A real-time monitoring method and system for urban atmospheric pollution

Abstract

The application discloses a kind of urban atmospheric pollution real-time monitoring method and system.Method includes: obtaining pollutant concentration, weather and geographic position data by multi-source monitoring equipment;Data fusion is carried out using spatiotemporal kriging interpolation and wavelet threshold denoising, to generate high-quality initial field of urban pollution distribution;Prediction is carried out using machine learning model with encoder-decoder structure, wherein the encoder is convolutional long short-term memory network, and the decoder is dynamic graph convolution network that fuses weather factors, to generate dynamic pollution trend map;Pollution area identification is carried out based on wind field streamline, and precise tracing is carried out by combining control variable method through Lagrangian particle diffusion model reverse simulation.It realizes the whole process optimization of pollution monitoring from data fusion, accurate prediction to quantitative tracing, significantly improves the accuracy and timeliness of urban atmospheric pollution supervision.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring technology, and in particular relates to a method and system for real-time monitoring of urban air pollution. Background Technology

[0002] With the acceleration of urbanization, air pollution has become increasingly severe, posing a serious threat to residents' health and the ecological environment. Real-time and accurate monitoring of pollution distribution and rapid source tracing are crucial for effective pollution control.

[0003] Existing urban air pollution monitoring methods primarily rely on data from fixed monitoring stations, which suffers from drawbacks such as low spatial coverage and inability to reflect pollution gradients within a region. While some technologies have introduced mobile monitoring equipment to supplement data, at the data fusion level, simple spatiotemporal interpolation (such as inverse distance weighting and ordinary kriging) is often used, failing to fully consider the complex influence of urban wind fields and building layouts on pollutant transport, resulting in limited accuracy of the generated pollution distribution field. In terms of prediction, traditional physical models are computationally complex and time-consuming, while early machine learning models struggle to capture the spatiotemporal dynamic evolution of pollutants. Regarding source tracing, simple static probability models based on wind direction and distance are often used, which cannot handle complex scenarios such as intermittent emissions from pollution sources and multiple source superposition, resulting in low source tracing accuracy. Summary of the Invention

[0004] This invention provides a method and system for real-time monitoring of urban air pollution, which addresses the technical problem of low source tracing accuracy in complex scenarios such as intermittent emissions from pollution sources and multiple sources overlapping.

[0005] In a first aspect, the present invention provides a method for real-time monitoring of urban air pollution, comprising:

[0006] Acquire multi-source air pollution data for the target area, including pollutant concentration data, meteorological data, and geographical location data;

[0007] Based on the timestamp and geographic location information of the geographic location data, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused using preset data processing rules to obtain the initial field of urban pollution distribution in the target area.

[0008] The initial field of urban pollution distribution is input into a pre-trained atmospheric pollution diffusion machine learning model. Combined with the meteorological data, the diffusion trend and concentration change of pollutants in the future preset time period are inferred to generate a dynamic pollution situation map.

[0009] Based on the dynamic pollution situation map, a preset pollution area identification strategy is used to identify areas with abnormal concentrations. According to the pollution source list in the preset urban geographic information system, the spatial correlation probability between the areas with abnormal concentrations and potential pollution sources is calculated to determine the final pollution source.

[0010] Secondly, the present invention provides a real-time urban air pollution monitoring system, comprising:

[0011] The acquisition module is configured to acquire multi-source air pollution data of the target area, wherein the multi-source air pollution data includes pollutant concentration data, meteorological data, and geographical location data;

[0012] The fusion module is configured to perform spatiotemporal alignment and fusion processing on the geographic location data and the pollutant concentration data based on the timestamp and geographic location information of the geographic location data, using preset data processing rules, to obtain the initial field of urban pollution distribution in the target area.

[0013] The generation module is configured to input the initial field of urban pollution distribution into a pre-trained atmospheric pollution diffusion machine learning model, and combine it with the meteorological data to infer the diffusion trend and concentration changes of pollutants within a preset time period in the future, and generate a dynamic pollution situation map.

[0014] The determination module is configured to identify areas of abnormal concentration based on the dynamic pollution situation map using a preset pollution area identification strategy, and calculate the spatial correlation probability between the areas of abnormal concentration and potential pollution sources according to the pollution source list in the preset urban geographic information system, so as to determine the final pollution source.

[0015] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the urban air pollution real-time monitoring method according to any embodiment of the present invention.

[0016] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the urban air pollution real-time monitoring method according to any embodiment of the present invention.

[0017] This application's method and system for real-time urban air pollution monitoring generates a high-quality initial pollution field that is spatiotemporally continuous, noise-suppressed, and sensitive to real pollution events by combining non-uniform grid partitioning, spatiotemporal Kriging interpolation, and adaptive wavelet denoising, fundamentally improving the reliability of the data source. Furthermore, it combines the spatiotemporal feature extraction capabilities of ConvLSTM with a dynamic graph convolutional network that integrates meteorological and physical laws, enabling the model to not only learn data patterns but also possess reasoning capabilities guided by physical mechanisms. This significantly improves the accuracy and physical rationality of pollution diffusion prediction, especially in simulating pollution transmission paths under complex wind fields, achieving an accuracy that is difficult to achieve with traditional methods. Additionally, it identifies truly valuable anomaly areas through flow field segmentation and clustering, and combines this with a three-level source tracing system of reverse trajectory simulation, potential source area delineation, and quantitative verification of control variables. This achieves precise determination from "potentially polluted areas" to "specific responsible sources" with quantitative contribution, completely changing the limitations of traditional methods that rely on static parameters and qualitative inference. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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 illustrating a real-time monitoring method for urban air pollution according to an embodiment of the present invention;

[0020] Figure 2 This is a structural block diagram of a real-time urban air pollution monitoring system provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. 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] Please see Figure 1 The diagram shows a flowchart of a real-time monitoring method for urban air pollution according to this application.

[0024] like Figure 1As shown, the real-time monitoring method for urban air pollution specifically includes the following steps:

[0025] Step S101: Obtain multi-source air pollution data for the target area, including pollutant concentration data, meteorological data, and geographical location data.

[0026] Step S102: Based on the timestamp and geographic location information of the geographic location data, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused using preset data processing rules to obtain the initial field of urban pollution distribution in the target area.

[0027] In this step, the geographic location data and the pollutant concentration data are uniformly mapped to a preset spatiotemporal grid coordinate system, where each grid cell represents a spatial region of the target area within a fixed time interval. For grid cells with missing data, a spatiotemporal kriging interpolation algorithm is used to fill them in, and the filled grid data is then denoised using a wavelet threshold denoising algorithm to preserve the true abrupt change features in the data, thus obtaining the target geographic location data and the target pollutant concentration data. The target geographic location data and the target pollutant concentration data are then weighted and fused on a unified grid to generate the initial field of urban pollution distribution.

[0028] In one specific embodiment, firstly, a multi-scale spatiotemporal grid coordinate system covering the target area is constructed. The spatial grid is set to a non-uniform structure based on terrain and pollution source distribution (e.g., a 500m×500m high-density grid for city centers and a 1km×1km standard grid for suburbs), with a fixed time interval of 1 hour. All multi-source data from fixed monitoring stations and mobile monitoring devices are quickly allocated to corresponding spatiotemporal grid units using a spatial hash mapping algorithm, based on the high-precision geographic location information recorded by their GPS / BeiDou modules and the timestamp synchronized by atomic clocks. For grid units with missing data, a spatiotemporal kriging interpolation algorithm is used to fill in the gaps. This algorithm dynamically adjusts the spatial correlation distance and time decay coefficient through an adaptive variogram model. Specifically for areas with abrupt wind field changes caused by urban building clusters, it introduces anisotropic parameters based on wind direction perception, making the interpolation results more consistent with the actual transport patterns of pollutants in the wind field. After filling, threshold denoising based on wavelet modulus maxima theory is applied to the full grid data: a three-level decomposition using the Db4 wavelet basis is performed, and an adaptive threshold function is applied to high-frequency coefficients. This function can distinguish between random noise (manifested as uniformly distributed small fluctuations) and real pollution events (manifested as spatiotemporally continuous large abrupt changes), thus effectively filtering out noise while fully preserving the pollution front characteristics caused by sudden emissions. The denoised data is the target geographic location data and target pollutant concentration data. Finally, dynamic reliability-weighted fusion is performed on a unified grid: appropriate weights are assigned to data from different sources (such as fixed stations and mobile devices) in each grid cell. The weight values ​​are dynamically calculated based on the data source type precision coefficient (fixed stations have a higher base weight than mobile devices), temporal proximity (data closer to the current moment has a higher weight), and the local signal-to-noise ratio calculated after denoising. Through this weighted fusion, a spatiotemporally continuous, noise-suppressed initial field of urban pollution distribution that can sensitively respond to changes in real pollution is finally generated.

[0029] In summary, by constructing a non-uniform spatiotemporal grid system, an adaptive representation of the complex urban underlying surface was achieved. The spatiotemporal kriging interpolation algorithm significantly improved the physical rationality and accuracy of missing data imputation, especially overcoming the inaccuracy of traditional methods in wind field abrupt change zones. By introducing adaptive denoising based on wavelet modulus maxima theory, background noise was effectively filtered out while crucially preserving the true abrupt change signals characterizing sudden pollution events, solving the problem of conventional filtering methods easily smoothing out important pollution features. Finally, through a dynamic credibility-weighted fusion strategy, the complementary advantages of multi-source data were fully utilized, resulting in an initial urban pollution distribution field with not only high spatiotemporal resolution but also significantly improved data quality and reliability, laying a very solid data foundation for accurate prediction and source tracing of subsequent pollution trends.

[0030] Step S103: Input the initial field of urban pollution distribution into the pre-trained atmospheric pollution diffusion machine learning model, combine it with the meteorological data, and infer the diffusion trend and concentration change of pollutants in the future preset time period to generate a dynamic pollution situation map.

[0031] In this step, the atmospheric pollution diffusion machine learning model is a spatiotemporal sequence prediction model with an encoder-decoder structure. The encoder of the atmospheric pollution diffusion machine learning model is a convolutional long short-term memory network, which is used to extract the spatial features and their temporal dependencies in the initial field of urban pollution distribution. The decoder of the atmospheric pollution diffusion machine learning model is a graph convolutional network. The nodes of the graph convolutional network are the grid center points of the spatiotemporal grid coordinate system, and the edge weights are determined by the actual distance between grids and the wind direction and speed, which is used to simulate the cross-grid transport process of pollutants under complex urban wind fields.

[0032] Specifically, meteorological data for a future preset time period, including wind direction, wind speed, air pressure, and mixing layer height, are used as the edge weight update conditions for the graph convolutional network and input into the decoder. The decoder extrapolates the pollutant concentration change sequence for each grid within the future preset time period step by step and outputs it as a heat map overlaid on the electronic map to generate the dynamic pollution situation map.

[0033] In one specific embodiment, the encoder is a convolutional long short-term memory network that receives the initial field sequence of urban pollution distribution over the past six consecutive time steps (e.g., the past six hours). This ConvLSTM introduces the gating mechanism of LSTM into the convolutional operation, simultaneously capturing the spatial correlation of pollution distribution (e.g., the spatial morphology of pollution plumes) and its dynamic evolution over time (e.g., the transport direction and rate of pollutants), ultimately outputting a hidden state vector containing spatiotemporal features.

[0034] Next, the decoder is a dynamic graph convolutional network, constructed as follows: the center point of each grid in the spatiotemporal grid coordinate system is used as a graph node, and the node feature is the pollutant concentration at that location. The edges of the graph structure are determined by the actual Euclidean distance between grids and the real-time wind direction and speed. The specific edge weight calculation formula is as follows:

[0035] ,

[0036] in, Let the edge weights be those of grid i and grid j. Let L be the distance between grid i and grid j, and L be the feature length scale. Let φ be the azimuth angle from grid i to grid j, and φ be the current wind direction.

[0037] This design significantly enhances the downwind side weight, accurately simulating the dominant advection transport effect of pollutants under the influence of the wind field.

[0038] During the extrapolation phase, gridded forecast meteorological data (including wind direction, wind speed, air pressure, and mixing layer height) for a predetermined future time period (e.g., the next 24 hours) are used as the dynamic update conditions for the edge weights of the graph convolutional network and input into the decoder in real time. The decoder uses the hidden state output by the encoder as its initial state, combines it with the current pollution distribution, aggregates neighborhood information through graph convolution operations, and iterates the extrapolation step-by-step (e.g., once per hour) to finally output the pollutant concentration change sequence for each grid within the predetermined future time period. The system is trained using a course learning strategy, first learning short-term predictions and gradually increasing the prediction duration to ensure model stability. Finally, the extrapolated concentration sequence is overlaid on an electronic map as a dynamic heatmap, generating a visualized dynamic pollution situation map. Different colors in the map represent different concentration levels, and a time slider is used to show the evolution of the pollution situation.

[0039] Step S104: Based on the dynamic pollution situation map, an abnormal concentration area is identified using a preset pollution area identification strategy. Based on the pollution source list in the preset urban geographic information system, the spatial correlation probability between the abnormal concentration area and potential pollution sources is calculated to determine the final pollution source.

[0040] In this step, based on the dynamic pollution situation map, the dominant wind direction and wind speed information for a future preset time period are extracted to construct a dynamic wind field streamline; all grid points with pollution concentrations exceeding a first concentration threshold are identified from the dynamic pollution situation map as a set of potential anomalies; cluster analysis based on flow field segmentation is performed on the set of potential anomalies: potential anomalies located downstream of the same or a group of converging streamlines are merged into the same candidate region, and isolated high points in the streamline direction are removed; the comprehensive pollution intensity index I of each candidate region is calculated, and the calculation formula is: I = (average concentration × regional projected area × duration of continuous exceedance) / Euclidean distance between the region centroid and the nearest potential pollution source; the candidate regions with the comprehensive pollution intensity index exceeding a second concentration threshold are finally identified as concentration anomaly regions that need to be traced to their source.

[0041] Furthermore, based on the wind field data corresponding to the dynamic pollution situation map, starting from the geometric center of the identified concentration anomaly area, a preset Lagrange particle diffusion model is used to simulate the reverse time step, tracking the trajectory of air masses arriving at the concentration anomaly area within a preset backtracking time, forming one or more reverse trajectory lines; the spatial distribution of all reverse trajectory lines within the backtracking time is superimposed to generate a spatial distribution map of potential source contribution function values, and the geographical area with a PSCF value higher than a preset value is defined as a probability source area; within the probability source area, combined with the pollution source list, all potential pollution sources located within the concentration anomaly area are screened; using the atmospheric pollution diffusion machine learning model, a set of control variable simulations is performed: the emissions of each potential pollution source in the pollution source list are sequentially set to zero, the atmospheric pollution diffusion machine learning model is run and the concentration reduction amount in the concentration anomaly area is calculated at this time; the concentration reduction amount corresponding to each pollution source is normalized, and each normalized concentration reduction amount is used as the spatial correlation probability between each pollution source and the concentration anomaly area, and the pollution source with the highest correlation probability is determined as the final pollution source.

[0042] In one specific embodiment, firstly, from the dynamic pollution situation map for the next 6 hours, the hourly gridded dominant wind direction and speed are extracted using a weighted vector average algorithm, and then a dynamic wind field streamline map covering the entire area is generated using a streamline integration method. The PM2.5 concentration threshold is set to 75 μg / m³. 3 As the first concentration threshold, a spatial neighborhood search algorithm is used to identify all grid points exceeding the threshold from the situation map, forming a set of potential anomalies. Subsequently, physical constraint clustering based on flow field segmentation is performed: guided by the aforementioned streamlines, an improved DBSCAN clustering algorithm is used, with its core parameter—the neighborhood search range—set to expand along the streamline direction. Only anomalies downstream of the same or a group of converging streamlines and within 15% of the streamline length are grouped into the same candidate contamination region. This process effectively eliminates isolated high points without physical connection in the streamline direction caused by model noise.

[0043] For each candidate region generated by clustering, its comprehensive pollution intensity index I is calculated: I = (average concentration × projected area of ​​the region × duration of continuous exceedance) / Euclidean distance between the region's centroid and the nearest potential pollution source. An index threshold of I0 = 400 is set, and candidate regions with I > I0 are ultimately identified as areas with abnormal concentrations requiring in-depth source tracing.

[0044] Taking a identified area of ​​abnormal concentration as an example, 10,000 virtual particles are released from its geometric center. Using the HYSPLIT Lagrange particle diffusion model and hourly wind field data corresponding to the dynamic pollution situation map, a 72-hour reverse time step simulation is performed to track the trajectories of all particles arriving in the area. The kernel density estimation method is used to overlay the spatial distribution of all reverse trajectories to generate a potential source contribution function (PSCF) map. A PSCF threshold of 0.7 is set, and geographical areas with PSCF values ​​higher than this threshold are designated as high-probability source areas.

[0045] Within the designated probabilistic source area, and using the pollution source inventory from the city's geographic information system, spatial intersection analysis was employed to identify all potential industrial pollution sources located within this area (such as chemical plant A, thermal power plant B, and steel plant C). Subsequently, a controlled variable simulation engine was activated: keeping all other conditions constant, the emissions of each potential pollution source in the inventory were sequentially set to zero, and an atmospheric pollution diffusion machine learning model was run to calculate the concentration reduction in the target anomaly area under three scenarios. Finally, the relative contribution probability of each pollution source was calculated using the min-max normalization method, yielding the contribution probabilities of plant A, thermal power plant B, and steel plant C.

[0046] In summary, the method in this application, by combining non-uniform grid partitioning, spatiotemporal kriging interpolation, and adaptive wavelet denoising, generates a high-quality initial pollution field that is spatiotemporally continuous, noise-suppressed, and can sensitively respond to real pollution events, fundamentally improving the reliability of the data source. Furthermore, by combining the spatiotemporal feature extraction capability of ConvLSTM with a dynamic graph convolutional network that integrates meteorological and physical laws, the model not only learns data patterns but also possesses reasoning capabilities guided by physical mechanisms, significantly improving the accuracy and physical rationality of pollution diffusion prediction. In particular, it achieves a level of accuracy in simulating pollution transmission paths under complex wind fields that is difficult to achieve with traditional methods. Moreover, by identifying truly valuable anomaly areas through flow field segmentation and clustering, and combining this with a three-level source tracing system of reverse trajectory simulation, potential source area delineation, and quantitative verification of control variables, it achieves accurate determination from "potentially polluted areas" to "specific responsible sources" with quantitative contribution, completely changing the limitations of traditional methods that rely on static parameters and qualitative inference.

[0047] Please see Figure 2 The diagram shows a structural block diagram of a real-time urban air pollution monitoring system according to this application.

[0048] like Figure 2 As shown, the urban air pollution real-time monitoring system 200 includes an acquisition module 210, a fusion module 220, a generation module 230, and a determination module 240.

[0049] The acquisition module 210 is configured to acquire multi-source air pollution data of the target area, including pollutant concentration data, meteorological data, and geographic location data. The fusion module 220 is configured to perform spatiotemporal alignment and fusion processing on the geographic location data and the pollutant concentration data based on the timestamp and geographic location information of the geographic location data, using preset data processing rules, to obtain an initial field of urban pollution distribution in the target area. The generation module 230 is configured to input the initial field of urban pollution distribution into a pre-trained air pollution diffusion machine learning model, and combine it with the meteorological data to deduce the diffusion trend and concentration change of pollutants in a preset time period, generating a dynamic pollution situation map. The determination module 240 is configured to identify areas with abnormal concentrations based on the dynamic pollution situation map using a preset pollution area identification strategy, and calculate the spatial correlation probability between the areas with abnormal concentrations and potential pollution sources according to a preset list of pollution sources in the urban geographic information system, so as to determine the final pollution source.

[0050] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0051] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the urban air pollution real-time monitoring method in any of the above method embodiments.

[0052] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0053] Acquire multi-source air pollution data for the target area, including pollutant concentration data, meteorological data, and geographical location data;

[0054] Based on the timestamp and geographic location information of the geographic location data, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused using preset data processing rules to obtain the initial field of urban pollution distribution in the target area.

[0055] The initial field of urban pollution distribution is input into a pre-trained atmospheric pollution diffusion machine learning model. Combined with the meteorological data, the diffusion trend and concentration change of pollutants in the future preset time period are inferred to generate a dynamic pollution situation map.

[0056] Based on the dynamic pollution situation map, a preset pollution area identification strategy is used to identify areas with abnormal concentrations. According to the pollution source list in the preset urban geographic information system, the spatial correlation probability between the areas with abnormal concentrations and potential pollution sources is calculated to determine the final pollution source.

[0057] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the urban air pollution real-time monitoring system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, and these remote memories may be connected to the urban air pollution real-time monitoring system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0058] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the urban air pollution real-time monitoring method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the urban air pollution real-time monitoring system. The output device 340 may include a display screen or other display device.

[0059] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0060] In one implementation, the above-described electronic device is used in a real-time urban air pollution monitoring system as a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0061] Acquire multi-source air pollution data for the target area, including pollutant concentration data, meteorological data, and geographical location data;

[0062] Based on the timestamp and geographic location information of the geographic location data, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused using preset data processing rules to obtain the initial field of urban pollution distribution in the target area.

[0063] The initial field of urban pollution distribution is input into a pre-trained atmospheric pollution diffusion machine learning model. Combined with the meteorological data, the diffusion trend and concentration change of pollutants in the future preset time period are inferred to generate a dynamic pollution situation map.

[0064] Based on the dynamic pollution situation map, a preset pollution area identification strategy is used to identify areas with abnormal concentrations. According to the pollution source list in the preset urban geographic information system, the spatial correlation probability between the areas with abnormal concentrations and potential pollution sources is calculated to determine the final pollution source.

[0065] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0066] 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 method for real-time monitoring of urban air pollution, characterized in that, include: Acquire multi-source air pollution data for the target area, including pollutant concentration data, meteorological data, and geographical location data; Based on the timestamp and geographic location information of the geographic location data, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused using preset data processing rules to obtain the initial field of urban pollution distribution in the target area. The initial field of urban pollution distribution is input into a pre-trained atmospheric pollution diffusion machine learning model. Combined with the meteorological data, the diffusion trend and concentration change of pollutants in the future preset time period are inferred to generate a dynamic pollution situation map. Based on the dynamic pollution situation map, a preset pollution area identification strategy is used to identify areas with abnormal concentrations. Then, according to a pollution source list in a preset urban geographic information system, the spatial correlation probability between the areas with abnormal concentrations and potential pollution sources is calculated to determine the final pollution source. The step of identifying areas with abnormal concentrations based on the dynamic pollution situation map using the preset pollution area identification strategy includes: Based on the dynamic pollution situation map, the dominant wind direction and wind speed information for the future preset time period are extracted to construct dynamic wind field streamlines. All grid points whose pollution concentration exceeds the first concentration threshold are identified from the dynamic pollution situation map and used as a set of potential anomalies. Cluster analysis based on flow field segmentation is performed on the set of potential anomalies: potential anomalies located downstream of the same or a group of converging streamlines are grouped into the same candidate region, and isolated high points in the streamline direction are removed. Calculate the comprehensive pollution intensity index I for each candidate area. The calculation formula is: I = (average concentration × projected area of ​​the area × duration of continuous exceedance) / Euclidean distance between the centroid of the area and the nearest potential pollution source; Candidate areas whose comprehensive pollution intensity index exceeds the second concentration threshold are ultimately identified as areas with abnormal concentrations that require source tracing.

2. The method for real-time monitoring of urban air pollution according to claim 1, characterized in that, Based on the timestamps and geographic location information of the geographic location data, and using preset data processing rules, the geographic location data and the pollutant concentration data are spatiotemporally aligned and fused to obtain the initial field of urban pollution distribution in the target area, including: The geographic location data and the pollutant concentration data are uniformly mapped to a preset spatiotemporal grid coordinate system, wherein each grid cell represents a spatial region of the target area within a fixed time interval; For grid cells with missing data, a spatiotemporal kriging interpolation algorithm is used to fill them in, and the filled grid data is then denoised using a wavelet threshold denoising algorithm to preserve the real abrupt change features in the data, thus obtaining the target geographical location data and the target pollutant concentration data. The target geographic location data and the target pollutant concentration data are weighted and fused on a unified grid to generate the initial field of urban pollution distribution.

3. The method for real-time monitoring of urban air pollution according to claim 1, characterized in that, The atmospheric pollution diffusion machine learning model is a spatiotemporal sequence prediction model with an encoder-decoder structure. The encoder of the atmospheric pollution diffusion machine learning model is a convolutional long short-term memory network, which is used to extract the spatial features and their temporal dependencies in the initial field of urban pollution distribution. The decoder of the atmospheric pollution diffusion machine learning model is a graph convolutional network. The nodes of the graph convolutional network are the grid center points of the spatiotemporal grid coordinate system, and the edge weights are determined by the actual distance between grids and the wind direction and speed, which is used to simulate the cross-grid transport process of pollutants under complex urban wind fields. The step of inputting the initial field of urban pollution distribution into a pre-trained atmospheric pollution diffusion machine learning model, and combining it with meteorological data to predict the diffusion trend and concentration changes of pollutants within a preset time period, and generating a dynamic pollution situation map includes: Meteorological data for a future preset time period, including wind direction, wind speed, air pressure, and mixing layer height, are used as the edge weight update conditions for the graph convolutional network and input into the decoder. The decoder extrapolates the pollutant concentration change sequence for each grid within a preset time period by time step, and overlays it onto the electronic map in the form of a heat map to generate the dynamic pollution situation map.

4. The method for real-time monitoring of urban air pollution according to claim 1, characterized in that, The step of calculating the spatial correlation probability between the concentration anomaly area and potential pollution sources based on the pollution source list in the preset urban geographic information system, in order to determine the final pollution sources, includes: Based on the wind field data corresponding to the dynamic pollution situation map, starting from the geometric center of the identified concentration anomaly area, a preset Lagrange particle diffusion model is used to simulate the reverse time step, and the trajectory of the air mass that arrives at the concentration anomaly area within the preset backtracking time is tracked to form one or more reverse trajectory lines. The spatial distribution of all reverse trajectory lines over the backtracking time is superimposed to generate a spatial distribution map of potential source contribution function values. The geographical area where the PSCF value is higher than the preset value is defined as the probability source area. Within the probability source region, and in conjunction with the pollution source list, all potential pollution sources located within the concentration anomaly region are screened out; Using the aforementioned atmospheric pollution diffusion machine learning model, a set of control variables simulations were performed: the emissions of each potential pollution source in the pollution source inventory were sequentially set to zero, the atmospheric pollution diffusion machine learning model was run, and the concentration reduction in the concentration anomaly area was calculated at this time. The concentration reduction amount corresponding to each pollution source is normalized, and each normalized concentration reduction amount is used as the spatial correlation probability between each pollution source and the concentration anomaly area. The pollution source with the highest correlation probability is determined as the final pollution source.

5. A real-time urban air pollution monitoring system, characterized in that, include: The acquisition module is configured to acquire multi-source air pollution data of the target area, wherein the multi-source air pollution data includes pollutant concentration data, meteorological data, and geographical location data; The fusion module is configured to perform spatiotemporal alignment and fusion processing on the geographic location data and the pollutant concentration data based on the timestamp and geographic location information of the geographic location data, using preset data processing rules, to obtain the initial field of urban pollution distribution in the target area. The generation module is configured to input the initial field of urban pollution distribution into a pre-trained atmospheric pollution diffusion machine learning model, and combine it with the meteorological data to infer the diffusion trend and concentration changes of pollutants within a preset time period in the future, and generate a dynamic pollution situation map. The determination module is configured to identify areas of abnormal concentration based on the dynamic pollution situation map using a preset pollution area identification strategy, and calculate the spatial correlation probability between the areas of abnormal concentration and potential pollution sources according to a preset list of pollution sources in a city geographic information system, so as to determine the final pollution source. The step of identifying areas of abnormal concentration based on the dynamic pollution situation map using the preset pollution area identification strategy includes: Based on the dynamic pollution situation map, the dominant wind direction and wind speed information for the future preset time period are extracted to construct dynamic wind field streamlines. All grid points whose pollution concentration exceeds the first concentration threshold are identified from the dynamic pollution situation map and used as a set of potential anomalies. Cluster analysis based on flow field segmentation is performed on the set of potential anomalies: potential anomalies located downstream of the same or a group of converging streamlines are grouped into the same candidate region, and isolated high points in the streamline direction are removed. Calculate the comprehensive pollution intensity index I for each candidate area. The calculation formula is: I = (average concentration × projected area of ​​the area × duration of continuous exceedance) / Euclidean distance between the centroid of the area and the nearest potential pollution source; Candidate areas whose comprehensive pollution intensity index exceeds the second concentration threshold are ultimately identified as areas with abnormal concentrations that require source tracing.

Images