A dynamic tracking method of air pollution sources

Abstract

The application discloses an air pollution source dynamic tracking method, relates to the technical field of pollution source tracking, and collects pollution concentration grid data set, UAV sniffing set and meteorological field data set, carries out wind field reconstruction and pollution transportation prediction, generates candidate source assumption, carries out fingerprint matching and weight calculation, carries out sensor adaptive scheduling, carries out Bayes inversion fusion and result output, and the application generates unified pollution concentration and meteorological data through multi-source data collection and three-dimensional flow field reconstruction, realizes second-level wind field calculation by adopting dimension reduction CFD model, generates candidate source assumption by combining Lagrange backtracking and Euler transportation, and forms a weight matrix by utilizing two-way matching of UAV mass spectrum fingerprints. Through confidence level sorting, the second cruise of the unmanned aerial vehicle and the densification sampling of the ground node are driven, and the sampling density is adaptively optimized. Multi-source data are fused by adopting Bayes joint likelihood and Markov chain, dynamic source coordinates and emission rates are output, and a closed-loop iterative air pollution source tracking system is realized.

Description

Technical Field

[0001] This invention relates to the field of pollution source tracing technology, and more particularly to a method for dynamic tracing of air pollution sources. Background Technology

[0002] With the acceleration of urbanization and industrialization, air pollution events in urban-industrial mixed zones exhibit characteristics of multi-source, instantaneous, and spatial heterogeneous nature. Existing air quality monitoring and pollution tracking technologies mostly rely on fixed stations and average wind field models. Under complex meteorological conditions such as uncertain number of pollution sources, rapid changes in wind direction, and significant turbulence, it is difficult to achieve real-time location of pollution sources and inversion of emission intensity. Traditional source apportionment methods based on static sampling require long-term data accumulation and have long response cycles, which cannot meet the requirements of environmental regulation for second-level response and dynamic tracking. Especially in scenarios involving volatile pollutants such as VOCs, existing technologies lack a closed-loop processing mechanism that can simultaneously combine multi-source data, quickly reconstruct three-dimensional flow fields, and perform inversion calculations, making it difficult for regulatory authorities to promptly locate emission sources and intervene in the early stages of pollution diffusion.

[0003] Currently, Chinese patent application number CN202310243270.4 discloses a method for analyzing the sources of ambient air pollution. This method includes: acquiring pollution source component data within a first analysis period to form first pollution source component data; performing standard analysis processing to form standard analysis data; acquiring the standard analysis data and performing first feature deviation analysis processing to form first feature deviation analysis data; acquiring the standard analysis data and performing second feature deviation analysis processing to form second feature deviation analysis data; acquiring target pollution source component data; combining the standard analysis data with the second pollution source component data to perform feature state analysis to form feature state analysis data; and combining the standard analysis data, the first feature deviation analysis data, the second feature deviation analysis data, and the feature state analysis dataset to form a pollution source component feature dataset. This method can reasonably analyze pollution sources from a time perspective.

[0004] The relevant technologies are difficult to use in complex environments with multi-source pollution, rapidly changing wind fields, and significant turbulence to achieve real-time location of air pollution sources and dynamic inversion of emission rates. Summary of the Invention

[0005] The technical problem solved by this invention is that existing technologies are difficult to achieve real-time location of air pollution sources and dynamic inversion of emission rates in complex environments with multi-source pollution, rapidly changing wind fields, and significant turbulence.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A method for dynamically tracking air pollution sources includes the following steps:

[0008] Step S1: Collect pollution concentration raster dataset, UAV sniffing dataset, and meteorological field dataset;

[0009] Step S2, wind farm reconstruction and pollution transport prediction;

[0010] Step S3: Generate candidate source hypotheses. For each source coordinate in the candidate source hypothesis, call the Euler pollution transport solution module to simulate the pollutant concentration field in the three-dimensional flow field data file. Perform cell-by-cell difference calculation between the simulated concentration raster and the pollution concentration raster dataset according to the raster index. Output the difference score, error distribution matrix and time alignment label, and write them into the candidate source hypothesis library in descending order of difference score.

[0011] Step S4: Fingerprint matching and weight calculation;

[0012] Step S5, sensor adaptive scheduling;

[0013] Step S6: Bayesian inversion fusion and result output;

[0014] Step S4 includes the following sub-steps:

[0015] Step S401: Call the mass spectrometry fingerprint extraction module to perform baseline correction, noise threshold filtering and mass-charge ratio calibration on the full scan mass spectrometry data in the UAV sniffing set, record the normalized peak intensity array under the unified index, and attach the sampling time label and spatial coordinate field to generate a fingerprint feature set;

[0016] Step S402: Read the process emission fingerprint registered in the candidate source hypothesis library, perform cosine similarity calculation with the peak intensity array corresponding to the fingerprint feature set under the unified index, and write the similarity score into the cross cell of the source index and detection index of the matching weight matrix.

[0017] Step S5 includes the following sub-steps:

[0018] Step S501: Based on the matching weight matrix, sort the confidence values ​​of each candidate source, arrange the candidate source list from high to low confidence, and record the source coordinates, initial emission rate and corresponding confidence value of each candidate source.

[0019] Step S502: Based on the candidate source confidence ranking results and combined with the pollution concentration raster dataset, optimize the sampling frequency of each candidate source high confidence area, generate the UAV secondary cruise path, and generate the density sampling instruction in combination with the layout of ground monitoring nodes, optimize the sampling path and determine the density distribution of sampling nodes.

[0020] Step S503: Based on the optimized sampling path and the denser sampling instruction, execute the denser sampling task, collect pollution concentration data, and upload the sampling results to the data processing system in real time;

[0021] The sampling results are merged with the existing pollution concentration raster dataset, the pollution concentration values ​​of the raster cells are updated, and the updated data is synchronized to the pollution source dynamic tracking system.

[0022] Step S6 includes the following sub-steps:

[0023] Step S601: Construct a joint likelihood function and fuse the three-dimensional flow field reconstruction results, the pollution concentration raster dataset, and the intensified sampling results;

[0024] Step S602: Run the adaptive Markov chain Monte Carlo method to obtain the posterior distribution and output the dynamic source coordinates and emission rates;

[0025] Step S603: Generate a confidence assessment report and push the dynamic source coordinates, emission rates, and confidence intervals to the regulatory platform.

[0026] Preferably, step S1 includes the following sub-steps:

[0027] Step S101: Deploy a set of ground environmental monitoring nodes and continuously upload the pollution concentration raster dataset;

[0028] Step S102: Schedule the UAV sniffing queue to cruise in layers at a preset altitude to acquire UAV sniffing sets;

[0029] Step S103: Simultaneously record temperature, humidity, wind speed, wind direction, and boundary layer height, and aggregate them into a meteorological field dataset.

[0030] Preferably, step S2 includes the following sub-steps:

[0031] Step S201: Read the wind speed, wind direction, temperature, humidity and boundary layer height parameters from the meteorological field dataset, call the offline trained dimensionality reduction CFD proxy model, perform joint calculation of three-dimensional wind speed, turbulent kinetic energy and pressure gradient for the preset height range, and generate a three-dimensional flow field data file. The three-dimensional flow field data file includes velocity vector and turbulent energy distribution.

[0032] Step S202: Interpolate the real-time wind profile measurement sequence at one-second intervals and perform boundary layer constraint correction. Write the interpolation results into a wind profile grid with a unified time reference and align them with the fields of the three-dimensional flow field data file to generate a three-dimensional flow field reconstruction result.

[0033] Preferably, step S3 loads the velocity vector and turbulent energy distribution based on the three-dimensional flow field data file, sets terminal markers for the cells in the pollution concentration raster dataset whose concentration values ​​are not lower than a set threshold, calls the Lagrange particle trajectory backtracking module to integrate in the reverse direction along the velocity vector to the starting point of a preset time window, records the coordinates of the backtracking nodes, the arrival time and the corresponding concentration value, performs density clustering on the coordinates of the backtracking nodes, and outputs candidate source hypotheses, which include source coordinates, emission start time series and initial emission rate estimates;

[0034] Preferably, the dimension-reduced CFD proxy model used in step S201 is trained based on offline high-fidelity flow field samples.

[0035] Preferably, step S402 uses a bidirectional correlation matching strategy to cross-check the fingerprint feature set with the process emission fingerprint and calculate a similarity threshold.

[0036] Preferably, step S502 adjusts the sampling priority of the UAV track and ground nodes in real time based on the confidence ranking result so that the sampling density is increased by 100% or more in the area with the highest pollution concentration gradient.

[0037] The beneficial effects of this invention are as follows: This invention generates unified pollution concentration and meteorological data through multi-source data acquisition and three-dimensional flow field reconstruction. It employs a dimensionality-reduced CFD model to achieve second-level wind field estimation, combines Lagrange backtracking and Eulerian transport to generate candidate source hypotheses, and utilizes UAV mass spectrometry fingerprint bidirectional matching to form a weight matrix. It adaptively optimizes sampling density by driving secondary UAV patrols and ground node-based denser sampling through confidence ranking. Finally, it fuses multi-source data using Bayesian joint likelihood and Markov chains to output dynamic source coordinates and emission rates, realizing a closed-loop iterative air pollution source tracking system. Attached Figure Description

[0038] Figure 1 A flowchart illustrating the steps of a method for dynamically tracking air pollution sources according to an embodiment of the present invention;

[0039] Figure 2 This is a physical illustration of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0041] Example, refer to Figure 1 This paper provides a method for dynamic tracking of air pollution sources, including the following steps:

[0042] Step S1: Collect the pollution concentration raster dataset, UAV sniffing dataset, and meteorological field dataset.

[0043] Step S2, wind farm reconstruction and pollution transport prediction.

[0044] Step S3: Generate candidate source hypotheses.

[0045] Step S4: Fingerprint matching and weight calculation.

[0046] Step S5: Sensor adaptive scheduling.

[0047] Step S6: Bayesian inversion fusion and result output.

[0048] This invention generates unified pollution concentration and meteorological data through multi-source data acquisition and 3D flow field reconstruction. A dimensionality-reduced CFD model is used to achieve second-level wind field estimation. Candidate source hypotheses are generated by combining Lagrange backtracking and Eulerian transport, and a weight matrix is ​​formed using bidirectional matching of UAV mass spectrometry fingerprints. The sampling density is adaptively optimized by driving secondary UAV patrols and ground node-based denser sampling through confidence ranking. Multi-source data is fused using Bayesian joint likelihood and Markov chains to output dynamic source coordinates and emission rates, realizing a closed-loop iterative air pollution source tracking system.

[0049] Step S1 includes the following sub-steps:

[0050] Step S101: Deploy a set of ground environmental monitoring nodes and continuously upload the pollution concentration raster dataset.

[0051] Step S101 acquires high-resolution pollution concentration raster data within the ground area to provide a spatial reference benchmark for the spatiotemporal distribution of the concentration field and the location of pollution sources.

[0052] Step S102: Schedule the UAV sniffing queue to cruise in layers at a preset altitude to acquire UAV sniffing sets.

[0053] Step S102 establishes a pollutant distribution profile in the vertical direction and obtains airborne mass spectrometry fingerprint information, providing data support for multi-altitude concentration analysis and source feature identification.

[0054] Step S103: Simultaneously record temperature, humidity, wind speed, wind direction, and boundary layer height, and aggregate them into a meteorological field dataset.

[0055] Step S103 generates meteorological field parameters synchronized with the pollution concentration raster data, providing dynamic boundary conditions for three-dimensional wind field estimation and pollution transport path calculation.

[0056] Step S1 enables multi-dimensional synchronous acquisition of the spatial distribution of pollutants and meteorological driving parameters, forming a basic data set that can be used for subsequent flow field reconstruction and pollution source inversion.

[0057] Step S2 includes the following sub-steps:

[0058] Step S201: Read the wind speed, wind direction, temperature, humidity and boundary layer height parameters from the meteorological field dataset, call the offline trained dimensionality-reduced CFD proxy model, perform joint calculation of three-dimensional wind speed, turbulent kinetic energy and pressure gradient for the preset height range, and generate a three-dimensional flow field data file. The three-dimensional flow field data file includes velocity vector and turbulent energy distribution.

[0059] The dimension-reduced CFD proxy model used in step S201 is trained based on offline high-fidelity flow field samples.

[0060] Step S201 uses a dimensionality-reduced CFD proxy model to quickly generate a three-dimensional flow field data file containing velocity vectors and turbulent energy distribution, providing a refined flow field basis for pollutant transport calculations in complex turbulent environments.

[0061] Step S202: Interpolate the real-time wind profile measurement sequence at one-second intervals and perform boundary layer constraint correction. Write the interpolation results into a wind profile grid with a unified time reference and align them with the fields of the three-dimensional flow field data file to generate a three-dimensional flow field reconstruction result.

[0062] Step S202 standardizes the real-time wind profile sequence to a unified time reference and aligns it with the three-dimensional flow field data file to form a complete three-dimensional flow field reconstruction result for pollution transport prediction and trajectory inversion.

[0063] Step S2 enables high-precision reconstruction of the instantaneous three-dimensional wind field in the target area, providing dynamic meteorological driving conditions for pollutant transport path calculation and source inversion.

[0064] Step S3 includes the following sub-steps:

[0065] Step S301: Based on the three-dimensional flow field data file, load the velocity vector and turbulent energy distribution, set terminal markers for the cells in the pollution concentration raster data set whose concentration values ​​are not lower than a set threshold, call the Lagrange particle trajectory backtracking module to integrate in the reverse direction along the velocity vector to the starting point of the preset time window, record the coordinates of the backtracking nodes, the arrival time and the corresponding concentration value, and perform density clustering on the coordinates of the backtracking nodes to output candidate source hypotheses. The candidate source hypotheses include source coordinates, emission start time series and initial emission rate estimate.

[0066] Step S301 uses the velocity vector and turbulent energy distribution in the three-dimensional flow field data file to perform trajectory backtracking on the pollution concentration raster data, maps the raster cells above the concentration threshold to possible source locations, and generates candidate source hypotheses containing source coordinates, emission start time series and initial emission rate estimates, providing an initial solution space for pollution source spatial locking and time series reconstruction.

[0067] Step S302: For each source coordinate in the candidate source hypothesis, call the Euler pollution transport solution module to perform a forward simulation of the pollutant concentration field in the three-dimensional flow field data file. Perform a grid-by-grid difference calculation between the simulated concentration raster and the pollution concentration raster dataset according to the grid cell index, output the difference score, error distribution matrix and time alignment label, and write them into the candidate source hypothesis library in descending order of difference score.

[0068] Step S302 uses the Eulerian pollution transport solution module to perform a forward concentration field simulation for each candidate source and compares it cell by cell with the actual pollution concentration grid. It outputs the difference score, error distribution matrix and time alignment label. The results after difference ranking are written into the candidate source hypothesis library to achieve multi-dimensional screening and optimization of source location and emission time history, and provide an accurate set of candidate sources for subsequent inversion calculation.

[0069] Step S3 completes the preliminary estimation of the spatial location and emission time history of pollution sources. A candidate source hypothesis library is generated by combining reverse backtracking and forward simulation, providing basic data support for subsequent fingerprint matching and Bayesian inversion.

[0070] Step S4 includes the following sub-steps:

[0071] Step S401: Call the mass spectrometry fingerprint extraction module to perform baseline correction, noise threshold filtering and mass-charge ratio calibration on the full scan mass spectrometry data in the UAV sniffing set, record the normalized peak intensity array under the unified index, and attach the sampling time label and spatial coordinate field to generate a fingerprint feature set.

[0072] Step S401 performs baseline correction, noise threshold filtering, and mass-charge ratio calibration on the full-scan mass spectrometry data in the UAV sniffing set. After unifying the m / z index, a normalized peak intensity array is generated, and time and spatial fields are added to form a complete fingerprint feature set, providing standardized input for the chemical feature analysis of pollutants.

[0073] Step S402: Read the process emission fingerprints registered in the candidate source hypothesis library, perform cosine similarity calculation with the peak intensity array corresponding to the fingerprint feature set under the unified index, and write the similarity score into the cross cell of the source index and detection index of the matching weight matrix.

[0074] Step S402 uses a bidirectional correlation matching strategy to cross-check the fingerprint feature set with the process emission fingerprint and calculate the similarity threshold.

[0075] Step S402 performs bidirectional correlation matching between the process emission fingerprints in the candidate source hypothesis library and the fingerprint feature set. The weight values ​​are filled in the intersection cells of the source index and the detection index by calculating the cosine similarity. At the same time, the similarity threshold is calculated to screen stable matching relationships and generate a matching weight matrix. This provides quantitative support for the chemical feature verification of source localization and is used for confidence ranking calculation.

[0076] Step S4 verifies the correlation between candidate hypotheses of pollution sources and actual monitored chemical characteristics. A matching weight matrix is ​​generated through feature matching of mass spectrometry fingerprints, providing a quantitative basis for subsequent confidence ranking and sensor scheduling based on chemical composition.

[0077] Step S5 includes the following sub-steps:

[0078] Step S501: Based on the matching weight matrix, sort the confidence values ​​of each candidate source, arrange the candidate source list from high to low confidence, and record the source coordinates, initial emission rate and corresponding confidence value of each candidate source.

[0079] Step S501 converts the matching weight matrix output in step S4 into a quantized confidence score, performs global sorting on the candidate sources and outputs the sorted candidate source list, and binds the source coordinates, initial emission rate and corresponding confidence score value to establish a priority index system for scheduling optimization and subsequent Bayesian inversion.

[0080] Step S502: Based on the candidate source confidence ranking results and combined with the pollution concentration raster dataset, optimize the sampling frequency of each candidate source high confidence area, generate the UAV secondary cruise path, and generate the density increase sampling command in combination with the layout of ground monitoring nodes, optimize the sampling path and determine the density distribution of sampling nodes.

[0081] Step S502 adjusts the sampling priority of UAV tracks and ground nodes in real time based on the confidence ranking results so that the sampling density is increased by 100% or more in the area with the highest pollution concentration gradient.

[0082] Step S502, based on the candidate source confidence ranking results, calls the sampling optimization module to perform spatial gradient analysis on the pollution concentration raster data, determines high-confidence areas and generates the UAV secondary cruise path; at the same time, calculates the density increase sampling location and frequency according to the geographical distribution of ground monitoring nodes, outputs the optimized sampling path instruction containing UAV tracks and ground node priorities, realizes dynamic sampling density scheduling, and increases the sampling density of the area with the highest pollution concentration gradient by 100% or more.

[0083] Step S503: Based on the optimized sampling path and the denser sampling instruction, execute the denser sampling task, collect pollution concentration data, and upload the sampling results to the data processing system in real time.

[0084] The sampling results are merged with the existing pollution concentration raster dataset, the pollution concentration values ​​of the raster cells are updated, and the updated data is synchronized to the pollution source dynamic tracking system.

[0085] Step S503 executes the optimized UAV flight path and ground-based enhanced sampling instructions, collects multi-dimensional pollution concentration data and uploads it to the data processing system in real time, aligns and merges the newly sampled data with the existing pollution concentration raster data using spatiotemporal indexing, updates the raster cell concentration values ​​and synchronously writes the updated raster data into the pollution source dynamic tracking system, providing refined data input for real-time inversion calculation.

[0086] Step S5 achieves dynamic collaborative sampling between UAVs and ground nodes by matching the confidence ranking driven by the weight matrix and scheduling the sensors. This significantly improves the spatial resolution of high-confidence areas, forming real-time updated raster data of pollution concentration, and providing high-precision observation input for subsequent inversion.

[0087] Step S6 includes the following sub-steps:

[0088] Step S601: Construct a joint likelihood function to integrate the three-dimensional flow field reconstruction results, the pollution concentration raster dataset, and the intensified sampling results.

[0089] Step S601: Based on the three-dimensional flow field reconstruction results output in step S2, the pollution concentration raster dataset obtained in step S1, and the increased sampling results generated in step S5, a joint likelihood function is constructed to couple the source coordinates, emission rate, and observed concentration data into a unified statistical framework, forming an objective function for Bayesian inversion and providing consistent constraints for posterior distribution solution.

[0090] Step S602: Run the adaptive Markov chain Monte Carlo method to obtain the posterior distribution and output the dynamic source coordinates and emission rates.

[0091] Step S602 runs the adaptive Markov chain Monte Carlo algorithm based on the joint likelihood function in step S601. Through iterative sampling and parameter updates in the state space, it obtains the posterior distribution of the dynamic coordinates and emission rates of the pollution source, and extracts the maximum posterior estimate and statistical interval to provide a quantitative expression of uncertainty for the source tracing results.

[0092] Step S603: Generate a confidence assessment report and push the dynamic source coordinates, emission rates, and confidence intervals to the regulatory platform.

[0093] Step S603 organizes the dynamic source coordinates, emission rates and confidence intervals output in step S602 into a structured form, generates a confidence assessment report with time tags, and pushes it to the regulatory platform through the data interface to realize the real-time delivery of pollution source location and emission information, providing a callable data foundation for emergency response and long-term monitoring.

[0094] Step S6 statistically fuses the three-dimensional flow field reconstruction results, the pollution concentration raster dataset, and the density sampling results. Based on joint likelihood and posterior inference, it generates the spatial coordinates and emission rates of the pollution source and outputs the tracking results with confidence intervals, providing quantifiable source inversion data for the regulatory platform.

[0095] This invention combines multi-source real-time data acquisition with 3D flow field reconstruction to form a unified pollutant concentration raster dataset and meteorological field dataset, achieving dynamic integration of pollutant spatial distribution and meteorological driving factors. A dimensionality-reduced CFD surrogate model generates 3D wind field data files within seconds, enabling real-time wind field estimation under complex turbulent conditions. This provides high-precision boundary conditions for pollutant transport prediction and inversion. By combining Lagrange trajectory backtracking and Eulerian pollution transport solutions, a candidate source hypothesis library is constructed. A difference matrix is ​​used to quantitatively screen source coordinates and emission time histories, ensuring the accuracy of source location calculations. A bidirectional correlation matching is performed between UAV sniffing mass spectrometry fingerprints and candidate source hypotheses to form a matching weight matrix, realizing the gas chemical composition... The coupling of information and spatial inversion models improves the ability to distinguish between multiple sources. By driving secondary drone patrols and ground node-based dense sampling through confidence ranking, adaptive sensor scheduling is achieved, enhancing the spatial resolution of pollution concentration gradient regions. Using Bayesian joint likelihood function and adaptive Markov chain Monte Carlo inversion method, the three-dimensional flow field, concentration grid, and dense sampling results are fused to output dynamic source coordinates and emission rates, and generate confidence assessment reports. This provides real-time, quantitative decision data for the regulatory platform. The entire process adopts a closed-loop data utilization mechanism to ensure that the collected data is called and updated at each processing stage, forming a continuously iterative dynamic pollution source tracking system, which significantly shortens the response time from monitoring to source location.

[0096] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamic tracking of air pollution sources, characterized in that, Includes the following steps: Step S1: Collect pollution concentration raster dataset, UAV sniffing dataset, and meteorological field dataset; Step S2, wind farm reconstruction and pollution transport prediction; Step S3: Generate candidate source hypotheses. For each source coordinate in the candidate source hypothesis, call the Euler pollution transport solution module to simulate the pollutant concentration field in the three-dimensional flow field data file. Perform cell-by-cell difference calculation between the simulated concentration raster and the pollution concentration raster dataset according to the raster index. Output the difference score, error distribution matrix and time alignment label, and write them into the candidate source hypothesis library in descending order of difference score. Step S4: Fingerprint matching and weight calculation; Step S5, sensor adaptive scheduling; Step S6: Bayesian inversion fusion and result output; Step S4 includes the following sub-steps: Step S401: Call the mass spectrometry fingerprint extraction module to perform baseline correction, noise threshold filtering and mass-charge ratio calibration on the full scan mass spectrometry data in the UAV sniffing set, record the normalized peak intensity array under the unified index, and attach the sampling time label and spatial coordinate field to generate a fingerprint feature set; Step S402: Read the process emission fingerprint registered in the candidate source hypothesis library, perform cosine similarity calculation with the peak intensity array corresponding to the fingerprint feature set under the unified index, and write the similarity score into the cross cell of the source index and detection index of the matching weight matrix. Step S5 includes the following sub-steps: Step S501: Based on the matching weight matrix, sort the confidence values ​​of each candidate source, arrange the candidate source list from high to low confidence, and record the source coordinates, initial emission rate and corresponding confidence value of each candidate source. Step S502: Based on the candidate source confidence ranking results and combined with the pollution concentration raster dataset, optimize the sampling frequency of each candidate source high confidence area, generate the UAV secondary cruise path, and generate the density sampling instruction in combination with the layout of ground monitoring nodes, optimize the sampling path and determine the density distribution of sampling nodes. Step S503: Based on the optimized sampling path and the denser sampling instruction, execute the denser sampling task, collect pollution concentration data, and upload the sampling results to the data processing system in real time; The sampling results are merged with the existing pollution concentration raster dataset, the pollution concentration values ​​of the raster cells are updated, and the updated data is synchronized to the pollution source dynamic tracking system. Step S6 includes the following sub-steps: Step S601: Construct a joint likelihood function and fuse the three-dimensional flow field reconstruction results, the pollution concentration raster dataset, and the intensified sampling results; Step S602: Run the adaptive Markov chain Monte Carlo method to obtain the posterior distribution and output the dynamic source coordinates and emission rates; Step S603: Generate a confidence assessment report and push the dynamic source coordinates, emission rates, and confidence intervals to the regulatory platform.

2. The method for dynamic tracking of air pollution sources as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S101: Deploy a set of ground environmental monitoring nodes and continuously upload the pollution concentration raster dataset; Step S102: Schedule the UAV sniffing queue to cruise in layers at a preset altitude to acquire UAV sniffing sets; Step S103: Simultaneously record temperature, humidity, wind speed, wind direction, and boundary layer height, and aggregate them into a meteorological field dataset.

3. The method for dynamic tracking of air pollution sources as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S201: Read the wind speed, wind direction, temperature, humidity and boundary layer height parameters from the meteorological field dataset, call the offline trained dimensionality reduction CFD proxy model, perform joint calculation of three-dimensional wind speed, turbulent kinetic energy and pressure gradient for the preset height range, and generate a three-dimensional flow field data file. The three-dimensional flow field data file includes velocity vector and turbulent energy distribution. Step S202: Interpolate the real-time wind profile measurement sequence at one-second intervals and perform boundary layer constraint correction. Write the interpolation results into a wind profile grid with a unified time reference and align them with the fields of the three-dimensional flow field data file to generate a three-dimensional flow field reconstruction result.

4. The method for dynamic tracking of air pollution sources as described in claim 3, characterized in that, Step S3 loads the velocity vector and turbulent energy distribution based on the three-dimensional flow field data file, sets terminal markers for the cells in the pollution concentration raster dataset whose concentration values ​​are not lower than a set threshold, calls the Lagrange particle trajectory backtracking module to integrate in the reverse direction along the velocity vector to the starting point of the preset time window, records the coordinates of the backtracking nodes, the arrival time and the corresponding concentration value, performs density clustering on the coordinates of the backtracking nodes, and outputs candidate source hypotheses. The candidate source hypotheses include source coordinates, emission start time series and initial emission rate estimate.

5. The method for dynamic tracking of air pollution sources as described in claim 4, characterized in that, The dimension-reduced CFD proxy model used in step S201 is trained based on offline high-fidelity flow field samples.

6. The method for dynamic tracking of air pollution sources as described in claim 5, characterized in that, Step S402 uses a bidirectional correlation matching strategy to cross-check the fingerprint feature set with the process emission fingerprint and calculate the similarity threshold.

7. The method for dynamic tracking of air pollution sources as described in claim 6, characterized in that, Step S502 adjusts the sampling priority of the UAV track and ground node in real time based on the confidence ranking results so that the sampling density is increased by 100% or more in the area with the highest pollution concentration gradient.

Images