessed to remove impurities, and then the source data format is unified.

[0015] Preferably, in step S1, the uncertainty data is calculated as follows: the standard degree of the instrument used to monitor the mass concentration data of chemical components is obtained, and the PM2.5 concentration per unit time is collected. 2.5 Substitute the mass concentration data of the chemical components into equation (1) to calculate the uncertainty data corresponding to the mass concentration data of each chemical component;

[0016]

[0017] In the formula, EF is the error ratio; A is the unit time (PM). 2.5 Chemical component mass concentration data, μg / m 3MDL is the detection limit of the instrument used to acquire mass concentration data of chemical components; B is the uncertainty data, μg / m 3 .

[0018] Preferably, in S1, the chemical components include water-soluble ions, carbon components, and metal elements; wherein the water-soluble ions include nitrate ions, sulfate ions, ammonium ions, magnesium ions, potassium ions, and calcium ions; the carbon components include organic carbon and elemental carbon; and the metal elements include mercury, iron, arsenic, calcium, silicon, aluminum, manganese, chromium, and lead.

[0019] The components listed here are PM 2.5 The PM collected in this application is a common major component, but not all components. 2.5 The chemical composition actually encompasses almost all possible single components.

[0020] Preferably, in step S2, after substituting into the positive definite matrix factorization model for calculation, the ratio of the mass concentration of a single chemical component to the total mass concentration of chemical components per unit time is obtained, and the ratio is used to determine the PM per unit time. 2.5 The source, ultimately obtained from PM 2.5 The source contribution rate.

[0021] Preferably, in step S3, a circle with center A as the center and a radius of 50 kilometers is drawn to define the area, and the area is divided into collection point B1, collection point B2, and collection point B with a range of 9 kilometers × 9 kilometers. n (n = 2, 3, 4, 5, ..., a; a is a positive integer greater than 2).

[0022] More preferably, when each collection point has two or more monitoring stations for acquiring meteorological data or atmospheric pollutant emission data per unit time, the meteorological data per unit time of each station is averaged to obtain the meteorological data per unit time of that collection point, and the atmospheric pollutant emission data per unit time of each station is averaged to obtain the atmospheric pollutant emission data per unit time of that collection point.

[0023] This application aims to establish PM 2.5 A PMF source apportionment simulation model is constructed by relating chemical component mass concentration data, meteorological data, air pollutant emission data, and source contribution rate data. High correlation is required between the chemical component mass concentration data, meteorological data, and air pollutant emission data; otherwise, the training results will lack specificity and accuracy. Therefore, the regional restrictions on data sources are high. If the region is too large, the correlation between chemical component mass concentration data and meteorological and air pollutant emission data becomes too weak to establish an accurate relationship.

[0024] Preferably, in step S3, the meteorological data includes wind speed data, wind direction data, temperature data, humidity data, and pressure data; the air pollutant emission data includes SO2 data, NO... x Data, PM 10 Data, PM 2.5 Data, BC data, OC data, NH3 data, and NMVOCs data.

[0025] Preferably, in step S3, the sampling method for the collection points is Latin hypercube sampling.

[0026] Preferably, in step S3, the collected meteorological data and atmospheric pollutant emission data per unit time are cleaned and then sampled at the collection points.

[0027] Preferably, in step S4, the simulation training method is as follows: input the collection point B near the input center A. x (x is a positive integer ≥ 1) Meteorological data and atmospheric pollutant emission data per unit time during a specific period, and PM2.5 per unit time at center A. 2.5 Chemical component mass concentration data, PM 2.5 The data on chemical component mass concentration, meteorological data, and atmospheric pollutant emission data were collected at the same unit time and time; PM 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data are substituted into the PMF source apportionment simulation model of S4 to predict the data collection point B. x PM during that period 2.5 Source contribution rate, relative to collection point B x The actual PM during this period 2.5 Compare source contribution rates; if the predicted PM 2.5 Source contribution rate and actual PM 2.5 If the source contribution rates differ significantly, adjust the parameters of the extreme gradient boosting model and return to training until the predicted PM is reached. 2.5 Source contribution rate and actual PM 2.5 The source contribution rates are consistent.

[0028] This invention also provides a machine learning-based PM 2.5 The particulate matter source apportionment system includes:

[0029] The data acquisition module is used to acquire samples of chemical component mass concentration data and PM. 2.5 Source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples;

[0030] Among them, PM 2.5 Source contribution rate data samples were obtained by calculation using chemical component mass concentration data samples and uncertainty data;

[0031] The simulation training module is used to process the chemical component mass concentration data samples and PM data acquired by the data acquisition module. 2.5 Simulation training was conducted using source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples to obtain a PMF source apportionment simulation model;

[0032] The analysis module uses the PMF source analysis simulation model to analyze PM. 2.5 The particulate matter source was analyzed.

[0033] This invention also provides a machine learning-based PM 2.5 A particulate matter source analysis apparatus includes: a memory for storing a computer program; and a processor for executing the computer program to implement the aforementioned machine learning-based PM analysis. 2.5 Methods for analyzing particulate matter sources.

[0034] The present invention also provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, implements the above-mentioned machine learning-based PM. 2.5 Methods for analyzing particulate matter sources.

[0035] Therefore, the beneficial effects of this invention are: the PMF source analysis simulation model and analysis method provided by this invention can realize the analysis of PM... 2.5 Rapid and effective source analysis; and does not involve the use of scarce equipment, applicable to PM in most areas. 2.5 Regulatory oversight and monitoring. Attached Figure Description

[0036] Figure 1 The machine learning-based PM provided by this invention 2.5 Flowchart of particulate matter source apportionment method;

[0037] Figure 2 Distribution map of XX branch stations and predicted stations in XX city;

[0038] Figure 3 This is a comparison chart of the errors between the actual observed values ​​and the model predicted values ​​at a sampling point in XX City, as shown in Example 1.

[0039] Figure 4 A comparison chart of predicted and observed PM2.5 source apportionment values ​​for XX City;

[0040] Figure 5 This is a map showing the source apportionment results of particulate matter at all grid points in XX City. Detailed Implementation

[0041] The present invention will be further described below with reference to specific embodiments. Those skilled in the art will be able to implement the present invention based on these descriptions. Furthermore, the embodiments of the present invention described below are generally only some, not all, of the embodiments of the present invention. Therefore, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0042] Example 1

[0043] A machine learning-based PM 2.5 Methods for analyzing particulate matter sources include:

[0044] S1. Data collection unit time PM 2.5 The chemical component mass concentration data were first preprocessed to remove outliers, blank values, and other interfering data. The source data format was then standardized to obtain the processed PM per unit time. 2.5 The chemical component mass concentration data is obtained, and then the uncertainty data corresponding to the processed chemical component mass concentration data is calculated.

[0045] The uncertainty data is calculated as follows: the standard degree of the instrument used to monitor the mass concentration data of chemical components is obtained, and the PM2.5 concentration per unit time is collected. 2.5 Substitute the mass concentration data of the chemical components into equation (1) to calculate the uncertainty data corresponding to the mass concentration data of each chemical component;

[0046]

[0047] In the formula, EF is the error ratio; A is the unit time (PM). 2.5 Chemical component mass concentration data, μg / m 3 MDL is the detection limit of the instrument used to acquire mass concentration data of chemical components; B is the uncertainty data, μg / m 3 .

[0048] The chemical components include water-soluble ions, carbon components, and metal elements; the water-soluble ions include nitrate ions, sulfate ions, ammonium ions, magnesium ions, potassium ions, and calcium ions; the carbon components include organic carbon and elemental carbon; and the metal elements include mercury, iron, arsenic, calcium, silicon, aluminum, manganese, chromium, and lead.

[0049] S2. The PM per unit time collected by S1 2.5The chemical component mass concentration data and the calculated uncertainty data are substituted into a positive definite matrix factorization model for calculation to obtain the ratio of the mass concentration of a single chemical component to the total mass concentration of chemical components per unit time. This ratio is then used to determine the PM per unit time. 2.5 The source, ultimately obtained from PM 2.5 The source contribution rate.

[0050] S3. In terms of unit time PM 2.5 The collection point for the chemical component mass concentration data is center A. A circle with center A and a radius of 50 kilometers is drawn to delineate the area. This area is then divided into collection points B1, B2, and B3, each measuring 9 kilometers x 9 kilometers. n (n = 2, 3, 4, 5, ..., a; a is a positive integer > 2); When each collection point has two or more monitoring stations to obtain meteorological data or atmospheric pollutant emission data per unit time, the meteorological data per unit time from each station is averaged to obtain the meteorological data per unit time for that collection point, and the atmospheric pollutant emission data per unit time from each station is averaged to obtain the atmospheric pollutant emission data per unit time for that collection point. Meteorological data includes wind speed, wind direction, temperature, humidity, and pressure data; atmospheric pollutant emission data includes SO2 data, NO... x Data, PM 10 Data, PM 2.5 Data, BC data, OC data, NH3 data, and NMVOCs data.

[0051] In PM 2.5 During the data collection period for the mass concentration of chemical components, the collection points B1, B2, and B3 near the collection center A were included. n (n = 2, 3, 4, 5, ..., a; a is a positive integer > 2) represent the meteorological data and atmospheric pollutant emission data per unit time, respectively, PM2.5. 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same time unit. Subsequently, the meteorological data and air pollutant emission data obtained from different collection points were cleaned and then subjected to Latin hypercube sampling at the collection points to obtain meteorological samples and air pollutant emission samples.

[0052] PM collected by S4.S1 2.5 The chemical component mass concentration data is used as the chemical component sample, the source contribution rate calculated by S2 is used as the source contribution rate sample, and the meteorological sample and atmospheric pollutant emission sample obtained by S3, totaling four samples, constitute a multi-source dataset; the multi-source dataset is then substituted into the extreme gradient boosting model for simulation training. The data collection point B near input center A is used. x(x is a positive integer ≥ 1) Meteorological data and atmospheric pollutant emission data per unit time during a specific period, and PM2.5 per unit time at center A. 2.5 Chemical component mass concentration data, PM 2.5 The data on chemical component mass concentration, meteorological data, and atmospheric pollutant emission data were collected at the same unit time and time; PM 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data are substituted into the PMF source apportionment simulation model of S4 to predict the data collection point B. x PM during that period 2.5 Source contribution rate, relative to collection point B x The actual PM during this period 2.5 Compare source contribution rates; if the predicted PM 2.5 Source contribution rate and actual PM 2.5 If the source contribution rates differ significantly, adjust the parameters of the extreme gradient boosting model and return to training until the predicted PM is reached. 2.5 Source contribution rate and actual PM 2.5 With consistent source contribution rates, a PMF source analysis simulation model was obtained through training.

[0053] S5. Collection point B near collection center A m (m is a positive integer ≥ 1) Meteorological data and atmospheric pollutant emission data per unit time during a certain period, and PM2.5 emissions per unit time for center A during that period. 2.5 Chemical component mass concentration data, PM 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same unit time; the collected PM2.5 data were then processed. 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data were substituted into the PMF source apportionment simulation model of S4 to obtain the data from collection point B. m PM during that period 2.5 The source contribution rate.

[0054] Example 2

[0055] A machine learning-based PM 2.5 The particulate matter source apportionment system includes:

[0056] The data acquisition module is used to acquire samples of chemical component mass concentration data and PM. 2.5 Source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples;

[0057] Among them, PM 2.5 Source contribution rate data samples were obtained by calculation using chemical component mass concentration data samples and uncertainty data;

[0058] The simulation training module is used to process the chemical component mass concentration data samples and PM data acquired in the data acquisition module. 2.5 Simulation training was conducted using source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples to obtain a PMF source apportionment simulation model;

[0059] The analysis module uses the PMF source analysis simulation model to analyze PM. 2.5 The particulate matter source was analyzed.

[0060] Example 3

[0061] A machine learning-based PM 2.5 A particulate matter source analysis apparatus includes: a memory for storing a computer program; and a processor for executing the computer program to implement the aforementioned machine learning-based PM analysis. 2.5 Methods for analyzing particulate matter sources.

[0062] Example 4

[0063] A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the above-mentioned machine learning-based PM. 2.5 Methods for analyzing particulate matter sources.

[0064] Application Example 1

[0065] This application example uses atmospheric PM2.5 in XX District, XX City. 2.5 Taking particulate matter pollution as an example, the analytical method of Example 1 was used to analyze PM2.5 in the atmosphere. 2.5 Particulate matter source apportionment is performed. Details are as follows:

[0066] ① Collection of chemical component mass concentration data

[0067] Data collected from the XX branch station in XX district of XX city for the entire month of January 2023, hourly PM. 2.5 The chemical component mass concentration data is first preprocessed to remove outliers, blank values, and other interfering data. Then, the source data format is standardized to obtain the processed hourly PM2.5 concentration. 2.5 The chemical component mass concentration data is obtained, and then the uncertainty data corresponding to the processed chemical component mass concentration data is calculated.

[0068] The uncertainty data is calculated as follows: the standard degree of the instrument used to monitor the mass concentration data of chemical components is obtained, and the hourly PM data is collected. 2.5 Substitute the mass concentration data of the chemical components into equation (1) to calculate the uncertainty data corresponding to the mass concentration data of each chemical component;

[0069]

[0070] In the formula, EF is the error ratio; A is the unit time (PM). 2.5 Chemical component mass concentration data, μg / m 3 MDL is the detection limit of the instrument used to acquire mass concentration data of chemical components; B is the uncertainty data, μg / m 3 .

[0071] Table 1. Instrumental Detection Limits for Fine Particulate Matter Components in XX City

[0072] Species <![CDATA[NO3 - ]]> <![CDATA[SO4 2- ]]> <![CDATA[NH4 + ]]> <![CDATA[K + ]]> <![CDATA[Mg 2+ ]]> <![CDATA[Ca 2+ ]]> OC EC Fe MDL 20.0 40.0 30.0 20.0 10.0 10.0 400.0 200.0 1.4 EF 6096.89 3818.99 4231.46 321.47 953.55 351.99 6377.16 1255.09 288.91 Species K Ca Zn Pb Mn Cu As Cr Ni MDL 4.2 1.6 0.41 0.39 0.51 0.48 0.20 0.52 0.4 EF 284.43 120.72 78.19 16.31 21.79 10.92 2.24 4.15 3.13

[0073] The chemical components include water-soluble ions, carbon components, and metal elements; among them, water-soluble ions include nitrate ions (NO3). - ), sulfate ions (SO4) 2- ), ammonium ions (NH4) + ), potassium ions (K) + ), magnesium ions (Mg 2+ ) and calcium ions (Ca 2 + The carbon component includes elemental carbon (EC) and organic carbon (OC); the metal elements include iron (Fe), potassium (K), calcium (Ca), zinc (Zn), lead (Pb), manganese (Mn), copper (Cu), arsenic (As), chromium (Cr), and nickel (Ni).

[0074] ② Source apportionment based on the traditional particulate matter source apportionment model (PMF)

[0075] PM2.5 per hour was collected at the XX branch station in XX District, XX City. 2.5 The chemical component mass concentration data and the calculated uncertainty data are substituted into a positive definite matrix factorization model for calculation to obtain the ratio of the mass concentration of a single chemical component to the total mass concentration of chemical components per hour. This ratio is then used to determine the PM2.5 concentration. 2.5 The source, ultimately obtained from PM 2.5 The source contribution rate.

[0076] The establishment of meteorological samples and air pollutant emission samples is based on hourly PM2.5. 2.5 The data collection point for the chemical component mass concentration is designated as center point A (XX substation). With the entire city of XX as the target grid, XX city is divided into 9 km × 9 km grids, as follows: Figure 2 As shown: The area is divided into collection points B1, B2, ... B1, with a range of 9 km x 9 km. n .

[0077] Because the monitoring equipment at each collection point can monitor different types of data, some collection points have multiple devices that can simultaneously obtain hourly meteorological data and hourly air pollutant emission data for that point, while others have only a single device that can only monitor hourly meteorological data or only hourly air pollutant emission data. The meteorological data includes wind speed, wind direction, temperature, humidity, and pressure data; the air pollutant data includes SO2 data, NO... x Data, PM 10 Data, PM 2.5 Data, BC data, OC data, particulate matter composition data.

[0078] Therefore, hourly meteorological data and hourly air pollutant data were collected from the monitoring stations of XX College, XX Primary School, XX District Disabled Persons' Federation, and XX Housing and Construction Bureau throughout January 2023 (these stations fall within the designated area of ​​Collection Point B after the division into collection points). Subsequently, the meteorological and air pollutant data from the different collection points were cleaned and subjected to Latin hypercube sampling to obtain meteorological and air pollutant dataset samples.

[0079] ③ Development of a source resolution simulation system based on machine learning

[0080] PM per hour 2.5 The chemical component mass concentration data was used as the chemical component sample, the calculated source contribution rate was used as the source contribution rate data sample, and meteorological samples and atmospheric pollutant dataset samples were used to form a multi-source dataset. The multi-source dataset was divided into training and testing datasets, and substituted into the Extreme Gradient Boosting (XGBoost) model for simulation training and testing. The hourly meteorological data and hourly atmospheric pollutant emission data of a sampling point in XX City in January 2023 were input and substituted into the ML-PMF source apportionment simulation model to predict the PM2.5 concentration at that sampling point in January 2023. 2.5 Source analysis results.

[0081] Comparison between predicted and actual observations from a machine learning-based fine particulate matter source apportionment model. Figure 3 As shown, the results indicate that the overall error between the model-based predicted values ​​and the actual values ​​is ≤ ±10%, while the error in the mean is ≤ ±5%. Therefore, the fine particulate matter source apportionment model based on machine learning in this study can effectively predict PM2.5 levels in XX city. 2.5 Source analysis.

[0082] like Figure 4As shown, the only particulate matter component monitoring station in XX City is the XX sub-station (i.e., point A), which observed the mass concentrations of PM2.5 and its chemical components in January 2023. Based on the traditional PMF source apportionment method, the PM2.5 concentration in XX City... 2.5 The pollution source contributions were as follows: secondary sources contributed 41%, vehicle emissions 21%, coal combustion 18%, biomass combustion 9%, dust 6%, and industrial sources 5%. The fine particulate matter source apportionment results based on machine learning were: secondary sources 34%, vehicle emissions 20%, coal combustion 24%, biomass combustion 9%, dust 7%, and industrial sources 6%. Comparing the source apportionment results calculated by the two methods, it can be seen that the machine learning-based source apportionment model performs better in apportioning biomass combustion, vehicle emissions, industrial sources, and dust sources, but shows some error in apportioning secondary and coal combustion sources. This is because secondary and coal combustion sources contain many correlated components, such as sulfates, elemental carbon, and organic carbon.

[0083] ④ Model Validation

[0084] The mean equation difference (MSE), root mean square error (RMSE), mean absolute error (MAE), and goodness of fit (R²) of the training and test datasets of the above machine learning-based particulate matter source apportionment model are analyzed. 2 The error analysis is shown in Table 2. The results show that the machine learning-based particulate matter source apportionment model (ML-PMF) performs excellently in both the training and test datasets. Specifically, for the test dataset, the goodness of fit for different sources ranges from 0.88 to 0.95, indicating a good correlation between the predicted and actual values, and a small RMSE error.

[0085] Table 2. Error Analysis of the ML-PMF Model

[0086]

[0087] ⑤ Model usage

[0088] Meteorological data and air pollutant emission data from other collection points in the vicinity of Collection Center A (XX sub-site) in January 2024 were substituted into the PMF source apportionment simulation model to obtain the PM2.5 concentration at that collection point in January 2024. 2.5 Source analysis results, such as Figure 5 As shown.

Claims

1. A machine learning-based PM 2.5 A method for analyzing particulate matter sources, characterized in that, include: S1. Data collection unit time PM 2.5 Given the mass concentration data of the chemical components, calculate the uncertainty data corresponding to the mass concentration data of each chemical component. The uncertainty data is calculated as follows: the standard accuracy of the instrument used to monitor the mass concentration data of chemical components is obtained, and the PM2.5 concentration per unit time is collected. 2.5 Substitute the mass concentration data of the chemical components into equation (1) to calculate the uncertainty data corresponding to the mass concentration data of each chemical component; （1）； In the formula, EF This is the error ratio; A PM as a unit of time 2.5 Chemical component mass concentration data, μg / m 3 ; MDL To determine the detection limit of instruments used for monitoring the mass concentration of chemical components; B For uncertainty data, μg / m 3 ; The chemical components include water-soluble ions, carbon components, and metal elements; The water-soluble ions include nitrate ions, sulfate ions, ammonium ions, magnesium ions, potassium ions, and calcium ions; the carbon components include organic carbon and elemental carbon; and the metallic elements include mercury, iron, calcium, aluminum, manganese, chromium, and lead. S2. The PM per unit time collected by S1 2.5 The chemical component mass concentration data and the calculated uncertainty data are substituted into the positive definite matrix factorization model for calculation to obtain the ratio of the mass concentration of a single chemical component to the total mass concentration of chemical components per unit time. This ratio is then used to determine the PM per unit time. 2.5 The source, ultimately obtained from PM 2.5 The source contribution rate; S3. In terms of unit time PM 2.5 The data collection point for the chemical component mass concentration is center A, at PM 2.5 During the data collection period for the mass concentration of chemical components, the collection points B1, B2, and B3 near the collection center A were included. n The corresponding meteorological data per unit time and atmospheric pollutant emission data per unit time, respectively, PM 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same time unit; then, meteorological samples and air pollutant emission samples were obtained by sampling the meteorological data and air pollutant emission data from different collection points. Meteorological data includes wind speed, wind direction, temperature, humidity, and pressure data; air pollutant emission data includes SO2 and NO. x Data, PM 10 Data, PM 2.5 Data, including BC data, OC data, NH3 data, and NMVOCs data; the sampling method for the collection points was Latin hypercube sampling; In step S3, a circle with center A as the center and a radius of 50 kilometers is drawn to define the area. The area is then divided into collection points B1, B2, and B3, with a range of 9 kilometers × 9 kilometers. n ; S4. PM collected by S1 2.5 The chemical component mass concentration data is used as the chemical component sample, the source contribution rate calculated by S2 is used as the source contribution rate sample, and the meteorological sample and atmospheric pollutant emission sample obtained by S3 are used as the source contribution rate sample. A total of four samples are used to form a multi-source dataset. The multi-source dataset is used for simulation training to obtain the PMF source apportionment simulation model. The simulation training method is as follows: input the collection point B near the center A. x Meteorological data and atmospheric pollutant emissions per unit time during a specific period, as well as PM2.5 per unit time at Center A. 2.5 Chemical component mass concentration data, PM 2.5 The data on chemical component mass concentration, meteorological data, and atmospheric pollutant emission data were collected at the same unit time and time; PM 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data are substituted into the PMF source apportionment simulation model of S4 to predict the data collection point B. x PM during that period 2.5 Source contribution rate, relative to collection point B x The actual PM during this period 2.5 Compare source contribution rates; if the predicted PM 2.5 Source contribution rate and actual PM 2.5 If the source contribution rates differ significantly, adjust the parameters of the extreme gradient boosting model and return to training until the predicted PM is reached. 2.5 Source contribution rate and actual PM 2.5 Consistent source contribution rates; S5. Collection point B near collection center A m Meteorological data and atmospheric pollutant emissions per unit time during a certain period, and PM2.5 per unit time for Center A during that period. 2.5 Chemical component mass concentration data, PM 2.5 The chemical component mass concentration data, meteorological data, and air pollutant emission data were collected at the same unit time; the collected PM2.5 data were then processed. 2.5 The chemical component mass concentration data, meteorological data, and atmospheric pollutant emission data were substituted into the PMF source apportionment simulation model of S4 to obtain the data from collection point B. m PM during that period 2.5 The source contribution rate; Among them, collection point B n n = 2, 3, 4, 5..., a, where a is a positive integer greater than 2; Collection point B x x is a positive integer ≥ 1; Collection point B m m is a positive integer ≥ 1.

2. The analytical method as described in claim 1, characterized in that, In S1, PM is collected per unit time. 2.5 After obtaining the chemical component mass concentration data, the data is first preprocessed to remove impurities, and then the source data format is unified.

3. The analytical method as described in claim 1, characterized in that, In step S3, when each collection point has two or more monitoring stations for acquiring meteorological data or atmospheric pollutant emission data per unit time, the meteorological data per unit time of each station is averaged to obtain the meteorological data per unit time of that collection point, and the atmospheric pollutant emission data per unit time of each station is averaged to obtain the atmospheric pollutant emission data per unit time of that collection point.

4. A Machine Learning-Based PM 2.5 A particulate matter source analysis system, characterized in that, For use in conjunction with the machine learning-based PM as described in any one of claims 1 to 3 2.5 Methods for analyzing particulate matter sources include: The data acquisition module is used to acquire samples of chemical component mass concentration data and PM. 2.5 Source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples; Among them, PM 2.5 Source contribution rate data samples were obtained by calculation using chemical component mass concentration data samples and uncertainty data; The simulation training module is used to process the chemical component mass concentration data samples and PM data acquired by the data acquisition module. 2.5 Simulation training was conducted using source contribution rate data samples, meteorological data samples, and atmospheric pollutant emission data samples to obtain a PMF source apportionment simulation model; The analysis module uses the PMF source analysis simulation model to analyze PM. 2.5 The particulate matter source was analyzed.

5. A Machine Learning-Based PM 2.5 The particulate matter source analysis apparatus is characterized in that, include: Memory, used to store computer programs; Processor, configured to execute the computer program to implement the machine learning-based PM as described in any one of claims 1 to 3 2.5 Methods for analyzing particulate matter sources.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the machine learning-based PM as described in any one of claims 1 to 3. 2.5 Methods for analyzing particulate matter sources.

Images