Dynamic tracking method for air pollution source

Abstract

The invention discloses a dynamic tracking method for an air pollution source, which relates to the technical field of pollution source tracking and comprises the following steps of: acquiring a pollution concentration raster data set, a UAV sniffing set and a meteorological field data set, reconstructing a wind field and predicting pollution transportation, generating a candidate source hypothesis, matching fingerprints and calculating weights, adaptively scheduling sensors, and performing Bayesian inversion fusion and result output. Unified pollution concentration and meteorological data are generated through multi-source data collection and three-dimensional flow field reconstruction, second-level wind field calculation is achieved through a dimensionality reduction CFD model, candidate source hypotheses are generated in combination with Lagrange backtracking and Euler transport, and a weight matrix is formed through UAV mass spectrum fingerprint bidirectional matching. Secondary cruise of the unmanned aerial vehicle and densification sampling of ground nodes are driven through confidence ranking, and sampling density is adaptively optimized. Bayesian joint likelihood and Markov chain are adopted to fuse multi-source data, dynamic source coordinates and emission rate are output, and a closed-loop iterative air pollution source tracking system is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pollution source tracking, and in particular to an air pollution source dynamic tracking method. BACKGROUND

[0002] With the acceleration of urbanization and industrialization, air pollution events in urban-industrial mixed areas show the characteristics of multi-source, instantaneousness and spatial heterogeneity. The existing air quality monitoring and pollution tracking technology mainly relies on fixed stations and average wind field models. In the complex meteorological conditions of uncertain pollution source quantity, rapid wind direction change and significant turbulence, it is difficult to realize real-time positioning and emission intensity inversion of pollution sources. The traditional source analysis method based on static sampling needs a long time to accumulate data, and the response period is long, which cannot meet the requirements of environmental supervision for second-level response and dynamic tracking. Especially in the scene involving VOCs and other volatile pollutants, the existing technology lacks a closed-loop processing mechanism that can combine multi-source data, quickly reconstruct a three-dimensional flow field and perform inversion calculation, making it difficult for regulatory departments to timely lock the emission source and intervene in the early stage of pollution diffusion.

[0003] At present, the Chinese patent with the application number CN202310243270.4 discloses an environmental air pollution source analysis method, which comprises the following steps: obtaining pollution source component data in a first analysis period to form first pollution source component data, performing standard analysis processing to form standard analysis data; obtaining the standard analysis data, performing first characteristic deviation analysis processing to form first characteristic deviation analysis data; obtaining the standard analysis data, performing second characteristic deviation analysis processing to form second characteristic deviation analysis data; obtaining target pollution source component data, combining the standard analysis data to perform characteristic state analysis on the second pollution source component data to form characteristic state analysis data; and combining the standard analysis data, the first characteristic deviation analysis data, the second characteristic deviation analysis data, and the characteristic state analysis data set to form a pollution source component characteristic data set. The method can reasonably analyze the pollution source from the time level.

[0004] The related technology is difficult to realize real-time positioning and emission rate dynamic inversion of air pollution sources in a complex environment with multi-source pollution, rapid wind field change and significant turbulence. SUMMARY

[0005] The technical problem solved by the present application is that the existing technology is difficult to realize real-time positioning and emission rate dynamic inversion of air pollution sources in a complex environment with multi-source pollution, rapid wind field change and significant turbulence.

[0006] To solve the above technical problems, the present application provides the following technical solutions: An air pollution source dynamic tracking method, comprising the following steps: Step S1, collecting pollution concentration grid dataset, UAV sniffing set and meteorological field dataset; Step S2, wind field reconstruction and pollution transport prediction; Step S3, generating candidate source hypothesis; Step S4, fingerprint matching and weight calculation; Step S5, sensor adaptive scheduling; Step S6, Bayesian inversion fusion and result output.

[0007] Preferably, the step S1 comprises the following sub-steps: Step S101, deploying ground environmental monitoring node set and continuously uploading pollution concentration grid dataset; Step S102, scheduling UAV sniffing queue to cruise at preset height layer, obtaining UAV sniffing set; Step S103, synchronously recording temperature, humidity, wind speed, wind direction and boundary layer height, and gathering into meteorological field dataset.

[0008] Preferably, the step S2 comprises the following sub-steps: Step S201, reading wind speed, wind direction, temperature, humidity and boundary layer height parameters in meteorological field dataset, calling dimension reduction CFD proxy model trained offline, performing three-dimensional wind speed, turbulent kinetic energy and pressure gradient joint calculation on preset height interval, generating three-dimensional flow field data file, the three-dimensional flow field data file comprising velocity vector and turbulent energy distribution; Step S202, performing one-second interval interpolation on real-time wind profile measurement sequence and executing boundary layer constraint correction, writing interpolation result into wind profile grid of unified time reference, and aligning with three-dimensional flow field data file field, generating three-dimensional flow field reconstruction result.

[0009] Preferably, the step S3 comprises the following sub-steps: Step S301, based on three-dimensional flow field data file loading velocity vector and turbulent energy distribution, setting terminal mark for grid elements in pollution concentration grid dataset with concentration value not lower than set threshold, calling Lagrangian particle trajectory backtracking module to integrate reversely along velocity vector to preset time window starting point, recording backtracking node coordinates, arrival time and corresponding concentration value, and performing density clustering on backtracking node coordinates, outputting candidate source hypothesis, the candidate source hypothesis comprising source coordinates, start-up time sequence and initial emission rate estimate; Step S302, calling Euler pollution transport calculation module for each source coordinate in candidate source hypothesis to forward simulate pollution concentration field in three-dimensional flow field data file, performing per-grid element difference calculation on simulated concentration grid and pollution concentration grid dataset according to grid element index, outputting difference score, error distribution matrix and time alignment label, and writing into candidate source hypothesis library in descending order of difference score.

[0010] Preferably, the step S4 comprises the following sub-steps: Step S401, calling the mass spectrum fingerprint extraction module to perform baseline correction, noise threshold filtering and mass-to-charge ratio calibration on the full scan mass spectrum data in the UAV sniffing set, record the normalized peak intensity array under unified indexing, and attach the sampling time label and spatial coordinate field to generate the fingerprint feature set; Step S402, reading the process emission fingerprints registered in the candidate source hypothesis library, and performing cosine similarity calculation on the corresponding peak intensity array of the fingerprint feature set under unified indexing, and writing the similarity score into the source index and detection index cross unit of the matching weight matrix.

[0011] Preferably, the step S5 comprises the following sub-steps: Step S501, based on the matching weight matrix, sorting the confidence value of each candidate source, arranging the candidate source list from high to low according to the confidence, and recording the source coordinates, initial emission rate and corresponding confidence value of each candidate source; Step S502, according to the candidate source confidence sorting result, combining the pollution concentration grid data set, sampling the high confidence area of each candidate source to optimize the sampling frequency, generating the second UAV cruise path, and combining the layout of the ground monitoring node to generate the densification sampling instruction, optimizing the sampling path and determining the density distribution of the sampling node; Step S503, according to the optimized sampling path and the densification sampling instruction, performing the densification sampling task, collecting the pollution concentration data, and uploading the sampling results to the data processing system in real time; Merging the sampling results with the existing pollution concentration grid data set, updating the pollution concentration value of the grid cell, and synchronizing the updated data to the pollution source dynamic tracking system.

[0012] Preferably, the step S6 comprises the following sub-steps: Step S601, constructing a joint likelihood function to fuse the three-dimensional flow field reconstruction result, the pollution concentration grid data set and the densification sampling result; Step S602, running the adaptive Markov chain Monte Carlo to obtain the posterior distribution, outputting the dynamic source coordinates and emission rate; Step S603, generating a confidence evaluation report, and pushing the dynamic source coordinates, emission rate and confidence interval to the supervision platform.

[0013] Preferably, the step S201 adopts a dimension reduction CFD proxy model trained based on offline high-fidelity flow field samples.

[0014] Preferably, the step S402 uses a bidirectional correlation matching strategy to detect and calculate the similarity threshold between the fingerprint feature set and the process emission fingerprint.

[0015] Preferably, the step S502 adjusts the UAV flight path and ground node sampling priority in real time based on the confidence ranking result to make the sampling density in the area with the highest pollution concentration gradient increase by more than one time.

[0016] The present application has the following beneficial effects: the present application generates unified pollution concentration and weather data through multi-source data collection and three-dimensional flow field reconstruction, realizes second-level wind field calculation by using a reduced dimension CFD model, generates candidate source hypotheses by combining Lagrange backtracking and Euler transport, and forms a weight matrix by using UAV mass spectrum fingerprint bidirectional matching. The sampling density is adaptively optimized by driving the second cruise of the UAV and the densification sampling of the ground nodes through confidence ranking. The dynamic source coordinates and emission rate are output by fusing multi-source data through Bayesian joint likelihood and Markov chain, so that the air pollution source tracking system is realized in a closed loop. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A step flow chart of an air pollution source dynamic tracking method provided for an embodiment of the present application is shown in the figure. Figure 2 A physical diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0019] Embodiments, with reference to Figure 1 , an air pollution source dynamic tracking method is provided, which comprises the following steps: Step S1, collecting pollution concentration grid data set, UAV sniffing set and weather field data set.

[0020] Step S2, wind field reconstruction and pollution transport prediction.

[0021] Step S3, generating candidate source hypotheses.

[0022] Step S4, fingerprint matching and weight calculation.

[0023] Step S5, sensor adaptive scheduling.

[0024] Step S6, Bayesian inversion fusion and result output.

[0025] The application generates unified pollution concentration and meteorological data through multi-source data acquisition and three-dimensional flow field reconstruction, realizes second-level wind field calculation by using dimension reduction CFD model, generates candidate source hypothesis by combining Lagrange backtracking and Euler transport, and forms a weight matrix by using UAV mass spectrum fingerprint bidirectional matching. Through confidence ranking driving unmanned aerial vehicle secondary cruise and ground node densification sampling, the sampling density is adaptively optimized. The Bayesian joint likelihood and Markov chain are used to fuse multi-source data, output dynamic source coordinates and emission rate, and realize the closed-loop iterative air pollution source tracking system.

[0026] Step S1 includes the following sub-steps: Step S101, laying out a set of ground environment monitoring nodes and continuously uploading a pollution concentration grid data set.

[0027] Step S101 acquires high-resolution pollution concentration grid data within the ground range, providing a spatial reference benchmark for concentration field spatiotemporal distribution and pollution source positioning.

[0028] Step S102, scheduling the UAV sniffing team to cruise at a preset height, acquiring a UAV sniffing set.

[0029] Step S102 establishes a pollution distribution profile in the vertical direction and acquires air mass spectrum fingerprint information, providing data support for multi-height concentration analysis and source feature identification.

[0030] Step S103, synchronously recording temperature, humidity, wind speed, wind direction and boundary layer height, and gathering into a meteorological field data set.

[0031] Step S103 forms a meteorological field parameter synchronized with the pollution concentration grid data, providing dynamic boundary conditions for three-dimensional wind field calculation and pollution transport path calculation.

[0032] Step S1 realizes multi-dimensional synchronous acquisition of pollution spatial distribution and meteorological driving parameters, forming a basic data set that can be used for subsequent flow field reconstruction and pollution source inversion.

[0033] Step S2 includes the following sub-steps: Step S201, reading the wind speed, wind direction, temperature, humidity and boundary layer height parameters in the meteorological field data set, calling the dimension reduction CFD proxy model trained offline, performing three-dimensional wind speed, turbulent kinetic energy and pressure gradient joint calculation on the preset height interval, generating a three-dimensional flow field data file, and the three-dimensional flow field data file includes velocity vector and turbulent energy distribution.

[0034] The dimension reduction CFD proxy model used in step S201 is trained based on offline high-fidelity flow field samples.

[0035] Step S201 rapidly generates a three-dimensional flow field data file containing velocity vector and turbulent energy distribution by dimension reduction CFD proxy model, to provide fine flow field basis for pollutant transport calculation under complex turbulent flow environment.

[0036] Step S202 interpolates the real-time wind profile measurement sequence at one-second intervals and performs boundary layer constraint correction, writes the interpolation results into the wind profile grid of the unified time reference, and aligns with the three-dimensional flow field data file field to generate three-dimensional flow field reconstruction results.

[0037] Step S202 standardizes the real-time wind profile sequence to a unified time reference and aligns with the three-dimensional flow field data file to form a complete three-dimensional flow field reconstruction result for pollutant transport prediction and trajectory inversion.

[0038] Step S2 realizes high-precision reconstruction of the instantaneous three-dimensional wind field of the target area, and provides dynamic meteorological driving conditions for pollutant transport path calculation and source inversion.

[0039] Step S3 includes the following sub-steps: Step S301, based on the three-dimensional flow field data file, sets a terminal mark for the grid elements in the pollutant concentration grid data set whose concentration values are not lower than the set threshold, calls the Lagrangian particle trajectory backtracking module to integrate the velocity vector in reverse to the starting point of the preset time window, records the backtracking node coordinates, arrival time and corresponding concentration value, and performs density clustering on the backtracking node coordinates, outputs the candidate source hypothesis, and the candidate source hypothesis includes source coordinates, start-up time sequence and initial emission rate estimate.

[0040] Step S301 uses the velocity vector and turbulent energy distribution in the three-dimensional flow field data file to perform trajectory backtracking on the pollutant concentration grid data, maps the grid elements above the concentration threshold to possible source locations, and generates a candidate source hypothesis containing source coordinates, start-up time sequence and initial emission rate estimate, to provide an initial solution space for spatial locking and time sequence reconstruction of the pollution source.

[0041] Step S302, for each source coordinate in the candidate source hypothesis, call the Euler pollutant transport calculation module to simulate the pollutant concentration field in the three-dimensional flow field data file, perform per-grid element difference calculation on the simulated concentration grid and the pollutant concentration grid data set according to the grid element 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.

[0042] Step S302 performs forward concentration field simulation on each candidate source by the Euler pollution transport solution module, and compares with the actual pollution concentration grid by grid, outputs the difference degree score, error distribution matrix and time alignment label, writes the sorted difference degree result into the candidate source hypothesis library, realizes multi-dimensional screening and optimization of source location and emission time, and provides an accurate candidate source set for subsequent inversion calculation.

[0043] Step S3 completes the preliminary calculation of the spatial location and emission time of the pollution source, generates a candidate source hypothesis library by combining backward tracing and forward simulation, and provides basic data support for subsequent fingerprint matching and Bayesian inversion.

[0044] Step S4 includes the following sub-steps: Step S401, calling the mass spectrum fingerprint extraction module, performs baseline correction, noise threshold filtering and mass charge ratio calibration on the full scan mass spectrum data in the UAV sniffing set, records the normalized peak intensity array under unified indexing, and adds the sampling time label and spatial coordinate field, to generate a fingerprint feature set.

[0045] Step S401 performs baseline correction, noise threshold filtering and mass charge ratio calibration on the full scan mass spectrum data in the UAV sniffing set, generates a normalized peak intensity array after unified m / z indexing, and adds the time and space fields to form a complete fingerprint feature set, which provides standardized input for chemical feature analysis of pollutants.

[0046] Step S402, read the process emission fingerprints registered in the candidate source hypothesis library, and perform cosine similarity calculation on the corresponding peak intensity array of the fingerprint feature set under unified indexing, and write the similarity score into the source index and detection index cross unit of the matching weight matrix.

[0047] Step S402 uses a bidirectional correlation matching strategy to mutually detect and calculate the similarity threshold of the fingerprint feature set and the process emission fingerprints.

[0048] Step S402 performs bidirectional correlation matching of the process emission fingerprints in the candidate source hypothesis library and the fingerprint feature set, fills the weight value in the cross unit of the source index and the detection index by cosine similarity calculation, and calculates the similarity threshold to screen stable matching relationship, generates a matching weight matrix, provides quantitative support for chemical feature verification of source positioning, and is used for confidence order calculation.

[0049] Step S4 realizes the correlation verification of the candidate hypothesis of the pollution source and the actual monitored chemical characteristics, generates a matching weight matrix through the feature matching of the mass spectrum fingerprint, and provides a quantitative basis in the chemical composition dimension for subsequent confidence order and sensor scheduling.

[0050] Step S5 includes the following sub-steps: Step S501, based on the matching weight matrix, the confidence value of each candidate source is sorted, and the candidate source list is arranged from high to low confidence, and the source coordinates, initial emission rate and corresponding confidence value of each candidate source are recorded.

[0051] Step S501 converts the matching weight matrix output by step S4 into a quantitative confidence value, globally sorts the candidate sources and outputs the sorted candidate source list, while binding the source coordinates, initial emission rate and corresponding confidence value, and establishes a priority index system for scheduling optimization and subsequent Bayesian inversion.

[0052] Step S502, according to the candidate source confidence sorting result, combined with the pollution concentration grid data set, the high confidence area of each candidate source is sampled frequency optimized, the second cruise path of unmanned aerial vehicle is generated, and the densification sampling instruction is generated combined with the layout of ground monitoring node, the sampling path is optimized and the density distribution of sampling node is determined.

[0053] Step S502 based on the confidence sorting result real-time adjusts the unmanned aerial vehicle flight path and the ground node sampling priority to make the sampling density in the highest pollution concentration gradient area increase by one time or more.

[0054] Step S502 based on the candidate source confidence sorting result, calls the sampling optimization module to analyze the spatial gradient of the pollution concentration grid data, determines the high confidence area and generates the second cruise path of unmanned aerial vehicle; at the same time, according to the geographical distribution of ground monitoring node, the densification sampling position and frequency are calculated, the optimized sampling path instruction containing the unmanned aerial vehicle flight path and the ground node priority is output, the dynamic sampling density scheduling is realized, and the sampling density in the highest pollution concentration gradient area is increased by one time or more.

[0055] Step S503, according to the optimized sampling path and densification sampling instruction, executes the densification sampling task, collects the pollution concentration data, and uploads the sampling results to the data processing system in real time.

[0056] The sampling results are combined with the existing pollution concentration grid data set, the pollution concentration value of the grid cell is updated, and the updated data is synchronized to the pollution source dynamic tracking system.

[0057] Step S503 executes the optimized unmanned aerial vehicle flight path and ground densification sampling instruction, collects multi-dimensional pollution concentration data and uploads to the data processing system in real time, aligns and combines the newly added sampling data with the existing pollution concentration grid data in space-time index, updates the grid concentration value and synchronously writes the updated grid data to the pollution source dynamic tracking system, provides fine data input for real-time inversion calculation.

[0058] Step S5 realizes dynamic collaborative sampling of the UAV and the ground node by matching the confidence ranking driven by the weight matrix and the sensor scheduling, so that the spatial resolution of the high-confidence area is significantly improved, and real-time updated pollution concentration grid data is formed, providing high-precision observation input for subsequent inversion.

[0059] Step S6 includes the following sub-steps: Step S601, construct a joint likelihood function, fuse the three-dimensional flow field reconstruction result, the pollution concentration grid data set and the densified sampling result.

[0060] Step S601 is based on the three-dimensional flow field reconstruction result output in step S2, the pollution concentration grid data set obtained in step S1 and the densified sampling result generated in step S5, constructs a joint likelihood function, couples the source coordinates, the emission rate and the observed concentration data to a unified statistical framework, forms a target function for Bayesian inversion, and provides consistency constraints for solving the posterior distribution.

[0061] Step S602, run adaptive Markov chain Monte Carlo to obtain the posterior distribution, and output the dynamic source coordinates and emission rate.

[0062] Step S602 runs the adaptive Markov chain Monte Carlo algorithm based on the joint likelihood function of step S601, iteratively samples and updates parameters in the state space, obtains the posterior distribution of the dynamic coordinates and emission rate of the pollution source, and extracts the maximum a posteriori estimation value and statistical interval, to provide quantitative uncertainty expression for the source tracking result.

[0063] Step S603, generate a confidence evaluation report, and push the dynamic source coordinates, emission rate and confidence interval to the supervision platform.

[0064] Step S603 structures and organizes the dynamic source coordinates, emission rate and confidence interval output in step S602, generates a time-labeled confidence evaluation report, and pushes it to the supervision platform through a data interface, to realize real-time delivery of the pollution source positioning and emission information, and provide a callable data basis for emergency response and long-term monitoring.

[0065] Step S6 statistically fuses the three-dimensional flow field reconstruction result, the pollution concentration grid data set and the densified sampling result, generates the spatial coordinates and emission rate of the pollution source based on the joint likelihood and posterior inference, and outputs the tracking result with a confidence interval, to provide quantifiable source inversion data for the supervision platform.

[0066] The application combines multi-source real-time data acquisition and three-dimensional flow field reconstruction to form a unified pollution concentration grid data set and a meteorological field data set, realizes dynamic integration of the spatial distribution of pollutants and meteorological driving factors, generates a three-dimensional wind field data file in seconds by using a dimension reduction CFD proxy model, can realize real-time wind field calculation under complex turbulent conditions, provides high-precision boundary conditions for pollutant transport prediction and inversion, constructs a candidate source hypothesis library through the joint of Lagrangian trajectory backtracking and Euler pollution transport calculation, and quantitatively screens the source coordinates and emission time by using a difference matrix, ensures the calculation accuracy of source positioning, realizes two-way correlation matching of UAV sniffing mass spectrum fingerprints and candidate source hypotheses to form a matching weight matrix, realizes the coupling of gas chemical component information and spatial inversion model, improves the discrimination ability in the multi-source scene, realizes the adaptive scheduling of sensors through confidence order driving UAV secondary cruise and ground node densification sampling, enhances the spatial resolution of the pollution concentration gradient area, uses the Bayesian joint likelihood function and the adaptive Markov chain Monte Carlo inversion method to fuse the three-dimensional flow field, the concentration grid and the densification sampling result, outputs the dynamic source coordinates and the emission rate, and generates a confidence evaluation report, provides real-time and quantitative decision data for the supervision platform, the whole process adopts a closed-loop data utilization mechanism, ensures that the collected data is called and updated in each processing stage, forms a continuous iteration of the pollution source dynamic tracking system, and significantly shortens the response time from monitoring to source positioning.

[0067] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (or computer- readable storage media) having computer-usable program code embodied in the medium. The medium can be any available storage media that can be accessed by a computer. By way of example, and not limitation, such computer-usable storage media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other storage medium(s) that can be used to carry or store desired computer program code in the form of instructions or data structures and that can be accessed by a computer. Also, the present application can be embodied in a computer program product that can be traded as goods or merchandise, through the storage medium described above or any other suitable medium. Accordingly, the present application can be embodied in a computer program product that can be traded as goods or merchandise, through the storage medium described above or any other suitable medium. Computer program code embodied in a storage medium is said (referring to a program or code) to "cause a computer" (or Figure 1 one or more functions specified in the flow Figure 1 one or more functions specified in the flow

[0068] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A dynamic tracking method of an air pollution source, characterized by, The method comprises the following steps: Step S1, collecting pollution concentration grid data set, UAV sniffing set and meteorological field data set; Step S2, wind field reconstruction and pollution transport prediction; Step S3, generating candidate source hypothesis; Step S4, fingerprint matching and weight calculation; Step S5, sensor adaptive scheduling; Step S6, Bayesian inversion fusion and result output.

2. The method of claim 1, wherein the air pollution source is a vehicle. The step S1 comprises the following sub-steps: Step S101, laying out ground environmental monitoring node set and continuously uploading pollution concentration grid data set; Step S102, scheduling UAV sniffing queue to cruise at preset height layer to obtain UAV sniffing set; Step S103, synchronously recording temperature, humidity, wind speed, wind direction and boundary layer height to gather meteorological field data set.

3. The method of claim 2, wherein the air pollution source is a vehicle. The step S2 comprises the following sub-steps: Step S201, reading wind speed, wind direction, temperature, humidity and boundary layer height parameters in meteorological field data set, calling offline trained dimension reduction CFD proxy model, performing three-dimensional wind speed, turbulent kinetic energy and pressure gradient joint calculation on preset height interval to generate three-dimensional flow field data file, wherein the three-dimensional flow field data file comprises velocity vector and turbulent energy distribution; Step S202, performing one-second interval interpolation on real-time wind profile measurement sequence and performing boundary layer constraint correction, writing the interpolation result into wind profile grid of unified time reference and aligning with three-dimensional flow field data file field to generate three-dimensional flow field reconstruction result.

4. The method of claim 3, wherein the air pollution source is a vehicle. The step S3 comprises the following sub-steps: Step S301, based on three-dimensional flow field data file loading velocity vector and turbulent energy distribution, setting terminal mark for grid element in pollution concentration grid data set with concentration value not lower than set threshold, calling Lagrangian particle trajectory backtracking module to perform reverse integration along velocity vector to preset time window starting point, recording backtracking node coordinates, arrival time and corresponding concentration value, and performing density clustering on backtracking node coordinates to output candidate source hypothesis, wherein the candidate source hypothesis comprises source coordinates, start-up and discharge time sequence and initial discharge rate estimate; Step S302, calling Euler pollution transport calculation module for each source coordinate in candidate source hypothesis to simulate pollution concentration field in three-dimensional flow field data file, performing per-grid element difference calculation on simulated concentration grid and pollution concentration grid data set according to grid element index, outputting difference score, error distribution matrix and time alignment label, and writing into candidate source hypothesis library in descending order of difference score.

5. The method of claim 4, wherein the air pollution source is a vehicle. The step S4 comprises the following sub-steps: Step S401, calling mass spectrum fingerprint extraction module to perform baseline correction, noise threshold filtering and mass-to-charge ratio calibration on full scan mass spectrum data in UAV sniffing set, recording normalized peak intensity array under unified index, and adding sampling time label and space coordinate field to generate fingerprint feature set; Step S402, reading process emission fingerprint registered in candidate source hypothesis library, performing cosine similarity calculation on corresponding peak intensity array of fingerprint feature set under unified index, and writing similarity score into source index and detection index cross unit of matching weight matrix.

6. The method of claim 5, wherein the air pollution source is a vehicle. The step S5 comprises the following sub-steps: Step S501, based on the matching weight matrix, the confidence value of each candidate source is sorted, and the candidate source list is arranged from high to low confidence, and the source coordinates, initial emission rate and corresponding confidence value of each candidate source are recorded; Step S502, according to the candidate source confidence sorting result, combined with the pollution concentration grid data set, the high confidence area of each candidate source is sampled frequency optimization, the unmanned aerial vehicle secondary cruise path is generated, and the densification sampling instruction is generated combined with the layout of the ground monitoring node, the sampling path is optimized and the density distribution of the sampling node is determined; Step S503, according to the optimized sampling path and the densification sampling instruction, the densification sampling task is executed, the pollution concentration data is collected, and the sampling result is uploaded to the data processing system in real time; The sampling result is combined with the existing pollution concentration grid data set, the pollution concentration value of the grid element is updated, and the updated data is synchronized to the pollution source dynamic tracking system.

7. The method of claim 6, wherein the air pollution source is dynamically tracked by: The step S6 includes the following substeps: Step S601, construct a joint likelihood function, fuse three-dimensional flow field reconstruction results, pollution concentration grid data set and densification sampling results; Step S602, run adaptive Markov chain Monte Carlo to obtain posterior distribution, output dynamic source coordinates and emission rate; Step S603, generate confidence evaluation report, push dynamic source coordinates, emission rate and confidence interval to the supervision platform.

8. The method of claim 7, wherein the air pollution source is a vehicle. The dimension reduction CFD proxy model used in step S201 is trained based on offline high-fidelity flow field samples.

9. The method of claim 8, wherein the air pollution source is a vehicle. The step S402 uses a bidirectional correlation matching strategy to detect and calculate the similarity threshold between the fingerprint feature set and the process emission fingerprint.

10. The method of claim 9, wherein the air pollution source is a vehicle. The step S502 adjusts the unmanned aerial vehicle flight path and ground node sampling priority in real time based on the confidence sorting result to make the sampling density in the area with the highest pollution concentration gradient increase by more than one time.

Images