Pollutant emission monitoring and early warning method based on physical prior and information value driving

Abstract

The invention discloses a pollution emission monitoring and early warning method based on physical prior and information value driving, and relates to the technical field of environment monitoring and pollution control, and the method comprises the following steps: collecting production side certificate data, executing time alignment and self-calibration drift removal, and constructing a standardized multi-source database; judging whether a heat island effect occurs or not, and if yes, collecting volume heat capacity, solar irradiance and heat island intensity data of a single building; according to the method, the heat island effect is introduced to judge and generate a new time-space diagram network, real-time heat island intensity changes caused by building materials and solar irradiation can be dynamically captured, and building wind field and plume data are accurately corrected. The problem that a conventional method cannot adapt to irregular fluctuation of vertical wind shear caused by the real-time heat island effect is solved, the pollution plume low-altitude compression phenomenon caused by the heat island effect is effectively corrected, and the monitoring direction is always aligned with the real diffusion path of pollutants.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of environmental monitoring and pollution control technology, and in particular to a pollution emission monitoring and early warning method based on physical priori and information value driving. BACKGROUND

[0002] In the field of industrial park pollution emission monitoring, the core pain points in daily operation often include multiple points alarming at the same time, limited resources of law enforcement personnel and portable monitoring vehicles, and short window of pollution source evidence. The scheduling logic of existing monitoring and early warning technology has significant defects, and only allocates law enforcement resources according to distance or fixed priority, without considering key factors such as pollution prediction uncertainty, dynamic migration law of wind direction, and evidence success probability. Moreover, the strategy lacks a unified measure of pollution prediction, uncertainty, action cost, and evidence success rate, resulting in misallocation and waste of law enforcement resources, frequent missed key pollution evidence, and direct impact on the subsequent punishment execution and rectification promotion effect of the illegal enterprise.

[0003] Industrial park buildings are densely built and have various materials, which are prone to form heat island effect under certain meteorological conditions. The conventional heat island effect causes pollution diffusion to deviate to the low-temperature area due to the stable temperature difference between the building and the surrounding environment, although it affects the prediction accuracy, the law can be found, and only the basic wind field needs to be corrected to adapt. However, real-time heat island effect, especially the situation caused by sudden increase of daytime solar radiation, has an impact far beyond the conventional norm. Real-time heat island effect triggers a reverse vertical wind shear, and its intensity shows irregular fluctuations every 10 minutes or so. This intense and unstable vertical wind field structure directly compresses the pollution plume to 2-5 meters in low altitude, which is completely deviated from the 10-15 meter vertical diffusion range usually predicted by the threshold probability field based on conventional meteorological conditions.

[0004] This phenomenon leads to two results: first, the pollution monitoring equipment at a fixed height cannot align with the actual aggregation height of the pollution plume, and the concentration data collected by the equipment is distorted, which cannot reflect the true pollution situation, resulting in incorrect monitoring direction. Second, the pollution source tracing and early warning based on distorted data will deviate, making it impossible to accurately locate and obtain the real pollution behavior. The existing technology lacks an effective mechanism for sensing, modeling, and compensating for this real-time and irregular heat island effect, and cannot solve the problems of pollution monitoring blind area and scheduling failure caused by it. SUMMARY

[0005] The present application aims to solve the problem that the pollution monitoring equipment of the existing technology has a fixed collection height, which cannot solve the sudden change of pollution diffusion law caused by real-time heat island effect, and proposes a new pollution emission monitoring and early warning method based on physical priori and information value driving.

[0006] In order to achieve the above object, the present application adopts the following technology: a pollution emission monitoring and early warning method based on physical priori and information value driving, comprising the following steps:

[0007] Collecting production side data, performing time alignment and self-calibration de-drifting, and constructing a standardized multi-source database;

[0008] Judging whether the heat island effect will occur, if so:

[0009] Collecting single building volume heat capacity, solar irradiance and heat island intensity data, based on the standardized multi-source database, single building volume heat capacity, solar irradiance and heat island intensity data, embedding a space-time graph network, and obtaining a new space-time graph network;

[0010] Based on the new space-time graph network, the building wind field and plume data are corrected; based on the corrected building wind field and plume data, the adjusted threshold probability field is obtained; based on the obtained adjusted threshold probability field and real-time pollution concentration data, the optimized threshold probability field is obtained;

[0011] Based on the optimized threshold probability field, the collection height of the pollution monitoring equipment is adjusted, and the suspected low-altitude concentration data is collected by the pollution monitoring equipment; the authenticity of the suspected low-altitude concentration data is judged; if correct, the suspected low-altitude concentration data is the real low-altitude concentration data;

[0012] Based on the real low-altitude concentration data, the optimal scheduling scheme is generated.

[0013] Further, the method of embedding the space-time graph network based on the standardized multi-source database, single building volume heat capacity, solar irradiance and heat island intensity data comprises:

[0014] Obtaining a building attribute list; based on the building attribute list, calculating the single building volume heat capacity;

[0015] Based on the ultrasonic wind field data and fixed station data and spatial coordinates in the standardized multi-source database, combined with real-time heat island intensity data and solar irradiance data; obtaining a three-dimensional data set;

[0016] Mapping the heat island intensity data and solar irradiance data in the three-dimensional data set to the wind shear correction coefficient;

[0017] Then inputting the single building volume heat capacity data and the three-dimensional data set into the space-time graph network to obtain a new space-time graph network.

[0018] Further, the method of correcting the building wind field and plume data based on the new space-time graph network comprises:

[0019] Obtaining the ultrasonic wind field and pollution concentration data of the standardized multi-source database, combining the single building volume heat capacity, real-time heat island intensity and solar irradiance data, obtaining the space-time graph network input tensor;

[0020] Using buildings and monitoring points as nodes and spatial distance as edges, the law of conservation of mass and the Gaussian diffusion mechanism are embedded in the graph convolutional layer. The temporal dynamic characteristics of wind field and plume are captured through the temporal convolutional layer to obtain the vertical component of the basic wind field.

[0021] Based on the new spatiotemporal map network, real-time heat island intensity and solar irradiance data, wind shear correction coefficients are obtained; based on the wind shear correction coefficients and the vertical component of the basic wind field, corrected building wind field data are obtained.

[0022] Based on the corrected building wind field data, the pollution source emissions from the standardized multi-source database, and the volumetric heat capacity of a single building, plume data were calculated.

[0023] Furthermore, methods for determining whether a heat island effect will occur include:

[0024] Real-time wind speed and wind direction data are obtained from a standardized multi-source database, the average wind speed is calculated, and the average wind speed is compared with a wind speed threshold.

[0025] If the average wind speed is less than the wind speed threshold, the corresponding time period is a period of weak wind.

[0026] The building surface temperature and ambient temperature are collected during periods of low wind, and the difference between the building surface temperature and the ambient temperature is calculated to obtain the real-time heat island intensity.

[0027] Within the same preset time window for calculating the real-time heat island intensity, solar irradiance data is collected; based on the solar irradiance data within the window, the magnitude of the sudden increase in irradiance is calculated.

[0028] Compare the real-time heat island intensity with the intensity threshold; compare the sudden increase in irradiance with the amplitude threshold;

[0029] If the real-time heat island intensity is greater than the intensity threshold, and the sudden increase in irradiance is greater than the amplitude threshold, the heat island effect will occur.

[0030] Furthermore, based on the corrected building wind field and plume data, methods for adjusting the overthreshold probability field include:

[0031] The area containing building wind field and plume data is divided into multiple grid cells, and the building location and pollution source coordinates are labeled in each grid cell.

[0032] Based on the corrected building wind field and plume data, combined with the building location and pollution source coordinates of each grid cell, the hourly predicted concentration of each grid cell is calculated; based on the hourly predicted concentration of each grid cell and the preset threshold for exceeding the threshold, the probability of exceeding the threshold for each grid cell is calculated; thus, the adjusted probability field for exceeding the threshold is obtained.

[0033] Furthermore, based on the obtained adjusted threshold probability field and real-time pollution concentration data, methods for optimizing the threshold probability field include:

[0034] Real-time pollution concentration data is divided into grid cells according to the adjusted threshold probability field and matched with timestamps. Outliers in the concentration data are removed to ensure that the spatiotemporal dimensions of the two types of data are fully aligned, resulting in a spatiotemporally aligned threshold probability field.

[0035] An initial sample set is obtained based on the spatiotemporally aligned over-threshold probability field. The initial sample set is then input into a new spatiotemporal graph network, which outputs the predicted concentration field corresponding to each initial sample. The initial error is calculated based on the predicted concentration field of each initial sample and real-time pollution concentration data.

[0036] Based on real-time pollution concentration data, an observation error matrix is ​​constructed, and based on the initial error, a model error matrix is ​​constructed. Based on the observation error matrix and the model error matrix, the overthreshold probability value of the grid cells of each initial sample is adjusted to obtain the optimized overthreshold probability field.

[0037] Furthermore, methods for adjusting the acquisition height of pollution monitoring equipment based on optimizing the overthreshold probability field include:

[0038] Based on the optimized overthreshold probability field, grid cells with a probability greater than a preset threshold are extracted to obtain the overthreshold spatial region, and vertical pollution distribution characteristic data of the overthreshold spatial region are collected; based on the vertical pollution distribution characteristic data, the initial vertical centroid position of the pollution plume in the overthreshold spatial region is calculated.

[0039] Based on the corrected building wind field data and the initial vertical centroid position, the corrected vertical centroid position of the plume is obtained; based on the corrected vertical centroid position of the plume, the acquisition height of the pollution monitoring equipment is obtained.

[0040] Furthermore, methods for verifying the authenticity of suspected low-altitude concentration data include:

[0041] Based on suspected low-altitude concentration data, a multi-scale data assimilation system is constructed to output the analysis field.

[0042] Based on the optimized superthreshold probability field and analysis field, the probabilistic concentration field is obtained; based on the probabilistic concentration field, the simulated concentration spatiotemporal sequence is obtained; based on the simulated concentration spatiotemporal sequence, the reanalysis concentration field is obtained.

[0043] Resampling based on the reanalysis concentration field yields a spatial uncertainty distribution map of the real low-altitude concentration data, and key uncertainty regions are identified based on the uncertainty distribution map.

[0044] Collect vertical profile concentration data in key uncertainty areas; if the vertical profile concentration data is within the confidence interval of the uncertainty distribution, the suspected low-altitude concentration data is the true low-altitude concentration data.

[0045] Furthermore, methods for deriving key uncertainty regions based on uncertainty distribution maps include:

[0046] Based on the uncertainty distribution map, the uncertainty quantification index of each grid cell is obtained;

[0047] The uncertainty quantification index of each grid cell is compared with a preset uncertainty threshold; grid cells with uncertainty quantification index greater than the preset uncertainty threshold are selected.

[0048] The selected grid cells are clustered and connected in space to obtain continuous spatial patches;

[0049] Based on the corrected building wind field data, the flow field location of the spatial patch was determined;

[0050] Based on the spatial patch area and flow field location, key uncertainty areas are identified.

[0051] Furthermore, methods for collecting vertical profile concentration data in key uncertainty areas include:

[0052] Based on the spatial coordinates of the key uncertainty area, the initial flight path of the UAV is obtained;

[0053] Based on the corrected building wind field data, the instantaneous flow field characteristics in the key uncertainty area are obtained;

[0054] Based on the instantaneous flow field characteristics, the initial flight path is adjusted to obtain a determined flight path;

[0055] The drone climbs along a predetermined flight path, collecting pollutant concentration data at corresponding horizontal and vertical positions to obtain vertical profile concentration data.

[0056] In summary, due to the adoption of the above-mentioned pollution emission monitoring and early warning method based on physical priors and information value, the beneficial effects of this invention are:

[0057] This invention effectively integrates production evidence data, meteorological data, and monitoring data by constructing a standardized multi-source database and embedding it into a spatiotemporal graph network, overcoming the limitations of existing technologies that rely on single data sources and fixed preprocessing methods. This deep integration and standardized processing of multi-source data lays a reliable data foundation for subsequent accurate forecasting, improving the quality and consistency of monitoring data from the source.

[0058] This invention introduces a new spatiotemporal map network to determine the urban heat island effect, enabling dynamic capture of real-time heat island intensity changes caused by building materials and solar radiation, and precise correction of building wind field and plume data. This solves the problem that conventional methods cannot adapt to the irregular fluctuations in vertical wind shear caused by the real-time heat island effect, effectively corrects the low-altitude compression of pollution plumes caused by the heat island effect, and ensures that the monitoring direction is always aligned with the actual diffusion path of pollutants.

[0059] This invention combines the adjusted overthreshold probability field with real-time pollution concentration data, and generates an optimized overthreshold probability field through an ensemble Kalman assimilation algorithm, which significantly reduces the uncertainty of model prediction. This process dynamically integrates the physical model with real-time observation data, breaking through the prediction bottleneck of existing technologies that rely solely on static models or single data sources, and achieving continuous optimization and self-correction of the spatial distribution of pollution overthreshold probability.

[0060] This invention is based on dynamically adjusting the acquisition height of pollution monitoring equipment by optimizing the probability field beyond the threshold. By performing multi-scale data assimilation and computational fluid dynamics simulation on suspected low-altitude concentration data, a complete quality control closed loop from data acquisition to authenticity judgment is constructed. This effectively identifies and filters out true low-altitude concentration data, eliminates false alarms and missed alarms caused by improper monitoring height or data distortion, and ensures the accuracy and reliability of monitoring results. Attached Figure Description

[0061] Figure 1 The flowchart of the pollution emission monitoring and early warning method based on physical priors and information value driven by the present invention is shown.

[0062] Figure 2 The flowchart illustrating the new spatiotemporal graph network obtained by this invention is shown.

[0063] Figure 3 The flowchart of the present invention for determining the authenticity of suspected low-altitude concentration data is shown. Detailed Implementation

[0064] The pollution emission monitoring and early warning method based on physical prior knowledge and information value driven by 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0065] To more clearly and intuitively demonstrate the practical application effects and advantages of the pollution emission monitoring and early warning method based on physical prior knowledge and information value, and to verify its feasibility and effectiveness, the invention is further described below with reference to embodiments. Through specific scenario simulations and data calculations, the method is explained in detail how it plays a role in actual pollution emission monitoring and early warning, helping readers better understand the technical details and practical value of the invention. The invention is further described below with reference to embodiments;

[0066] Example 1:

[0067] See Figures 1-2 First, production verification data, pollution concentration data from fixed stations / perimeter micro-stations, and wind field data from lidar / ultrasonic anemometers are collected. Time alignment is performed on the three types of data, i.e., they are unified to UTC time, and high-frequency data are sampled at 5-minute intervals. Then, the three types of data are self-calibrated to remove drift, i.e., the amount of drift in the concentration data is reverse-checked based on the production verification flow data, in order to build a standardized multi-source database with complete dimensions. The production verification data includes valve position, flow rate, and power load.

[0068] It should be noted that methods for constructing a standardized multi-source database include:

[0069] The data types from multiple sources in the industrial park are classified into monitoring, meteorological, and production categories. Pollution concentration data from fixed stations / perimeter micro-stations, wind field data from lidar / ultrasonic anemometers, and valve position / flow / power load data from production side are classified into monitoring, meteorological, and production categories. The hardware parameters of each data acquisition device are clarified to lay the foundation for subsequent targeted processing.

[0070] Wavelet transform is used to denoise monitoring data, moving average is used to smooth fluctuations in meteorological data, and interpolation is used to fill in short-term missing data in production data to ensure that the original data is free of interference. The classification results from the first step are used directly to solve the problem of poor adaptability of existing single preprocessing technology and obtain clean preprocessed data.

[0071] Based on the preprocessed clean data, time alignment is performed with UTC time as the reference. That is, high-frequency production data is downsampled according to the timestamp of low-frequency monitoring data, and meteorological data is matched by linear interpolation to achieve full data time synchronization, eliminate data misalignment caused by time deviation of existing technology, and obtain time-aligned data.

[0072] Cross-device self-calibration to remove drift is performed on the time-aligned data; that is, the concentration drift of the monitoring data is reverse-checked based on the flow data of the production side, the wind direction data of the ultrasonic anemometer is used to correct the wind profile deviation of the lidar, and the calibration correlation is established based on the data aligned in the third step, which breaks through the limitation of single-device calibration in the existing technology and obtains calibrated data.

[0073] The calibrated data is subjected to correlation verification, which involves constructing a causal correlation model between production verification data and monitoring data, calculating the correlation coefficient, and removing data with correlation coefficients lower than the correlation threshold, thereby eliminating unrelated abnormal data, such as the sudden increase in concentration when the valve is closed, and obtaining the valid data after verification.

[0074] The validated data is packaged according to the environmental monitoring data standardization specifications; that is, metadata tags such as device number, collection time, and verification result are added to each type of data to form a structured database that can be directly adapted to the input format of the subsequent spatiotemporal graph network.

[0075] Determine whether the urban heat island effect will occur. If it does:

[0076] Data on the volumetric heat capacity, solar irradiance, and heat island intensity of individual buildings are collected and embedded into a spatiotemporal graph network based on a standardized multi-source database and data on the volumetric heat capacity, solar irradiance, and heat island intensity of individual buildings.

[0077] It should be noted that methods for determining whether the heat island effect has occurred include:

[0078] Real-time wind speed and direction data from an ultrasonic anemometer are extracted from a standardized multi-source database. The average wind speed is calculated in a 5-minute time window. Weak wind periods with average wind speed < 2 m / s are selected to define an effective time range for subsequent heat island data collection, which is different from the limitations of existing technologies that directly collect heat island data without combining wind field data.

[0079] Based on the selected low wind periods, heat island related data collection is triggered; that is, the infrared thermometer deployed on the top of the building and the environmental thermometer in the open space 10 meters away are activated to simultaneously collect the building surface temperature and the ambient temperature during the low wind period, ensuring that the collection timing matches the low wind scenario and solving the problem of the lack of targeted collection timing in the existing technology.

[0080] The difference between the collected building surface temperature and the ambient temperature is calculated to obtain the real-time heat island intensity; at the same time, the building material at the temperature measurement point is recorded, such as steel structure and concrete, to provide a basis for subsequent correlation of volumetric heat capacity;

[0081] Within the same 3-minute window for calculating the real-time heat island intensity, solar irradiance data is continuously collected using a radiometer. The difference between the maximum and minimum solar irradiance values ​​within this window is calculated to obtain the magnitude of the sudden increase in irradiance, thus clarifying the quantitative calculation method for the sudden increase.

[0082] Based on average wind speed, real-time heat island intensity, and sudden increase in irradiance, a three-dimensional judgment condition is constructed: average wind speed < 2m / s, heat island intensity > 1℃, and sudden increase in irradiance > 200W / ㎡. If these conditions are met, the heat island effect is determined to occur, breaking through the limitations of existing single threshold judgments.

[0083] It should be noted that the methods for embedding spatiotemporal graph networks based on standardized multi-source databases, single-building volumetric heat capacity, solar irradiance, and urban heat island intensity data include:

[0084] Basic information about buildings involved in the heat island scenario is extracted from the industrial park's geographic information system, including building materials, three-dimensional volume, and spatial coordinates, thereby establishing a list of building attributes and providing accurate basic data for volumetric heat capacity calculation; building materials include steel structure / concrete, and the three-dimensional volume is length × width × height.

[0085] Based on the building attribute list, the volumetric heat capacity of a single building is calculated according to material parameters; that is, the volumetric heat capacity of a single steel structure building is calculated as volume × material density × 480 J / (kg). (℃), the volumetric heat capacity of a single concrete building is calculated as volume × material density × 900 J / (kg) (℃), the volumetric heat capacity of a single building in a mixed-material building is calculated by weighting the volume ratio of each part to ensure that the data is adapted to the actual building structure;

[0086] Ultrasonic wind field data, fixed station concentration data and spatial coordinates of the corresponding area of ​​the heat island scenario are extracted from a standardized multi-source database. The ultrasonic wind field data, fixed station concentration data and real-time heat island intensity and solar irradiance data are matched with the spatial coordinates to form a three-dimensional dataset.

[0087] By employing an attention mechanism, the heat island intensity data and solar irradiance data in the 3D dataset are mapped to wind shear correction coefficients, establishing a dynamic mapping relationship between heat island parameters and wind shear direction / intensity, thus obtaining a scene-based association layer, overcoming the limitation of existing spatiotemporal graph networks lacking a scene-based association layer;

[0088] The volumetric heat capacity data of a single building and the 3D dataset are input into the spatiotemporal graph network. At the same time, a scenario-based association layer is enabled, allowing the spatiotemporal graph network to incorporate the influence of the heat island on wind shear when learning the mass conservation and diffusion mechanism, resulting in a new spatiotemporal graph network.

[0089] Based on the new spatiotemporal graph network, the building wind field and plume data are corrected; based on the corrected building wind field and plume data, the adjusted overthreshold probability field is obtained.

[0090] It should be noted that the methods for correcting building wind field and plume data based on spatiotemporal graph networks include:

[0091] Ultrasonic wind field and pollution concentration data under the heat island scenario are obtained from a standardized multi-source database. Combined with the volumetric heat capacity of a single building, real-time heat island intensity and solar irradiance data, the data are encapsulated in a three-dimensional structure according to spatial coordinates, timestamps and physical parameters to form a spatiotemporal graph network input tensor, which provides a structured basis for network computation and avoids the problem of chaotic data format in existing technologies.

[0092] A new spatiotemporal graph network framework with physical priors is constructed. Specifically, buildings and monitoring points are used as nodes, and spatial distance is used as edges. The law of mass conservation and the Gaussian diffusion mechanism are embedded in the graph convolutional layer. The temporal dynamic features of wind field and plume are captured through temporal convolutional layers to obtain the vertical component of the basic wind field. The initialization network of the spatiotemporal graph network input tensor is directly used to break through the limitations of the existing pure data-driven approach. The law of mass conservation means that the total mass of pollutants remains unchanged, and the Gaussian diffusion mechanism is a small-scale diffusion model.

[0093] The heat island and vertical wind shear correlation layer in the new spatiotemporal map network is activated, and the heat island intensity and solar irradiance data are converted into wind shear correction coefficients. The vertical component of the basic wind field is adjusted in reverse to correct the wind shear deviation of lower heat and upper cold caused by the heat island. Using the network structure and input data of the previous steps, the corrected building wind field data is obtained, where the vertical component of the basic wind field is the rate of change of wind speed with height.

[0094] Based on the corrected building wind field data, combined with the pollution source emissions from production verification data taken from a standardized multi-source database, the three-dimensional trajectory of the pollution plume is calculated through the plume diffusion equation, and the plume edge concentration is corrected according to the building bypass coefficient to ensure consistency with the actual diffusion law. The three-dimensional trajectory includes the planar path + vertical height, and the building bypass coefficient is dynamically adjusted based on the volume heat capacity of a single building.

[0095] The corrected building wind field data and plume data are compared with the measured wind profiles from lidar and the real-time concentration data from microstations. The wind speed deviation rate is calculated to ensure it is not greater than the deviation rate threshold of 5% and the concentration error value is not greater than the error value threshold of 8%. If these conditions are not met, the samples are marked as anomalous and are directly verified based on the output data from the previous steps. The plume data includes the three-dimensional trajectory of the pollution plume and the concentration at the plume edge. The anomalous samples are fed back to the new spatiotemporal graph network. The attention weights and physical prior embedding coefficients of the heat island and vertical wind shear association layer are adjusted using the gradient descent method. Steps two through five are then re-executed until the error reaches the target, forming a dynamic correction closed loop.

[0096] It should be noted that, based on the corrected building wind field and plume data, the methods for obtaining the adjusted overthreshold probability field include:

[0097] Based on the corrected spatial resolution of the building wind field and the three-dimensional range of the plume diffusion, the industrial park is divided into multiple structured grid units. The location of the building and the coordinates of the pollution source are marked in each grid unit to avoid the problem of the grid unit division and the wind field plume being disconnected in the existing technology. For example, the spatial resolution is 10m×10m grid unit, and the three-dimensional range of the plume diffusion is 2 to 5 meters at low altitude.

[0098] Based on the pollution concentration prediction sub-model, the hourly predicted concentration of each grid cell is calculated; at the same time, for the building bypass area, an additional bypass correction term is added to correct the predicted concentration value of the area, ensuring that the prediction results match the actual characteristics of the plume being compressed to the low altitude. The model parameters and grid cell information of the previous steps are used, where the building bypass area is taken from the corrected wind field data, and the bypass correction term is such as the wind speed attenuation coefficient.

[0099] Referring to national air pollutant emission standards, a threshold for exceeding the threshold concentration is preset for each grid cell. Based on the hourly predicted concentration and the preset threshold concentration, the percentage of times the hourly predicted concentration exceeds the preset threshold concentration within the prediction period is statistically analyzed to calculate the probability of exceeding the threshold for each grid cell. At the same time, the wind speed fluctuation error of the corrected wind field is introduced as a probability correction factor to improve the reliability of the results. The predicted concentration and threshold data from the previous steps are directly used, with a period of 60 minutes and a wind speed fluctuation error of ±5%, thus obtaining the adjusted probability field of exceeding the threshold.

[0100] Specifically, methods for constructing pollution concentration prediction sub-models include:

[0101] Using the wind speed and direction data from the corrected building wind field data as diffusion dynamic parameters, and the diffusion coefficient and initial concentration of the plume data as basic parameters, and incorporating the real-time emissions of pollution sources from production evidence data in a standardized multi-source database, a grid cell-level concentration prediction formula is established, thereby obtaining a pollution concentration prediction sub-model.

[0102] Based on the obtained adjusted overthreshold probability field and real-time pollution concentration data, and combined with the ensemble Kalman assimilation algorithm, an optimized overthreshold probability field is obtained;

[0103] It should be noted that, based on the obtained adjusted threshold probability field and real-time pollution concentration data, the methods for optimizing the threshold probability field include:

[0104] Spatiotemporal consistency processing is performed on the adjusted threshold probability field and the real-time pollution concentration data. That is, the real-time pollution concentration data is divided into grid cells according to the adjusted threshold probability field and matched with the timestamps. Outliers in the concentration data are removed by the 3σ criterion to ensure that the spatiotemporal dimensions of the two types of data are completely aligned. This is different from the problem of existing technologies ignoring data matching and obtains a spatiotemporally aligned threshold probability field. For example, the grid cell division is 10m×10m and the timestamp is 5 minutes / time.

[0105] Based on the spatiotemporally aligned overthreshold probability field, the initial sample set of the ensemble Kalman assimilation algorithm is obtained: according to the uncertainty distribution of the overthreshold probability field, 50-100 sets of probability field samples are randomly generated. Each set of initial samples retains the overthreshold probability and concentration prediction value of the grid cell. The probability field data after the first step of alignment is used directly, which breaks through the limitation of the single initial value in the existing technology. For example, the uncertainty distribution fluctuates by ±15% in the edge region.

[0106] The generated initial sample set is input into the new spatiotemporal graph network to perform the assimilation prediction step: the network outputs the predicted concentration field corresponding to each set of initial samples, and the initial error between the predicted concentration field of each set of initial samples and the real-time pollution concentration data is calculated to provide an error benchmark for subsequent observation updates.

[0107] Construct observation error matrix and model error matrix; that is, observation error matrix is ​​based on instrument accuracy setting of real-time pollution concentration data in the first step, and model error matrix is ​​based on statistical variance calculation of initial error in the third step. Clarify the two types of error quantification values ​​and distinguish them from the fuzzy processing of existing technologies that ignore error quantification.

[0108] Based on the observation error matrix and the model error matrix, the initial sample set in the second step is updated by observation: the observation information of real-time pollution concentration data is incorporated into each initial sample by using the ensemble Kalman filter formula, and the over-threshold probability value of the grid cell of each initial sample is adjusted so that the error between the predicted concentration and the real-time concentration of the updated sample is reduced.

[0109] Based on the optimized threshold probability field, the acquisition height of the pollution monitoring equipment is adjusted, and the pollution monitoring equipment acquires suspected low-altitude concentration data; the authenticity of the suspected low-altitude concentration data is judged; if it is correct, the suspected low-altitude concentration data is the true low-altitude concentration data.

[0110] It should be noted that the methods for adjusting the sampling height of pollution monitoring equipment based on optimizing the overthreshold probability field include:

[0111] First, grid cells with a probability greater than a preset probability threshold are extracted from the optimized overthreshold probability field to obtain overthreshold spatial regions. Vertical pollution distribution characteristic data of these overthreshold spatial regions are collected to clarify the aggregation trend of pollution concentration at different vertical levels. This helps to lock the core target area for subsequent height adjustment, which is different from the limitation of existing technologies that only focus on planar areas and ignore vertical distribution.

[0112] Based on the vertical pollution distribution characteristics data, the initial vertical centroid position of the pollution plume in the high superthreshold probability area is calculated. This initial vertical centroid position reflects the vertical layer where the pollutants are most concentrated. The vertical distribution data from the first step is used directly, avoiding the blindness of setting the height based on experience in existing technologies.

[0113] Based on the corrected building wind field data and the initial vertical centroid position, the vertical variation law of the wind field in the high superthreshold probability area is analyzed, the influence trend of the wind field on the vertical centroid position of the plume is determined, and the corrected vertical centroid position of the plume is obtained, overcoming the defect of existing technology that does not consider the vertical influence of the wind field; among them, the influence trend is, for example, whether it stabilizes the centroid or slightly shifts it.

[0114] Based on the corrected vertical center of gravity of the plume and the monitoring accuracy characteristics of the pollution monitoring equipment, the sampling height of the pollution monitoring equipment is determined to ensure that this range can completely cover the core aggregation layer of the pollution plume.

[0115] Specifically, the vertical pollution distribution characteristics data include:

[0116] Pollutant concentration data from multiple vertical height nodes are selected by choosing 5-6 consecutive nodes covering key heights within the super-threshold space region. The grid average concentration value and standard deviation of each height node are recorded. The nodes at key heights need to cover both the low-altitude accumulation zone and the normal heights in the heat island scenario. Typical key height nodes are 1m, 3m, 5m, 8m, 12m, and 15m. The vertical centroid position is essentially the weighted center of concentration in the vertical direction. The concentration difference at different heights needs to reflect the pollution accumulation trend. If only 1-2 height nodes are selected, the concentration information of the intermediate layer will be lost, such as the high concentration zone at 2-5 meters in the low-altitude region during the heat island effect, causing the centroid position to be biased towards a certain extreme height. Selecting 5-6 consecutive nodes that cover 1-15 meters, including the near-ground, low-altitude core area, and high-altitude blank layer, can completely capture the concentration gradient of pollution from the near-ground to the high altitude, ensuring that the weighted average result can reflect the true accumulation core.

[0117] Precise positioning data for vertical height nodes is crucial, including the absolute altitude and relative ground height for each concentration collection node. The absolute altitude error should be ≤0.5 meters, and the relative ground height error ≤0.1 meters. This prevents inconsistent concentration data calculations across different nodes due to height reference deviations. A unified height reference is required for calculating the vertical center of gravity. If the positioning references for different height nodes are inconsistent (e.g., node A is based on relative building roof height while node B is based on relative ground height), the height values ​​cannot be directly used in weighted calculations; for example, 1m relative roof height might correspond to 5m relative ground height. Accurate absolute / relative height data ensures consistent height references for all concentration nodes, preventing errors in center of gravity calculations due to positioning deviations. An error >0.3 meters will cause subsequent equipment to collect data at heights deviating from the core pollution area.

[0118] The data includes the vertical emission height of pollution sources in the over-threshold space area, the vertical height of the emission center of all active pollution sources in the over-threshold space area, and the emission flow rate. Active pollution sources include exhaust vents of production units and breather valves of storage tanks. It is necessary to indicate whether each pollution source is a major emission source, based on the emission flow rate ratio. If it is greater than 30% of the total emission in the area, it is considered a major source. The emission height of the pollution source directly determines the initial diffusion starting point of the pollution plume. If the emission height is ignored and only environmental monitoring concentration data is used for calculation, the high concentration near the emission outlet may be misjudged as the pollution accumulation core. For example, if the emission height of a pollution source is 4m, the high concentration at a height of 4m near it is a direct effect of the emission, rather than a natural accumulation of the plume. The emission height and flow rate data are included, and abnormal concentration values ​​directly affected by the emission outlet are removed before calculation. For example, the concentration within ±0.5m of the emission outlet height needs to be corrected according to the emission flow rate weight to ensure that the center of gravity reflects the natural accumulation trend after the plume diffusion, rather than the direct interference of the emission source.

[0119] The optimal scheduling scheme is obtained based on real low-altitude concentration data.

[0120] It should be noted that the methods for obtaining the optimal scheduling scheme include:

[0121] Based on real low-altitude concentration data and combined with an optimized threshold probability field, the contribution probability of each potential pollution source to the real low-altitude concentration data is calculated, a dynamic source tracing probability map is generated, and high-probability discharge units are identified.

[0122] Based on the dynamic source tracing probability map, the spatial distribution of high-probability pollution discharge units is extracted, and production verification data in the standardized multi-source database is associated to identify the real-time operating conditions and emission intensity of the corresponding enterprises.

[0123] By combining the corrected building wind field data, we analyze the plume transport path from the location of the high-probability sewage discharge unit to the actual low-altitude concentration data collection point, and predict the movement trajectory and concentration change trend of the pollution plume in a specific future period.

[0124] By integrating the movement trajectory of pollution plumes, concentration change trends, and real-time operating conditions of enterprises, a dynamic dispatch value index is constructed that integrates the probability of source tracing, the expected success of evidence collection, the cost of dispatch actions, and the urgency of timeliness.

[0125] All targets to be inspected are sorted according to the dynamic scheduling value index, law enforcement resources are prioritized to the area with the highest index, and the optimal inspection route covering the core time period and key path is automatically generated.

[0126] The optimal inspection route is sent to the mobile monitoring terminal, simultaneously triggering real-time tracking and recording of production certification data of target enterprises along the route, forming a closed-loop management system covering the entire chain from monitoring and early warning to on-site certification.

[0127] Example 2:

[0128] Please see Figure 3 After obtaining suspected low-altitude concentration data, monitoring personnel determine the authenticity of the suspected low-altitude concentration data using methods including:

[0129] Based on the suspected low-altitude concentration data collected by pollution monitoring equipment, a multi-scale data assimilation system is constructed. The initial observation field of the multi-scale data assimilation system is the actual low-altitude concentration data. Its purpose is to integrate the discrete actual low-altitude concentration data into a continuous spatial distribution model.

[0130] Specifically, multi-scale data assimilation systems include:

[0131] Define the core components and input data of the multi-scale data assimilation system; that is, the system uses suspected low-altitude concentration data as the initial observation field, the optimized overthreshold probability field generated by the preprocess as the prior constraint of the background field, and the corrected building wind field data as the driving parameters of the physical transport model.

[0132] An assimilation framework for a multi-scale data assimilation system is constructed. This framework uses an ensemble Kalman filter algorithm as the core assimilation method. By setting up an assimilation set containing dozens to hundreds of members, the uncertainty between the observed data and the model background field is characterized.

[0133] A physical constraint module is configured for the multi-scale data assimilation system. This module embeds the corrected building wind field data into the pollutant advection diffusion equation, ensuring that the assimilation process is carried out under mass conservation and dynamic constraints, thus guaranteeing that the generated analytical field conforms to the physical laws of atmospheric diffusion.

[0134] The system employs a multi-scale nested structure for data assimilation. By setting grid layers with different spatial resolutions, the system can transmit and receive information between the overall scale of the industrial park and the local scale of the building complex, enabling multi-resolution analysis from macro to micro levels.

[0135] Initialize the operating parameters of the multi-scale data assimilation system, including setting the assimilation time window and calculation step size to keep them synchronized with the acquisition frequency of suspected low-altitude concentration data and the update cycle of the optimized overthreshold probability field, ensuring that all input data are fully matched in the spatiotemporal dimensions.

[0136] By running this multi-scale data assimilation system, an analysis field is output, providing a spatially fused concentration distribution for the subsequent generation of a probabilistic concentration field.

[0137] The analysis field output by the multi-scale data assimilation system is coupled with the optimized superthreshold probability field obtained based on the corrected building wind field and plume data. Specifically, the optimized superthreshold probability field is used as a weight matrix to apply probabilistic constraints to the analysis field, generating a concentration spatial distribution modulated by prior probability, which is called the probabilistic concentration field.

[0138] Specifically, the probabilistic concentration field includes:

[0139] The analysis field output by the multi-scale data assimilation system is obtained, which is the optimal spatial distribution obtained from suspected low-altitude concentration data through data assimilation methods.

[0140] The analysis field is coupled with the optimized superthreshold probability field obtained from the modified building wind field and plume data. Specifically, the optimized superthreshold probability field is used as a weight matrix to impose probabilistic constraints on the analysis field.

[0141] By incorporating prior information of the optimized super-threshold probability field into the analysis field through probabilistic constraints, the concentration value of each grid cell in the analysis field is probabilistically modulated, thereby strengthening the concentration characteristics of high-probability regions and weakening the concentration performance of low-probability regions.

[0142] After probability modulation, a concentration spatial distribution modulated by prior probability is generated, namely the probabilistic concentration field. This probabilistic concentration field not only retains the core features of the assimilation results of the observation data, but also embeds the prior knowledge of spatial distribution provided by the optimized superthreshold probability field.

[0143] A high-fidelity computational fluid dynamics model is driven by the generated probabilistic concentration field. This model uses the modified building wind field as the initial boundary condition to simulate the dynamic transport and diffusion of pollutants in a complex building complex environment and outputs the simulated spatiotemporal concentration sequence.

[0144] Specifically, computational fluid dynamics models include:

[0145] Based on the actual industrial park building complex area studied in this invention, an accurate three-dimensional geometric model is established. This model must completely include the external shape, size, and spatial location of all relevant buildings. Subsequently, a computational domain that fully surrounds the building complex is defined. The dimensions of this computational domain in both the horizontal and vertical directions must be large enough to ensure that the boundary conditions do not cause artificial interference to the flow field in the core region of interest. The core region of interest includes the interior of the building complex and the low-altitude region where the sampling height of the pollution monitoring equipment is adjusted based on the optimized overthreshold probability field.

[0146] The established computational domain is discretized into a computational grid for computational fluid dynamics solutions. To ensure high fidelity, this grid is locally refined in key areas, particularly building surfaces, rooftops, and the low-altitude region where the pollution monitoring equipment's sampling height is determined based on an optimized overthreshold probability field. The grid density is verified for grid independence to ensure that the final calculation results do not depend on the grid size.

[0147] We selected a physical model and governing equations to solve the physical governing equations describing fluid motion and scalar transport. These include the transient Navier-Stokes equations for solving the unsteady velocity and pressure fields of air, and a closed set of equations using a turbulence model to simulate the complex turbulence generated around a building; an energy equation for solving the temperature field, which is crucial for simulating thermally buoyant-driven flow caused by the heat island effect; and a species transport equation for solving the concentration field of pollutants. The convection and diffusion terms in this equation are directly driven by the solved velocity field and turbulence parameters.

[0148] The inlet boundary conditions utilize building wind field data corrected using a new spatiotemporal graph network. Specifically, the wind speed, direction, and turbulence parameters at the computational domain inlet from this data are used as time-varying inlet conditions for the computational fluid dynamics model. No-slip wall conditions are applied to all building surfaces and the ground, and wall functions are used to handle turbulence effects in the near-wall region. Building surface temperature conditions are set based on heat island intensity data. The ground and building wall temperatures or heat fluxes are set based on heat island intensity and solar irradiance data to accurately simulate the thermal effects driving thermal circulation. The probabilistic concentration field from the realism assessment process is used as the initial field for the pollutant concentration equation in the computational fluid dynamics model. Simultaneously, based on production verification data from a standardized multi-source database, corresponding pollutant emission source terms are set at the identified pollution source locations. The initial flow field of the entire computational domain is initialized by interpolation using the corrected building wind field data to accelerate computational convergence.

[0149] Select a solver suitable for incompressible, transient, turbulent, and buoyancy-driven flows; set a high-precision numerical discretization scheme and time step to capture the transient characteristics of flow and diffusion. Then, start transient calculation iterations until the flow field and concentration field reach quasi-steady state or the simulation is completed for a predetermined physical time, and output the spatiotemporal sequence of simulated concentration.

[0150] The simulated concentration spatiotemporal sequence output by the computational fluid dynamics model is substituted back into the multi-scale data assimilation system as the background field for the second round of assimilation calculation. This process utilizes the physical consistency of the simulated data to correct for possible observation errors in the first round of assimilation, generating a reanalysis concentration field constrained by the physical model.

[0151] Design an uncertainty quantification module based on the bootstrap method; the uncertainty quantification module resamples the generated reanalysis concentration field to generate multiple statistically consistent concentration field sets, and by calculating the standard deviation among the members of the set, quantitatively depicts the spatial uncertainty distribution of the real low-altitude concentration data;

[0152] It should be noted that methods for obtaining key uncertainty areas based on uncertainty distribution maps include:

[0153] Uncertainty quantification index is extracted for each grid cell from the uncertainty distribution map. This index characterizes the confidence level of the reanalysis concentration field at that location.

[0154] The uncertainty quantification index of each grid cell is compared with the preset uncertainty threshold, and all grid cells with uncertainty quantification index greater than the preset uncertainty threshold are selected.

[0155] The selected high-uncertainty grid cells are clustered and connected in space to form continuous spatial patches, avoiding the dispersion of verification resources caused by isolated high points in existing technologies;

[0156] Based on the corrected building wind field data, the flow field location of these highly uncertain spatial patches is analyzed and determined, such as the downwind direction or the building bypass area, to further confirm their causes and verification priorities.

[0157] Based on the spatial patch area and flow field location, the final key uncertainty area is determined, and its spatial coordinates are output as the target area for deploying UAVs to collect vertical profile concentration data.

[0158] Specifically, the uncertainty distribution diagram includes:

[0159] Based on the reanalysis concentration field constrained by the physical model, the bootstrap resampling technique is used to generate hundreds of statistically consistent concentration field set members by randomly sampling the grid cell concentration values ​​in the reanalysis concentration field with replacement.

[0160] For each grid cell, the statistical standard deviation of the concentration values ​​of that grid cell across all members of the concentration field set is calculated, and the standard deviation of each grid cell is used as a quantitative indicator of the uncertainty at that location.

[0161] The uncertainty quantification indicators of all grid units are spatially visualized to form an uncertainty distribution map that covers the entire study area and can intuitively reflect the range of concentration value fluctuations at various points in space.

[0162] Based on the numerical distribution characteristics of the uncertainty distribution map, regions where the uncertainty quantification index is higher than a preset threshold are extracted and marked as key uncertainty regions, serving as key targets for subsequent external verification.

[0163] By overlaying and comparing the uncertainty distribution map with vertical profile concentration data obtained through an independent external validation data source, the degree of agreement between the key uncertainty areas marked by the uncertainty distribution map and the deviation areas of the measured data is verified, thus completing the accuracy assessment of the uncertainty quantification results.

[0164] An independent external validation data source is introduced: vertical profile concentration data collected by drones equipped with portable monitoring devices in key uncertainty areas. The vertical profile concentration data is overlaid and analyzed with the obtained uncertainty distribution map. If the vertical profile concentration data is within the confidence interval of the uncertainty distribution, the validation is passed, and the suspected low-altitude concentration data is the real low-altitude concentration data. If a systematic bias occurs, a re-correction process for building wind field and plume data is triggered, thus forming a closed-loop feedback from the validation results to the initial model parameters.

[0165] It should be noted that the methods for collecting vertical profile concentration data in key uncertainty areas include:

[0166] Based on the spatial coordinates of the key uncertainty area, plan the initial flight path of the UAV equipped with portable monitoring equipment. The initial flight path should cover the core area of ​​the key uncertainty area and include multiple hovering sampling points with different absolute altitudes.

[0167] By combining the corrected building wind field data, analyzing the instantaneous flow field characteristics in key uncertainty areas, and fine-tuning the initial flight path, a definite flight path is obtained, ensuring that the UAV is upwind of the main path of the pollution plume when collecting data.

[0168] Control the drone to climb along a predetermined flight path, hover at each preset absolute altitude, and activate portable monitoring equipment to collect pollutant concentration data at the corresponding horizontal position and vertical height, forming a vertical profile concentration data sequence;

[0169] While the drone collects data, it simultaneously records the GPS coordinates, absolute altitude and corresponding timestamp of each sampling point to ensure that the vertical profile concentration data and the reanalysis concentration field are fully aligned in spatiotemporal reference.

[0170] The collected raw vertical profile concentration data is transmitted back to the ground station in real time. The data is then processed by the data parsing module into a standardized format required for overlay analysis with the uncertainty distribution map, thus completing the preparation of the external validation data source.

[0171] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention based on the pollution emission monitoring and early warning method driven by physical prior knowledge and information value, and the inventive concept thereof, should be covered within the scope of protection of the present invention.

Claims

1. A pollution emission monitoring and early warning method based on physical priori and information value driving, characterized in that, The method comprises the following steps: Collecting production side data, performing time alignment and self-calibration de-drifting, and constructing a standardized multi-source database; Judging whether the heat island effect will occur, if so: Collecting single building volume heat capacity, solar irradiance and heat island intensity data, embedding the standardized multi-source database, single building volume heat capacity, solar irradiance and heat island intensity data into a space-time graph network to obtain a new space-time graph network; Based on the new space-time graph network, the building wind field and plume data are corrected; based on the corrected building wind field and plume data, the adjusted threshold probability field is obtained; based on the obtained adjusted threshold probability field and real-time pollution concentration data, the optimized threshold probability field is obtained; Based on the optimized threshold probability field, the collection height of the pollution monitoring equipment is adjusted, and the suspected low-altitude concentration data is collected by the pollution monitoring equipment; the authenticity of the suspected low-altitude concentration data is judged; if correct, the suspected low-altitude concentration data is the real low-altitude concentration data; Based on the real low-altitude concentration data, the optimal scheduling scheme is generated. 2.The method of monitoring and early warning of pollution emission based on physical priori and information value driving according to claim 1, characterized in that, The method for embedding the standardized multi-source database, single building volume heat capacity, solar irradiance and heat island intensity data into a space-time graph network to obtain a new space-time graph network comprises: Obtaining a building attribute list; based on the building attribute list, calculating the single building volume heat capacity; Based on the ultrasonic wind field data and fixed station data and spatial coordinates in the standardized multi-source database, combined with real-time heat island intensity data and solar irradiance data; obtain a three-dimensional data set; Map the heat island intensity data and solar irradiance data in the three-dimensional data set to the wind shear correction coefficient; Then input the single building volume heat capacity data and the three-dimensional data set into the space-time graph network to obtain a new space-time graph network. 3.The method of claim 1, wherein, The method for correcting the building wind field and plume data based on the new space-time graph network comprises: Obtain the ultrasonic wind field and pollution concentration data of the standardized multi-source database, combine the single building volume heat capacity, real-time heat island intensity and solar irradiance data to obtain the space-time graph network input tensor; Taking the building and monitoring point as the node and the spatial distance as the edge, embedding the law of conservation of mass and the Gaussian diffusion mechanism in the graph convolution layer, capturing the time dynamic characteristics of the wind field and the plume through the time convolution layer to obtain the basic wind field vertical component; Based on the new space-time graph network, real-time heat island intensity and solar irradiance data, the wind shear correction coefficient is obtained; based on the wind shear correction coefficient and the basic wind field vertical component, the corrected building wind field data is obtained; Based on the corrected building wind field data, the pollution source emission amount of the standardized multi-source database and the single building volume heat capacity, the plume data is calculated.

4. The method of claim 1, wherein the method is characterized by, The method for judging whether the heat island effect will occur comprises: Obtain the real-time wind speed data and wind direction data in the standardized multi-source database, calculate the average wind speed, and compare the average wind speed with the wind speed threshold based on the average wind speed; If the average wind speed is less than the wind speed threshold, the corresponding period is a weak wind period; Collecting the building surface temperature and the ambient temperature in the weak wind period, calculating the difference between the building surface temperature and the ambient temperature to obtain the real-time heat island intensity; In the same preset time window as calculating the real-time heat island intensity, collect the solar radiation data; based on the solar radiation data in the window, the irradiance sudden rise amplitude is calculated; Comparing the real-time heat island intensity with the intensity threshold value and comparing the irradiance sudden increase amplitude with the amplitude threshold value; If the real-time heat island intensity is greater than the intensity threshold value and the irradiance sudden increase amplitude is greater than the amplitude threshold value, the heat island effect occurs.

5. The method of claim 1, wherein the method is characterized by, The method for obtaining the adjusted super-threshold probability field based on the corrected building wind field and plume data comprises: Dividing the region where the building wind field and plume data are located into a plurality of grid cells, and labeling the building position and pollution source coordinates of each grid cell; Based on the corrected building wind field and plume data, the hourly predicted concentration of each grid cell is calculated based on the building position and pollution source coordinates of each grid cell; the super-threshold probability of each grid cell is calculated based on the hourly predicted concentration of each grid cell and the preset super-threshold concentration threshold; and the adjusted super-threshold probability field is obtained.

6. The method of monitoring and early warning of pollution emission based on physical priori and information value driving according to claim 1, characterized in that, The method for obtaining the optimized super-threshold probability field based on the obtained adjusted super-threshold probability field and real-time pollution concentration data comprises: The real-time pollution concentration data is divided according to the grid cells of the adjusted super-threshold probability field and matched with the time stamp, and the abnormal values in the concentration data are removed to ensure that the two types of data are completely aligned in time and space dimensions, and the spatio-temporal aligned super-threshold probability field is obtained; Based on the spatio-temporal aligned super-threshold probability field, an initial sample set is obtained; the initial sample set is input into a new spatio-temporal graph network, and the predicted concentration field corresponding to each initial sample is output; based on the predicted concentration field of each initial sample and the real-time pollution concentration data, an initial error is calculated; Based on the real-time pollution concentration data, an observation error matrix is constructed, and based on the initial error, a model error matrix is constructed; based on the observation error matrix and the model error matrix, the grid cell super-threshold probability value of each initial sample is adjusted to obtain the optimized super-threshold probability field.

7. The method of monitoring and early warning of pollution emissions based on physical priori and information value driving according to claim 6, characterized in that, The method for adjusting the collection height of the pollution monitoring equipment based on the optimized super-threshold probability field comprises: Based on the optimized super-threshold probability field, grid cells greater than a preset probability threshold are extracted to obtain a super-threshold spatial region, and vertical pollution distribution characteristic data of the super-threshold spatial region are collected; based on the vertical pollution distribution characteristic data, the initial vertical barycenter position of the pollution plume in the super-threshold spatial region is calculated; Based on the corrected building wind field data and the initial vertical barycenter position based on the corrected building wind field data, a corrected plume vertical barycenter position is obtained; and based on the corrected plume vertical barycenter position, the collection height of the pollution monitoring equipment is obtained. 8.The method of monitoring and early warning of pollution emission based on physical priori and information value driving according to claim 1, characterized in that, The method for judging the authenticity of the suspected low-altitude concentration data comprises: Based on the suspected low-altitude concentration data, a multi-scale data assimilation system is constructed, and an analysis field is output; Based on the optimized super-threshold probability field and the analysis field, a probabilistic concentration field is obtained; based on the probabilistic concentration field, a simulated concentration spatio-temporal sequence is obtained, and based on the simulated concentration spatio-temporal sequence, a reanalyzed concentration field is obtained; Based on the reanalyzed concentration field, a resampling is performed to obtain a spatial uncertainty distribution map of the real low-altitude concentration data, and based on the uncertainty distribution map, a key uncertainty region is obtained; Vertical profile concentration data of the key uncertainty region are collected; if the vertical profile concentration data are within the confidence interval of the uncertainty distribution, the suspected low-altitude concentration data are real low-altitude concentration data. 9.The method of monitoring and early warning of pollution emission based on physical priori and information value driving according to claim 8, characterized in that, The method for obtaining the key uncertainty region based on the uncertainty distribution map comprises: Based on the uncertainty distribution map, an uncertainty quantification index of each grid cell is obtained; The uncertainty quantification index of each grid cell is compared with a preset uncertainty threshold; grid cells with an uncertainty quantification index greater than the preset uncertainty threshold are screened out; The screened grid cells are spatially clustered and connected to obtain continuous spatial patches; Based on the corrected building wind field data, the flow field position of the spatial patch is determined; Based on the spatial patch area and the flow field position, a key uncertainty area is determined.

10. The method of claim 8, wherein the method is characterized by, The method for collecting vertical profile concentration data of the key uncertainty area comprises: Based on the spatial coordinates of the key uncertainty area, an initial flight route of the unmanned aerial vehicle is obtained; Based on the corrected building wind field data, the instantaneous flow field characteristics in the key uncertainty area are obtained; Based on the instantaneous flow field characteristics, the initial flight route is adjusted to obtain a determined flight route; The unmanned aerial vehicle climbs along the determined flight route to collect pollutant concentration data corresponding to the horizontal position and vertical height, and obtains the vertical profile concentration data.

Images