Spatial refined source analysis method and system based on particulate matter chemical component information

Abstract

The invention discloses a spatial refined source analysis method and system based on particulate matter chemical component information, and relates to the technical field of atmospheric pollution source analysis. The method is realized through the following steps: acquiring the chemical component concentration of particulate matters at multiple points in a target area and performing diffusion standardization treatment; determining the number of pollution sources by using a diffusion standardization positive definite matrix factorization model; setting factor spectrum prior distribution of a Bayesian space multivariate receptor model based on an actually measured pollution source chemical component spectrum; constructing an integrated model fusing the two models, and inputting standardized data, prior distribution, point location information and spatial structure parameters; and finally operating the model to obtain the pollution source contribution value of any point in the region. According to the method, the problems of limited space coverage and high priori parameter dependence of a traditional method are solved, and high-resolution and high-accuracy pollution source space refined analysis can be realized. The system and the corresponding electronic equipment can realize automatic operation of the method.

Description

Technical Field

[0001] This invention relates to the field of atmospheric pollution source apportionment technology, and in particular to a spatially refined source apportionment method and system based on particulate matter chemical composition information. Background Technology

[0002] Atmospheric particulate matter pollution is one of the major environmental problems affecting urban air quality in my country. Accurately identifying and quantifying the sources of atmospheric particulate matter pollution (i.e., source apportionment) is a crucial prerequisite for formulating scientific and effective air pollution control strategies. Traditional source apportionment studies are mainly based on receptor models. This involves setting up one or more spatially representative observation points within the study area to monitor and obtain information on the chemical composition of particulate matter (such as elements, ions, and carbon components). Receptor models, such as chemical mass balance methods and positive definite matrix factorization, are then used to analyze the contribution of each pollution source to that observation point. When a sufficient number of monitoring points are set up, comparing the analysis results at each point can, to some extent, reflect the spatial variation characteristics of source contributions.

[0003] However, traditional methods have significant limitations. First, the number of monitoring points that can be deployed is limited by factors such as cost and geographical conditions, and full spatial coverage is usually impossible. Therefore, for areas without monitoring points, the contribution of pollution sources can only be roughly inferred from the results of neighboring points, resulting in insufficient spatial refinement. Second, traditional receptor models (such as PMF) typically analyze or simply merge data from each point when processing multi-point data, failing to fully utilize the spatial correlation information between multiple points and failing to effectively integrate prior knowledge about pollution source characteristics (such as measured source composition spectra) into the model. This may lead to increased uncertainty and subjectivity in the analysis results.

[0004] To improve the spatial characterization capabilities of source apportionment, research has begun to explore spatialized receptor models. The Bayesian spatial multivariate receptor model is an advanced model capable of integrating multi-site observation data and prior knowledge, and utilizing spatial statistical methods to estimate the contributions of pollution sources at unmonitored sites. However, the performance of this model is highly dependent on the accurate setting of prior parameters, particularly the number of pollution sources, the chemical composition characteristics (factor spectra) of each source, and implicit spatial structure parameters. How to accurately and objectively determine these key prior parameters has become a core challenge restricting the widespread application of Bayesian spatial receptor models in practice and hindering the improvement of their apportionment accuracy.

[0005] Therefore, there is an urgent need for a new technical solution that can organically integrate traditional factor decomposition methods with advanced Bayesian spatial models and systematically solve the problem of prior parameter setting, thereby achieving high-precision, high-spatial-resolution detailed analysis of pollution sources over large-scale regions. Summary of the Invention

[0006] This invention aims to overcome the shortcomings of existing technologies and provide a spatially refined source apportionment method and system based on particulate matter chemical composition information. By constructing an integrated method of "positive definite matrix factorization-Bayesian spatial multivariate receptor model," this invention effectively couples prior information such as multi-site observation data and measured pollution source chemical composition spectra, and provides a systematic method for determining key prior parameters (number of pollution sources, source chemical composition, implicit spatial structure). This enables the estimation of pollution source contributions at any location within the entire target area, achieving truly spatially refined source apportionment.

[0007] To solve the above-mentioned technical problems, the technical solution proposed in this application is as follows: This invention provides a spatially refined source apportionment method based on particulate matter chemical composition information, comprising the following steps: Acquire particulate matter chemical component concentration data from multiple monitoring points within the target area, and perform diffusion standardization processing on the concentration data to obtain standardized concentration data; Based on standardized concentration data, the number of pollution sources in the target area is determined using a diffusion-standardized positive definite matrix factor decomposition model. Based on the measured chemical composition spectrum of pollution sources and the determined number of pollution sources, the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model is determined. An integrated model was constructed, which combines a diffusion-normalized positive definite matrix factorization model with a Bayesian spatial multivariate receptor model. The inputs to the integrated model are standardized concentration data, prior distribution of factor spectrum chemical composition, geographical location information of monitoring points, pre-set spatial location information of potential processes, and latitude and longitude of the points to be predicted. By running the integrated model, the pollution source contribution value of any point to be predicted within the target area can be obtained, thus achieving spatially refined source apportionment.

[0008] Furthermore, the specific steps include: The chemical component concentration data from multiple monitoring sites are arranged end-to-end to form a long sequence of receptor data. By running the diffusion-normalized positive definite matrix factorization model with different preset factor numbers, multiple candidate solutions were obtained. Based on the preset evaluation criteria, an optimal solution is selected from multiple candidate solutions, and the number of factors corresponding to the optimal solution is determined as the number of pollution sources in the target area. The evaluation criteria include: the ratio of the model's robust objective function value to the true objective function value, the ratio of the model's robust objective function value to the expected objective function value, the results of permutation perturbation analysis, the results of bootstrapping, the coefficient of determination between observed concentration and model predicted concentration, and the interpretability of the factor spectrum.

[0009] Furthermore, the diffusion-normalized positive definite matrix factorization model reduces the interference of meteorological factors on observation data by introducing a ventilation coefficient.

[0010] Furthermore, the prior distribution of factor spectrum chemical composition determined based on measured pollution source chemical composition spectra includes: Identify the main chemical components corresponding to different pollution sources as labeling components; Based on the measured chemical composition spectrum of the pollution source, the contribution values ​​of non-major chemical components are pre-set to zero or close to zero in the prior distribution of chemical composition in the factor spectrum.

[0011] Furthermore, the spatial location information of the potential process is set using a Gaussian kernel function, which includes setting the number of potential process spatial locations and the size parameter of the kernel function. The size parameter of the kernel function is determined based on the distance between adjacent potential process locations.

[0012] Furthermore, the number of potential process spatial locations is determined through sensitivity analysis, which includes testing the impact of different numbers of potential process spatial locations on the model source resolution results and prediction accuracy.

[0013] Furthermore, the input also includes a pre-defined zero value in the source contribution matrix of the Bayesian spatial multivariate receptor model, where the zero value indicates that the contribution of a specific pollution source is zero within a specific time period.

[0014] Furthermore, particulate matter is defined as PM2.5 or PM10.

[0015] On the other hand, this application also claims protection for a system for implementing the above-described spatially refined source apportionment method based on particulate matter chemical composition information, comprising: The data acquisition and processing module is used to acquire particulate matter chemical component concentration data from multiple monitoring points within the target area, and to perform diffusion standardization processing on the concentration data to obtain standardized concentration data. The pollution source number determination module is used to determine the number of pollution sources in the target area based on standardized concentration data and using a diffusion-standardized positive definite matrix factor decomposition model. The prior distribution setting module is used to determine the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model based on the measured chemical composition spectrum of the pollution sources and the determined number of pollution sources. The integrated model construction and input module is used to construct the integrated model, which combines the diffusion-normalized positive definite matrix factorization model with the Bayesian spatial multivariate receptor model; and takes the normalized concentration data, the prior distribution of the chemical composition of the factor spectrum, the geographical location information of the monitoring points, the preset potential process spatial location information, and the latitude and longitude of the points to be predicted as inputs to the integrated model. The model solving and output module is used to run the integrated model, solve for the pollution source contribution value of any point to be predicted in the target area, and output the spatially refined source apportionment results.

[0016] On the other hand, this application also claims protection for an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above methods.

[0017] Compared with the prior art, the present invention achieves the following beneficial technical effects: This invention achieves a breakthrough in pollution source analysis, moving from discrete points to a continuous spatial surface, by integrating diffusion-normalized factor decomposition and Bayesian spatial modeling. Its innovation lies in the systematic integration of multi-point observation data, measured pollution source composition spectra, and spatial correlations to construct a refined analytical method capable of extrapolating and predicting the contribution of any unmonitored point source. This significantly improves spatial resolution and the objectivity and accuracy of the analytical results, providing a reliable spatial decision-making basis for precise pollution control. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a spatially refined source analysis method based on particulate matter chemical composition information provided in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the measured chemical composition of pollution sources in a certain area from 2015 to 2017, used in an embodiment of the present invention. In this diagram, (a) to (e) represent the chemical composition of dust sources, mobile sources, industrial sources, biomass sources and coal-fired sources, respectively.

[0021] Figure 3 The structural block diagram of the spatially refined source apportionment system based on particulate matter chemical composition information provided in the embodiments of the present invention is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: Spatial Refinement Source Apportionment Method Based on Particulate Chemical Composition Information like Figure 1 As shown, the method of the present invention mainly includes the following steps: S1: Obtain particulate matter chemical component concentration data from multiple monitoring points within the target area, and perform diffusion standardization processing on the concentration data to obtain standardized concentration data; This step involves data preparation and preprocessing. Specifically, 19 ambient air monitoring stations are set up and operated within the target area (a specific region in this example). PM2.5 samples are collected at these stations, and their chemical component concentrations are analyzed. The analyzed components include, but are not limited to: elements (Al, As, Ca, Cu, Fe, Mg, Mn, Pb, Si, Zn), carbon components (organic carbon OC, elemental carbon EC), and water-soluble ions (Cl). - NO 3- SO4 2- Na + NH 4+ K + A total of 18 chemical components were identified, including [list of components]. Daily average concentration data for each location were collected over a period of time (e.g., 437 valid sample days).

[0024] Because meteorological conditions (such as wind speed and atmospheric diffusion capacity) significantly affect observed particulate matter concentrations, diffusion normalization is necessary to more accurately reflect the chemical characteristics of emission sources. Specifically, a ventilation coefficient is introduced to normalize the concentration of each component in each sample. Detailed principles of diffusion normalization can be found in relevant literature; its core principle is to eliminate the influence of meteorological disturbances on receptor data, obtaining "normalized concentration data" that better represents the characteristics of the source components. This process also lays the foundation for subsequent use of diffusion-normalized PMF models.

[0025] S2: Based on the standardized concentration data, the number of pollution sources in the target area is determined using a diffusion-standardized positive definite matrix factorization model; The purpose of this step is to determine the number of pollution sources. The goal of this step is to objectively and accurately determine the number of pollution source categories (i.e., the number of factors p) that play a major role in the study area. This is the foundation for the subsequent construction of the Bayesian model.

[0026] In practice, determining the number of pollution sources using the diffusion-normalized positive definite matrix factorization model includes the following operations: First, the standardized concentration data of 19 monitoring sites, 437 sample days, and 18 chemical components were arranged end-to-end in the order of "all sample day data of site 1, all sample day data of site 2, ..." to form a long two-dimensional matrix (receptor data matrix). This processing method treats the multi-site data as a whole for factor analysis.

[0027] Next, the diffusion-normalized PMF model (DN-PMF) was used, which reduces the interference of meteorological factors on the observed data by incorporating a ventilation coefficient. In practice, the model was run for 5 to 9 different factor numbers (i.e., p=5, 6, 7, 8, 9). For each factor number, 20 independent calculations were performed, and the solution with the lowest objective function value (Q) was selected as the candidate solution for that factor number.

[0028] Then, based on a comprehensive evaluation criterion, an optimal solution is selected from the multiple candidate solutions, and the number of factors corresponding to this optimal solution is determined as the number of pollution sources within the target area. The evaluation criterion includes: the robust objective function value of the model (Q...). robust ) and the true objective function value (Q) true The ratio of ) and the robust objective function value of the model (Q) robust ) and the expected objective function value (Q) expected The ratio of the observed concentration to the model-predicted concentration, the results of permutation perturbation analysis (DISP), the results of bootstrap analysis, and the coefficient of determination (r) between the observed concentration and the model-predicted concentration. 2 ), and the interpretability of the factor spectrum.

[0029] Referring to the summary results of this embodiment shown in Table 1, the candidate solutions are compared. For example, when the number of factors is 8, Q... robust / Q true Q is 0.992 (close to 1 and ≤ 1). robust / Q expected The value was 0.89; Bootstrap results showed that the mapping rate for all 8 factors was 20 (i.e., 100%); DISP results were stable; most chemical components (such as Ca, Fe, OC, EC, NO) were also supported. 3- SO4 2- NH 4+ (etc.) of r 2 A value of 0.9 or higher indicates a good fit; and the eight extracted factors clearly correspond to typical local pollution sources such as secondary nitrates, secondary sulfates, vehicle sources, coal combustion sources, industrial sources, biomass combustion sources, dust sources, and sea salt sources. Taking all factors into consideration, the eight-factor scheme is selected as the optimal scheme, thus determining the number of main categories of PM2.5 pollution sources in the target area (a certain location) to be eight.

[0030] Table 1. Summary of PMF parameter settings and model results in the ensemble method of "Positive Definite Matrix Factorization-Bayesian Spatial Multivariate Receptor Model".

[0031] S3: Based on the measured chemical composition spectrum of pollution sources and the number of pollution sources determined in step S2, determine the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model. The purpose of this step is to define the prior distribution of the chemical composition of the factor spectrum, which aims to provide the Bayesian spatial multivariate receptor model with prior knowledge about the chemical composition (i.e., factor spectrum) of each pollution source, so as to constrain the model solution and improve the rationality and accuracy of the solution.

[0032] The prior distribution of factor spectral chemical composition determined based on measured pollution source chemical composition spectra includes the following: First, obtain measured chemical composition spectra of pollution sources from the target area during a period similar to the receptor sampling period (e.g., 2015-2017). These spectra can be obtained, for example, from relevant research institutions or source libraries. Figure 2 The measured source spectrum of a certain location is schematically shown, where (a) to (e) represent the chemical composition spectra of dust source, mobile source, industrial source, biomass source and coal source, respectively.

[0033] Secondly, based on these measured source spectra, the main chemical components corresponding to different pollution sources were determined as labeling components. Table 2 provides an example: the main labeling component for secondary nitrates is NO3. - and NH4 + Motor vehicle sources are EC and OC; coal-fired sources are OC, EC and SO4. 2- Industrial sources are OC and Cl. - Fe, Ca; biomass combustion source is K + Cl - OC, EC; dust sources are Ca, Si, Al, Fe; sea salt source is Na. + Cl - .

[0034] Table 2. Major chemical components (identified components) of different particulate matter pollution sources

[0035] Then, combining the number of pollution sources (8) determined in step S2 and the resolved factor spectrum information, a prior distribution of the chemical composition is set for each factor in the Bayesian model. Specifically, for a factor (e.g., one identified as a "dust source"), the prior distribution of its identifying components (e.g., Ca, Si, Al, Fe) is set to have a high probability. Based on the measured chemical composition spectrum of the pollution source, the contribution value of non-major chemical components in the spectrum (e.g., for dust sources, Pb is usually a minor component) is pre-set to zero or near-zero in the prior distribution of the factor spectrum's chemical composition. This "pre-zero" operation serves as an identifiability condition, helping the model to more clearly distinguish different pollution sources.

[0036] S4: Construct an integrated model that combines the diffusion-normalized positive definite matrix factorization model with the Bayesian spatial multivariate receptor model; the standardized concentration data, the prior distribution of the factor spectrum chemical composition, the geographical location information of the monitoring points, the preset potential process spatial location information, and the latitude and longitude of the points to be predicted are used as inputs to the integrated model; The purpose of this step is to build an integrated model and set up the inputs. This step integrates the results of the previous steps to build a complete analytical framework. The input data required for the Bayesian spatial multivariate receptor model includes: Observational data (Y): Concentration data of 18 chemical components from 19 monitoring sites, after diffusion standardization.

[0037] Prior distribution (θ): includes the number of factors p=8, and the prior distribution of the factor spectrum chemical composition set by S3.

[0038] Monitoring point coordinates (S) obs ): Latitude and longitude information of 19 monitoring points.

[0039] Potential process spatial location (S η ) and the coordinates of the point to be predicted (S) pred ).

[0040] The "potential process spatial location information" is defined using a Gaussian kernel function, including setting the number (L) of potential process spatial locations and the size parameter (σ) of the kernel function. k The size parameter of the kernel function is determined based on the distance between adjacent potential process locations. This embodiment tested three cases: L=9, L=16, and L=25, with corresponding σ... k The values ​​were 0.5596, 0.3730, and 0.2798, respectively. Through sensitivity analysis (testing the impact of different numbers of potential process spatial locations on the model source resolution results and prediction accuracy), no significant differences were found when L=9, 16, and 25. To save computational resources and prevent overfitting, L=9 was ultimately chosen.

[0041] In addition, in the input settings of the model, a zero value can be selectively preset in the source contribution matrix of the Bayesian spatial multivariate receptor model. The zero value indicates that the contribution of a specific pollution source (e.g., coal combustion during the non-heating season) is zero in a specific time period.

[0042] In a certain location S5: the integrated model is run to solve for the pollution source contribution value of any point to be predicted within the target area, thereby achieving spatially refined source analysis.

[0043] The purpose of this step is to run the model and output the results, specifically the constructed ensemble model (the core of which is the Bayesian spatial multivariate receptor model). The model is solved using the Markov chain Monte Carlo method, and the posterior distribution of all parameters is obtained after iteration.

[0044] The output results include: time series of source contributions from all monitoring points, and contribution values ​​of each pollution source from all preset unmonitored points.

[0045] The particulate matter referred to is PM2.5. This method is also applicable to PM10 particulate matter, and will not be elaborated here.

[0046] Example 2: Implementing the above-mentioned spatially refined source apportionment system based on particulate matter chemical composition information Based on the same inventive concept, this invention also provides a system for implementing the above-mentioned spatially refined source apportionment method based on particulate matter chemical composition information, see [link to relevant documentation]. Figure 3 The system includes: The data acquisition and processing module is used to acquire particulate matter chemical component concentration data from multiple monitoring points within the target area, and to perform diffusion standardization processing on the concentration data to obtain standardized concentration data. The data acquisition and processing module is used to execute S1, acquire concentration data and perform diffusion standardization processing.

[0047] The pollution source number determination module is used to determine the number of pollution sources in the target area based on the standardized concentration data and using a diffusion-standardized positive definite matrix factorization model. The pollution source number determination module is used to execute S2. Its operating logic includes concatenating the data from multiple locations, running DN-PMF models with different numbers of factors, and selecting the optimal number of factors according to the comprehensive evaluation criteria.

[0048] The prior distribution setting module is used to determine the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model based on the measured chemical composition spectrum of the pollution sources and the determined number of pollution sources. The prior distribution setting module is used to execute S3, which determines the identifier component and sets the factor spectrum prior distribution based on the measured source spectrum.

[0049] An integrated model construction and input module is used to construct an integrated model that combines the diffusion-normalized positive definite matrix factorization model with the Bayesian spatial multivariate receptor model. The module takes the normalized concentration data, the prior distribution of the factor spectrum chemical composition, the geographical location information of the monitoring points, the preset potential process spatial location information, and the latitude and longitude of the points to be predicted as inputs to the integrated model. The integrated model building and input module is used to execute S4, build the model framework and integrate all input data, including setting the potential process space location information based on the Gaussian kernel function.

[0050] The model solving and output module is used to run the integrated model, solve for the pollution source contribution value of any point to be predicted in the target area, and output the spatially refined source apportionment results. The model solver and output module is used to execute S5, run the model solver, and output the source contribution results for any point.

[0051] The various modules of the system work together to implement the above method.

[0052] Example 3: Electronic Equipment Based on the same inventive concept, the present invention also provides an electronic device for implementing the above-mentioned spatial fine-grained source apportionment method based on particulate matter chemical composition information, the structure of which supports the operation of the spatial fine-grained source apportionment system based on particulate matter chemical composition information.

[0053] The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it controls the electronic device to implement the steps of the spatially refined source analysis method based on particulate matter chemical composition information as described in Example 1. Specifically, after the computer program is loaded and executed, the electronic device can automatically or according to instructions complete a series of operations from S1 to S5. The electronic device can be a server, workstation, personal computer, or dedicated environmental data analysis equipment.

[0054] In summary, this invention provides a complete, systematic, and highly operable spatial refined source apportionment solution through multiple embodiments such as methods, systems, and electronic devices. It can significantly improve the spatial characterization capability and reliability of atmospheric particulate matter source apportionment work, and has important scientific value and application prospects.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A spatially refined source apportionment method based on particulate matter chemical composition information, characterized in that, Includes the following steps: S1: Obtain particulate matter chemical component concentration data from multiple monitoring points within the target area, and perform diffusion standardization processing on the concentration data to obtain standardized concentration data; S2: Based on the standardized concentration data, the number of pollution sources in the target area is determined using a diffusion-standardized positive definite matrix factorization model; S3: Based on the measured chemical composition spectrum of pollution sources and the number of pollution sources determined in step S2, determine the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model. S4: Construct an integrated model that combines the diffusion-normalized positive definite matrix factorization model with the Bayesian spatial multivariate receptor model; the standardized concentration data, the prior distribution of the factor spectrum chemical composition, the geographical location information of the monitoring points, the preset potential process spatial location information, and the latitude and longitude of the points to be predicted are used as inputs to the integrated model; S5: Run the integrated model to solve for the pollution source contribution value of any point to be predicted within the target area, thereby achieving spatially refined source analysis.

2. The method according to claim 1, characterized in that, Step S2 specifically includes: S21: Arrange the chemical component concentration data from multiple monitoring points end to end to form a long sequence of receptor data; S22: Run the diffusion-normalized positive definite matrix factorization model with different preset factor numbers to solve the problem and obtain multiple candidate solutions; S23: Based on the preset evaluation criteria, select an optimal solution from the multiple candidate solutions, and determine the number of factors corresponding to the optimal solution as the number of pollution sources in the target area; The evaluation criteria include: the ratio of the model's robust objective function value to the true objective function value, the ratio of the model's robust objective function value to the expected objective function value, the results of permutation perturbation analysis, the results of bootstrapping, the coefficient of determination between observed concentration and model predicted concentration, and the interpretability of the factor spectrum.

3. The method according to claim 2, characterized in that, The diffusion-normalized positive definite matrix factorization model reduces the interference of meteorological factors on observation data by introducing a ventilation coefficient.

4. The method according to claim 1, characterized in that, In step S3, determining the prior distribution of factor spectrum chemical composition based on measured pollution source chemical composition spectra includes: Identify the main chemical components corresponding to different pollution sources as labeling components; Based on the measured chemical composition spectrum of the pollution source, the contribution values ​​of non-major chemical components are pre-set to zero or close to zero in the prior distribution of the chemical composition of the factor spectrum.

5. The method according to claim 1, characterized in that, In step S4, the potential process spatial location information is set by a Gaussian kernel function, including setting the number of potential process spatial locations and the size parameter of the kernel function. The size parameter of the kernel function is determined based on the distance between adjacent potential process locations.

6. The method according to claim 5, characterized in that, The number of potential process spatial locations is determined through sensitivity analysis, which includes testing the impact of different numbers of potential process spatial locations on the model source resolution results and prediction accuracy.

7. The method according to claim 1, characterized in that, In step S4, the input also includes a zero value pre-set in the source contribution matrix of the Bayesian spatial multivariate receptor model, where the zero value indicates that the contribution of a specific pollution source is zero within a specific time period.

8. The method according to any one of claims 1 to 7, characterized in that, The particulate matter is PM2.5 or PM10.

9. A system for implementing the spatially refined source apportionment method based on particulate matter chemical composition information as described in claim 1, characterized in that, include: The data acquisition and processing module is used to acquire particulate matter chemical component concentration data from multiple monitoring points within the target area, and to perform diffusion standardization processing on the concentration data to obtain standardized concentration data. The pollution source number determination module is used to determine the number of pollution sources in the target area based on the standardized concentration data and using a diffusion-standardized positive definite matrix factorization model. The prior distribution setting module is used to determine the prior distribution of the factor spectrum chemical composition of the Bayesian spatial multivariate receptor model based on the measured chemical composition spectrum of pollution sources and the determined number of pollution sources. An integrated model construction and input module is used to construct an integrated model that combines the diffusion-normalized positive definite matrix factorization model with the Bayesian spatial multivariate receptor model. The module takes the normalized concentration data, the prior distribution of the factor spectrum chemical composition, the geographical location information of the monitoring points, the preset potential process spatial location information, and the latitude and longitude of the points to be predicted as inputs to the integrated model. The model solving and output module is used to run the integrated model, solve for the pollution source contribution value of any point to be predicted in the target area, and output the spatially refined source apportionment results.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images