High-precision atmospheric pollution prediction method

Abstract

The present application belongs to the technical field of environmental protection monitoring, and particularly relates to a high-precision atmospheric pollution prediction method, comprising the following steps: S1, dividing a monitoring area into a plurality of monitoring units; S2, marking potential atmospheric pollutant emission sources as fixed monitoring points; S3, determining random monitoring points in each monitoring unit; S4, constructing an atmospheric pollutant diffusion model for the monitoring area; S5, obtaining real-time atmospheric pollutant data of the fixed monitoring points and the random monitoring points, inputting the atmospheric pollutant diffusion model, and obtaining a prediction result of atmospheric pollution in the monitoring area; and S6, determining a prediction deviation of the atmospheric pollutant diffusion model based on current monitoring data and historical atmospheric pollution prediction results, and correcting the prediction result according to the prediction deviation. The atmospheric pollution prediction method has the advantages of simple and reasonable model structure, high prediction accuracy, strong timeliness, strong migration ability, and wide application range.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring technology, specifically relating to an air pollution monitoring technology, and more particularly to a high-precision air pollution prediction method. Background Technology

[0002] Air pollutants refer to substances released into the atmosphere due to human activities or natural processes that have harmful effects on humans and the environment. Although the composition of gases in clean, dry air is relatively stable, within a certain area, the presence of trace amounts of harmful substances that were not originally present or whose concentrations are significantly increased, and whose quantity and duration reach a certain level, can cause direct or potential harm to humans, animals, plants, and materials. When the concentration of pollutants in the atmosphere reaches or exceeds the hazard threshold, thereby disrupting the stability of the ecosystem and threatening the normal conditions for human survival and development, it constitutes air pollution.

[0003] Air pollution not only affects human physiological health, such as the respiratory and cardiovascular systems, but can also induce mental health problems, leading to social issues such as reduced work efficiency, decreased creativity, and decreased well-being. Therefore, it has attracted significant public attention. The migration and diffusion of pollutants in the atmosphere exhibits significant spatial heterogeneity and temporal dynamics. Their concentration changes are driven by multiple factors, including the accumulation and dilution of local pollutants, regional meteorological conditions, geographical features, and the transport of pollutants from surrounding areas. Therefore, constructing a high-precision, high-timeliness pollutant monitoring and diffusion prediction technology system, and promptly regulating air pollution, is a crucial means of addressing the air pollution problem.

[0004] Currently, the monitoring and prediction of air pollutants mainly rely on two types of technical systems: one is conventional monitoring technology, represented by fixed ambient air quality monitoring stations, which is mainly used to collect basic data such as pollutant concentrations and meteorological parameters over a long period of time at specific geographical locations; the other is pollutant migration and diffusion prediction technology based on meteorological driving and geographical factor modeling. This type of technology combines factors such as pollution source emissions, wind speed and direction, and topography to construct simulation models of the spatial-temporal distribution of pollutants, which can assist in the assessment and trend prediction of regional pollution situations. Traditional fixed monitoring methods have advantages such as high monitoring accuracy and good data continuity, but due to their limited number of monitoring points and insufficient spatial coverage, they are difficult to fully reflect the dynamic migration process of pollutants in the atmosphere over a large area, and generally lack the ability to couple with topography and real-time meteorological conditions, resulting in a lag or excessive deviation in the reflection of pollutant diffusion trends. On this basis, predictive modeling methods have been developed, which use gridded modeling and remote sensing data fusion, combined with meteorological factors and topographic data, to construct simulation models of pollutant diffusion paths in the spatiotemporal dimensions, representing the modern development direction of pollution prediction technology. However, these models still suffer from problems such as complex data processing, incomplete consideration of predictive factors, limited model accuracy, and bloated model structure, which restrict their timeliness and applicability in actual environmental monitoring and emergency response. Specifically, current predictive modeling methods mainly suffer from the following issues:

[0005] First, the data processing volume is large and the complexity is high: It is known that the accuracy and precision of the prediction model are directly related to the sample size and the number of monitoring stations. More stations can provide pollution data for more areas, enabling the model to learn spatial heterogeneity and improve prediction accuracy and precision. However, the number of existing air quality monitoring stations is limited, which cannot provide enough sample data for the prediction model. Therefore, existing high-precision prediction models mostly rely on remote sensing images to uniformly and continuously cover a large monitoring area, and then divide the images into multiple grids and perform calculations and analyses one by one to increase the sample size and sample coverage. However, this method has the drawback of large data processing volume and high complexity. This method not only requires the acquisition of a large number of images, but also requires highly complex processing such as image stitching, coordinate correction, spatial inversion, radiometric correction, cloud detection and declouding, and image enhancement after the images are transmitted back to the ground. This method not only leads to a huge amount of data and a cumbersome processing process, but also places extremely high demands on the performance of computing equipment and data communication bandwidth, thereby limiting the processing efficiency and real-time prediction capabilities of the model, making it difficult to meet the application requirements of rapid response to pollutant migration in large-scale areas.

[0006] Second, the predictive factors are not comprehensively considered, resulting in limited model precision and poor prediction accuracy. Most current models primarily focus on the impact of conventional meteorological and geographical elements such as topographic relief, wind speed and direction, and precipitation on pollutant dispersion. However, in reality, the dispersion of air pollutants is influenced by many other factors, including atmospheric stability, atmospheric temperature and temperature inversion layers, relative humidity, solar radiation, sea and land breezes, the urban heat island effect, the intensity and height of pollution source emissions, urban building layout, the molecular weight and density of pollutants, the water solubility and reactivity of pollutants, the relative content of primary and secondary pollutants, and land cover type. For example, existing predictive models generally neglect the moderating effect of land cover type on pollutant dispersion, especially the adsorption, retention, and transformation capabilities of vegetation systems such as mountain forests. The simplification and neglect of factors influencing pollutant dispersion in predictive models can lead to biases in the simulation, reducing the accuracy and adaptability of the predictive models.

[0007] Third, the model structure is redundant and the computational efficiency is low: In order to improve the stability and adaptability of the model in complex scenarios, existing air pollution prediction models often rely on introducing a large number of feature parameters and redundant modules. However, this approach leads to a large model size, computational complexity, decreased operating efficiency, and a tendency to overfit. It also results in poor generalization ability and is not conducive to efficient deployment and application in real-world scenarios.

[0008] Based on this, in order to improve the accuracy and timeliness of modern modeling techniques in predicting pollutant migration and diffusion processes, those skilled in the art have conducted multi-directional technical explorations, such as constructing multi-objective prediction models and using multi-pollutant collaborative prediction mechanisms to identify nonlinear coupling relationships between pollutants; enabling models to have adaptive update capabilities through online learning or incremental learning mechanisms; and improving the transferability of models in different cities or regions and enhancing the cross-regional generalization ability of models through transfer learning and graph neural networks. Among them, the most representative research result is the WRF-Chem+LSTM method. This method is a hybrid prediction method that combines the physical mechanism model (WRF-Chem) with the data-driven model (LSTM). It first uses WRF-Chem for preliminary prediction to obtain the predicted values ​​of air pollutant concentrations for future time periods. Then, it introduces the LSTM model for error learning and correction. Based on historical observation data and WRF-Chem prediction results, the LSTM is trained to identify the systematic bias of WRF-Chem and generate corrected predictions that are closer to the true values. In essence, it performs post-processing or error correction on the output of WRF-Chem. LSTM is used to correct WRF-Chem. Although LSTM can significantly improve prediction accuracy, it also has some technical limitations and potential drawbacks:

[0009] (1) LSTM models need to be retrained frequently and have poor timeliness: air pollution is significantly affected by factors such as season, emission sources, and policy intervention. LSTM models are sensitive to the distribution of training data. If external conditions change, the model may fail. Therefore, it is necessary to continuously update the training set and retrain to maintain accuracy, which results in high maintenance costs.

[0010] (2) Difficulty in cross-regional transfer of LSTM models: It is difficult to directly apply an LSTM model trained in one city or region to another region. Data needs to be collected and retrained for different geographical environments and emission structures. The transfer learning of LSTM models is still in the early stages of research.

[0011] (3) The LSTM model has the risk of “error amplification”: If the WRF-Chem prediction itself is abnormal, the LSTM may “learn” this error trend, overfit a certain type of deviation of WRF-Chem, and further amplify the error.

[0012] In summary, existing air pollution monitoring and forecasting technologies still have many shortcomings in terms of data processing efficiency, comprehensiveness of predictive factors, and rationality of model structure. Furthermore, there are numerous limitations to post-correction techniques for forecasting models, making it difficult to achieve efficient, low-cost, and rapid pollutant dispersion forecasting while ensuring forecast accuracy. Therefore, how to improve model forecast accuracy while effectively reducing data processing complexity and enhancing the timeliness and practicality of the system has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0013] The purpose of this invention is to address the aforementioned technical problems by providing a high-precision air pollution prediction method that can improve the accuracy of model predictions while effectively reducing data processing complexity, minimizing feature parameters and redundant modules, and enhancing the timeliness and practicality of the system.

[0014] In view of this, the present invention provides a high-precision air pollution prediction method, comprising the following steps:

[0015] S1, the monitoring area is divided into several monitoring units based on land use, land cover type and / or vegetation distribution data;

[0016] S2, identify potential sources of air pollutant emissions in the monitoring area and mark them as fixed monitoring points;

[0017] S3. Based on the accuracy requirements for air pollutant prediction, the number of random monitoring points in each monitoring unit is determined according to the area of ​​each monitoring unit and the gas diffusion variation factor. Random monitoring points and fixed monitoring points are coupled to establish a high-precision monitoring network.

[0018] S4. Construct an atmospheric pollutant diffusion model for the monitoring area based on the geographical information, historical meteorological information, and historical pollutant information of the fixed and random monitoring points.

[0019] S5. Based on remote sensing images of fixed and random monitoring points, real-time air pollutant data of fixed and random monitoring points are obtained and input into the air pollutant diffusion model to predict future air pollution and obtain the prediction results of air pollution in the monitoring area for a period of time in the future.

[0020] S6. Based on current monitoring data and historical air pollution prediction results, determine the prediction bias of the air pollutant diffusion model, and correct the air pollution prediction results for the monitoring area in the future period according to the determined prediction bias, so as to obtain the corrected air pollution prediction results.

[0021] Furthermore, step S2 specifically includes:

[0022] S21, Obtain or construct geographic information of the monitoring area;

[0023] S22, Identify and classify potential sources of air pollution;

[0024] S23 identifies potential sources of air pollutant emissions as candidate sites for fixed monitoring points;

[0025] S24, merge and screen the candidate fixed monitoring points to determine the final fixed monitoring points;

[0026] S25, Determine the specific location of the fixed monitoring point.

[0027] Furthermore, in step S3, the process of determining the number of random monitoring points in each monitoring unit is as follows:

[0028] (1) Establish a dynamic assessment mechanism for gas diffusion variation factors, wherein the gas diffusion variation factor of each monitoring unit = standard deviation of historical pollutant content / average value of historical pollutant content;

[0029] (2) Determine the basic number of random monitoring points for each monitoring unit based on the magnitude of the gas diffusion variation factor;

[0030] (3) Adjust the basic number of random monitoring points based on the area of ​​each monitoring unit.

[0031] Furthermore, in step S5, after obtaining the predicted air pollution situation in the monitoring area for a future period, the predicted air pollution situation is first analyzed and processed. The specific process includes the following steps:

[0032] S51, Obtain the pollutant distribution prediction results and determine the pollutant distribution boundary and center;

[0033] S52, starting from the pollutant distribution center and ending at the pollutant distribution boundary line, draw a pollutant distribution vector map C corresponding to the predicted future air pollution situation.

[0034] Furthermore, step S6 specifically includes:

[0035] S61, Obtain the atmospheric pollution situation prediction results corresponding to the current moment from the output of the atmospheric pollutant diffusion model;

[0036] S62, Obtain the measured results of air pollution at the current moment based on remote sensing image monitoring;

[0037] S63, analyze the predicted air pollution situation and the measured air pollution situation at the current time respectively, determine the pollutant distribution boundary and center, and then draw the pollutant distribution vector map A corresponding to the current air pollution situation and the pollutant distribution vector map B corresponding to the current air pollution situation, with the pollutant distribution center as the starting point and the pollutant distribution boundary line as the ending point.

[0038] S64, Determine the prediction bias of the atmospheric pollutant diffusion model based on the results of pollutant distribution vector maps A and B;

[0039] S65, based on the determined prediction deviation, correct the prediction results of air pollution in the monitoring area for a future period of time to obtain the corrected prediction results of air pollution.

[0040] Furthermore, in steps S52 and S63, the drawn pollutant distribution vector map includes multiple one-to-one corresponding sub-vectors with the same angle. The sub-vectors are drawn at predetermined angular intervals, with the pollutant distribution center as the starting point and the pollutant distribution boundary line as the ending point.

[0041] Furthermore, in step S64, the process of determining the prediction bias of the atmospheric pollutant diffusion model based on the pollutant distribution vector map results A and B is as follows:

[0042] S641, calculate the deviation coefficient α of the pollutant distribution area in the pollutant distribution vector maps A and B, wherein the deviation coefficient α = (1-K)*100%, and K is the overlap ratio, which is K = the overlap area of ​​vector maps A and B / the total area of ​​the pollutant distribution area in map B.

[0043] S642, compare the deviation coefficient α and the preset threshold α 阈 Relative size:

[0044] If α < α阈 Instead of correcting the air pollution prediction results for the next period output by the air pollutant diffusion model, step S65 is executed to directly output the air pollution prediction results of the air pollutant diffusion model.

[0045] If α≥α 阈 The atmospheric pollution forecast results for the next period output by the atmospheric pollutant diffusion model need to be corrected, and step S643 should be continued.

[0046] S643, using the center point of the pollutant distribution area in pollutant distribution vector map A as the starting point and the center point of the pollutant distribution area in pollutant distribution vector map B as the ending point, draw the prediction deviation vector between the center points of the pollutant distribution areas in pollutant distribution vector maps A and B. ;

[0047] Then along the vector The pollutant distribution vector map A is translated from the starting point to the ending point to obtain the translated pollutant distribution vector map A';

[0048] Calculate the differences between each component vector in the pollutant distribution vector map B and the pollutant distribution vector map A' respectively, to obtain a set of component vector deviation vectors corresponding to each angle.

[0049] And use the calculated prediction deviation vector between the center points , and the deviation vector This indicates the prediction bias of the atmospheric pollutant diffusion model.

[0050] Furthermore, step S65 includes:

[0051] When α < α 阈 At that time, the air pollution prediction results of the air pollutant diffusion model are directly output;

[0052] When α≥α 阈 First, the center point of the pollutant distribution vector diagram C is aligned with the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. The corrected component vectors are obtained by adding them together. The boundaries of the pollutant distribution vector map C are then adjusted according to the corrected component vectors to obtain the pollutant distribution vector map C'. The pollutant distribution vector map C' is then output as the final prediction result of air pollution.

[0053] Furthermore, step S65 also includes: when α ≥ α 阈 Then, determine whether a deep correction is needed for the prediction results:

[0054] First, the pollutant distribution vector map A is aligned with the prediction deviation vector. The pollutant distribution vector map A' is translated to obtain a vector map of pollutant distribution A', such that the center point of the translated vector map A' coincides with the center point of the pollutant distribution vector map B. Then, the deviation coefficient α' of the pollutant distribution area in the vector maps A' and B is calculated, where the deviation coefficient α' = (1-K')*100%, and K' is the overlap ratio, which is equal to the overlap area of ​​vector maps A' and B / the total area of ​​the pollutant distribution area in map B.

[0055] Then compare the deviation coefficient α' with the preset threshold α. 阈 Relative size of ':

[0056] If α' < α 阈 ', No depth correction is required for the pollutant distribution vector map C;

[0057] If α'≥α 阈 ', It is necessary to perform depth correction on the pollutant distribution vector map C.

[0058] Furthermore, the depth correction process is as follows:

[0059] First, the boundary line of the pollutant distribution vector map B is analyzed to obtain the extreme points on the boundary line. Then, the extreme value vectors of the pollutant distribution vector map B are drawn with the center point of the pollutant distribution vector map B as the starting point and each extreme point on the boundary line as the ending point. The vector set composed of the extreme value vectors is called the extreme value vector set.

[0060] Then, draw the extreme value vectors on the pollutant distribution vector map A, which correspond one-to-one with the angles of the extreme value vectors;

[0061] Next, the differences between the extreme value vectors in pollutant distribution vector map B and pollutant distribution vector map A are calculated respectively, resulting in a set of extreme value deviation vectors corresponding to each extreme value.

[0062] In α'≥α 阈 First, draw the extreme value vectors on the pollutant distribution vector map C, which correspond one-to-one with the angles of the extreme value vectors. Then, draw the center point of the pollutant distribution vector map C along the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. After addition, the corrected component vector is obtained. The extreme value vector in the pollutant distribution vector map C is then compared with the corresponding extreme value deviation vector. The corrected extreme value vector is obtained by adding them together;

[0063] The boundaries of the pollutant distribution vector map C are adjusted according to the corrected component vectors and extreme value vectors to obtain the pollutant distribution vector map C', and the pollutant distribution vector map C' is output as the final prediction result of air pollution.

[0064] The beneficial effects of the present invention are: the high-precision air pollution prediction method of the present invention has the advantages of simple and reasonable model structure, high prediction accuracy, strong timeliness, strong transferability and wide applicability. Attached Figure Description

[0065] Figure 1 This is a flowchart of the high-precision air pollution prediction method described in this invention;

[0066] Figure 2 This is a schematic diagram of the pollutant distribution vector diagrams A and B described in this invention;

[0067] Figure 3 This is a comparison chart of the prediction results, measured values, and prediction results of the prior art in the embodiments of the present invention. Detailed Implementation

[0068] The technical solutions of this application will be clearly described below with reference to specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0069] In the description of this application, it should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. For ease of description, techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification.

[0070] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0071] A high-precision method for predicting air pollution, such as Figure 1 As shown, the steps include:

[0072] S1, the monitoring area is divided into several monitoring units based on land use, land cover type and / or vegetation distribution data;

[0073] S2, identify potential sources of air pollutant emissions in the monitoring area and mark them as fixed monitoring points;

[0074] S3. Based on the accuracy requirements for air pollutant prediction, the number of random monitoring points in each monitoring unit is determined according to the area of ​​each monitoring unit and the gas diffusion variation factor. Random monitoring points and fixed monitoring points are coupled to establish a high-precision monitoring network.

[0075] S4. Construct an atmospheric pollutant diffusion model for the monitoring area based on the geographical information, historical meteorological information, and historical pollutant information of the fixed and random monitoring points.

[0076] S5. Based on remote sensing images of fixed and random monitoring points, real-time air pollutant data of fixed and random monitoring points are obtained and input into the air pollutant diffusion model to predict future air pollution and obtain the prediction results of air pollution in the monitoring area for a period of time in the future.

[0077] S6. Based on current monitoring data and historical air pollution prediction results, determine the prediction bias of the air pollutant diffusion model, and correct the air pollution prediction results for the monitoring area in the future period according to the determined prediction bias, so as to obtain the corrected air pollution prediction results.

[0078] Furthermore, in step S1, the monitoring area can first be divided according to land use, surface cover type and / or vegetation distribution data. For example, it can be divided into construction land, forest land, water land, cultivated land, etc. according to land use; it can be divided into artificial surface, farmland, forest land, grassland, wetland, water body, bare land, snow / glacier, etc. according to surface cover type; and it can be divided into forest, shrubland, grassland, cultivated land, wetland vegetation, bare land, glacier, etc. according to vegetation distribution data.

[0079] Preferably, in step S1, when dividing the monitoring units, the differences in topography should also be considered, and each monitoring unit should have the same or similar topography.

[0080] The diffusion of air pollutants exhibits strong spatial heterogeneity, with different landforms and land types directly influencing their diffusion rate, migration path, and deposition mode. In this invention, the smallest unit of the monitoring process is formed by classifying and dividing the monitoring units. This division is based not only on topography but also comprehensively considers the actual surface properties, giving each monitoring unit ecological and geographical significance. This allows for a more realistic reflection of the differences in pollutant diffusion behavior on different underlying surfaces, such as water bodies, forests, urban roads, and bare land, thereby improving the geographical adaptability and physical fit of the atmospheric pollutant diffusion model constructed accordingly.

[0081] Most importantly, these monitoring units have spatial and topographical consistency, resulting in a high degree of consistency in their impact on the diffusion process of air pollutants. This high degree of spatial and topographical consistency reduces the uncertainty of the diffusion process, making the dynamic mechanism of air pollutant diffusion within each monitoring unit more consistent, and reducing prediction errors caused by the heterogeneity of the regions within the monitoring unit.

[0082] Building upon this foundation, by using monitoring units as the smallest modeling unit, large-scale areas can be modularly divided and processed, reducing data and computational redundancy caused by unified modeling across the entire region. This supports parallel computing and local updates, significantly improving the system's prediction efficiency and real-time response capabilities. It also avoids unnecessary remote sensing image processing and global inversion, enhancing the model's timeliness and scalability, and adapting to the deployment needs of large areas and multiple scenarios. Furthermore, these monitoring units, based on ecological and geographical attributes, have clear semantic boundaries, facilitating subsequent pollutant source analysis, risk warning, pollution control strategy formulation, and refined policy implementation and regionally differentiated management.

[0083] Preferably, in step S1, multiple monitoring unit division schemes can be formed based on land use, surface cover type and / or vegetation distribution data, and then the optimal monitoring unit division scheme can be determined as needed.

[0084] It should be noted that when determining the optimal monitoring unit division scheme, users can freely choose according to their needs and priorities. For example, the computational efficiency, accuracy, and stability of the model can be used as references for selection. The computational efficiency of the model can be measured by the time taken to complete a simulation or training. The accuracy of the model can be measured by how close the model's prediction results are to the actual situation or observation data, such as the deviation between the prediction and the actual situation. The stability of the model can be measured by the model's ability to resist interference.

[0085] Furthermore, step S2 specifically includes:

[0086] S21, Obtain or construct geographic information of the monitoring area: Obtain or construct geographic information of the monitoring area through satellite remote sensing images, DEM terrain elevation models, field surveying and mapping, etc. The geographic information includes: elevation data, land use, land cover type and vegetation distribution data, etc.

[0087] S22, Identify and classify potential air pollution sources: Potential air pollutant emission sources can be marked or identified by referring to local emission inventories, national pollution source census data, remote sensing identification methods, etc. The potential air pollutant emission sources mentioned in this invention include industrial production facilities such as factories, boilers, power plants, and waste incineration plants; transportation facilities such as highways and transportation hubs; and occasional facilities such as construction sites, wildfires, and volatile organic compounds from vegetation.

[0088] S23, mark potential air pollutant emission sources as candidate fixed monitoring points; for example, the location of air pollutant emission sources can be imported into the topographic map of the monitoring area, and attribute fields such as pollution source type, emission type, emission level, emission time characteristics, etc. can be added to each potential air pollutant emission source, and the pollution intensity can be visualized using the hierarchical symbol method, such as using large red dots to represent heavy pollution sources and small red dots to represent light pollution sources.

[0089] S24. Merge and screen candidate fixed monitoring points to determine the final fixed monitoring points: For example, based on the distance between the candidate fixed monitoring points, merge multiple fixed monitoring points that are close to each other into one fixed monitoring point; cancel fixed monitoring points with sparse population, low emissions, and low monitoring value; or refer to historical meteorological information, historical pollutant information, and topography of the monitoring area to analyze the convergence zone, ventilation corridor, and pollution transmission channel of heavily polluted areas, and cancel fixed monitoring points in areas with ultra-low air pollution risk, etc., and determine the final fixed monitoring points through manual or machine intelligent screening.

[0090] S25, Determine the specific location of fixed monitoring points: When setting up monitoring points, arrange the specific locations of each candidate fixed monitoring point near point emission sources, such as industrial parks, factories, boilers, power plants, waste incineration plants, construction sites, etc.; near densely populated or sensitive areas (such as schools and hospitals) near linear emission sources such as highways and transportation hubs; and downwind of area emission sources such as wildfires and volatile organic compounds from vegetation.

[0091] In this invention, by identifying, screening, and scientifically deploying fixed monitoring points through steps S21-S25, more targeted and effective technical optimizations can be achieved at the pollution source identification and monitoring network construction level. Starting from the actual distribution of pollution sources, a candidate set of fixed monitoring points is dynamically determined, achieving comprehensive coverage of typical emission sources. Furthermore, it no longer relies on fixed standard grid deployment, but rather on targeted deployment based on the characteristics of pollution source distribution. The labeling of attributes such as pollution source type, emission type, intensity, and time characteristics helps to build a more representative emission characteristic database, making hierarchical visualization possible. This facilitates further optimization of monitoring point screening and subsequent visual analysis, avoids blind deployment, and improves the regional adaptability of pollution identification and prediction models.

[0092] Based on this, by merging and screening candidate fixed monitoring points, redundant and low-value monitoring points can be effectively filtered out, and limited sampling resources can be deployed in the most representative and valuable areas, thereby reducing monitoring point redundancy, controlling deployment costs, and improving the resource utilization efficiency and economy of the overall monitoring system.

[0093] Furthermore, this invention also adopts a classification and distribution strategy for different types of emission sources, such as point, line, and area, which can fully reflect the differences in diffusion behavior of various pollution sources, improve the model's identification and tracing capabilities, and enhance the model's adaptability and spatiotemporal sensitivity to complex multi-source pollution scenarios.

[0094] Furthermore, in step S2, the establishment of fixed monitoring points enables continuous monitoring and dynamic tracking of key pollution sources in the monitoring area. This allows for real-time and continuous acquisition of pollutant concentration data, tracking of temporal changes in pollutant emissions, and accurate identification of major pollutant types and their sources. This provides high-quality data support for subsequent model calibration, pollutant diffusion prediction, and pollution source tracing analysis. Simultaneously, it enhances the accuracy of the initial and boundary conditions of the atmospheric pollutant diffusion model. Fixed monitoring points provide spatially distributed and temporally continuous data sources, serving as data input for the atmospheric pollutant diffusion model. This improves the model's responsiveness to actual pollutant migration, enabling rolling correction and dynamic optimization of pollution trends, reducing model prediction errors, and improving the accuracy and timeliness of early warnings. This overcomes the problem of large prediction biases caused by existing models relying on static inputs or incomplete meteorological drivers.

[0095] Furthermore, the establishment of fixed monitoring stations can support the precise triggering of pollution early warning mechanisms. When the pollutant concentration at a fixed monitoring station reaches or approaches a preset threshold, a short-term early warning mechanism can be automatically triggered, initiating the local pollution emergency response process. Simultaneously, the establishment of fixed monitoring stations can also support pollution source tracing, assessment, and decision analysis.

[0096] Furthermore, in step S3, based on the deployment of fixed monitoring points, the number of random monitoring points in each monitoring unit can be determined according to the accuracy requirements of atmospheric pollutant prediction, based on the area of ​​each monitoring unit and the gas diffusion variation factor. Then, the random monitoring points and fixed monitoring points are coupled and linked together to establish a high-precision monitoring network.

[0097] Specifically, the number of random monitoring points in each monitoring unit is dynamically determined based on the area of ​​the monitoring unit and the gas diffusion variation factor. The specific process is as follows:

[0098] (1) First, a dynamic assessment mechanism for gas diffusion variation factors is established, wherein the gas diffusion variation factor of each monitoring unit is equal to the standard deviation of historical pollutant content / the average historical pollutant content; the historical pollutant content can be the monthly or annual average pollutant content of the monitoring unit.

[0099] (2) Then, the basic number of random monitoring points in each monitoring unit is determined according to the magnitude of the gas diffusion variation factor. Generally, the larger the gas diffusion variation factor, the more random monitoring points are set in the monitoring unit. For example, when the gas diffusion variation factor is large, such as greater than 0.5, the number of random monitoring points in the monitoring unit can be appropriately increased, such as 6 to 10. When the gas diffusion variation factor is small, such as less than 0.2, the number of random monitoring points in the monitoring unit can be appropriately reduced, such as 1 to 3. When the gas diffusion variation factor is appropriate, such as between 0.2 and 0.5, the number of random monitoring points in the monitoring unit can be set to 4 to 5.

[0100] (3) Subsequently, the basic number of random monitoring points is adjusted based on the area of ​​each monitoring unit. Generally, when the area of ​​the monitoring unit is large, the number of random monitoring points can be increased appropriately; conversely, when the area of ​​the monitoring unit is small, the number of random monitoring points can be decreased appropriately. For example, the monitoring units can be classified according to their area size, such as large-area monitoring units (>100km). 2 Medium-sized monitoring units (50–100 km²) 2 Small-area monitoring units (<50km) 2Then, the basic number of random monitoring points is adjusted according to the classification. For example, the number of random monitoring points in a large-area monitoring unit can be adjusted to the basic number of monitoring points * (2 to 5), the number of random monitoring points in a medium-area monitoring unit can be adjusted to the basic number of monitoring points * (1.2 to 2.5), and the number of random monitoring points in a small-area monitoring unit can be adjusted to the basic number of monitoring points * (0.8 to 1).

[0101] Furthermore, the location of random monitoring points can be determined using methods such as random sampling models.

[0102] As some examples of the present invention, the location of random monitoring points can be generated by Monte Carlo random sampling algorithm to generate a set of monitoring sampling points that meet a preset number and spatial distribution constraints, wherein multiple monitoring points in the same monitoring unit tend to be evenly distributed.

[0103] Preferably, after generating the location and number of random monitoring points, the locations of the random monitoring points and fixed monitoring points can be compared to eliminate duplicate monitoring points caused by excessively close proximity.

[0104] In this invention, the introduction of random monitoring points can fill the gaps in fixed monitoring points in non-key areas or areas prone to pollution changes, improve the spatial continuity and accuracy of the prediction model, and at the same time, random point placement can provide the model with richer and more diverse training samples, reduce dependence on specific points, prevent overfitting, and improve generalization ability.

[0105] Furthermore, in this invention, a gas diffusion variation factor is used as the basis for point selection, which can accurately identify monitoring units with drastic pollution changes and conduct "intensified monitoring" on them, thereby improving the sensitivity and accuracy of the overall prediction. Simultaneously, this invention uses area and variation factor to jointly determine the number of random points, enabling the model to identify spatial heterogeneity, thus more accurately simulating the migration and diffusion trajectory of pollutants.

[0106] Based on this, the monitoring point extraction method adopted in this invention can combine the accuracy of fixed monitoring points with the wide coverage of random monitoring points, making up for the shortcomings of sparse distribution of fixed points, while avoiding redundant deployment and improving the economy of point deployment and information coverage.

[0107] As some examples of the present invention, in step S4, the geographic information includes elevation DEM data, road network and traffic density map, river and lake distribution, administrative boundaries and grid division, etc.; the historical meteorological information includes data such as wind speed and direction, solar radiation and sunshine duration, humidity, temperature and temperature gradient, precipitation and precipitation intensity, etc.; the historical pollutant information includes spatial distribution map / raster data of pollutants, pollution source location, emission intensity, pollution concentration, type, diffusion trajectory, distribution, etc. The specific composition of the geographic information, historical meteorological information and historical pollutant information can be selected according to the needs of the model construction.

[0108] Preferably, based on the post-correction mechanism of the present invention, when constructing the atmospheric pollutant diffusion model, the contribution of each feature parameter to the prediction result can be sorted, and then several feature parameters with larger contributions, such as 5 to 8 feature parameters, can be selected for training to construct the atmospheric pollutant diffusion model of the monitoring area. In this process, the influence of feature parameters with lower contributions on the prediction result can be ignored, and their influence on the prediction result can be uniformly reflected through the post-correction procedure, making the model structure more reasonable and concise, avoiding the introduction of too many feature parameters, which would lead to model structure redundancy, low computational efficiency, and poor generalization ability, making the model lighter, with lower training cost, and more reasonable structure.

[0109] Furthermore, in step S4, after acquiring historical meteorological and pollutant information of the monitoring area, the data can be preprocessed first. The data preprocessing process includes at least: data format unification (converting data from different sources, such as CSV, GeoTIFF, and NetCDF, into a uniformly processable format), time alignment (aligning meteorological and pollutant data to the same time resolution), spatial interpolation / rasterization (performing spatial interpolation on pollutant location data, such as Kriging and IDW, to generate a continuous pollution field), and data cleaning (removing outliers and processing missing values ​​using interpolation methods, etc.). After the data preprocessing, the obtained dataset is used to train a preset machine learning model to obtain an atmospheric pollutant diffusion model.

[0110] Furthermore, the detailed process of constructing the atmospheric pollutant diffusion model in step S4 has been described in the prior art and will not be repeated here.

[0111] It should be noted that this invention mainly focuses on optimizing the input parameters of the prediction model and correcting the prediction results of the model. Therefore, the specific type of atmospheric pollutant diffusion model used in this invention is not limited.

[0112] As some examples of the present invention, the atmospheric pollutant diffusion model can be a physical mechanism model based on aerodynamics and diffusion equations, such as the Gaussian plume model, the Lagrange model, the WRF-Chem model, etc.; a statistical and machine learning model based on historical data patterns, such as Random Forest, SVM, XGBoos, Kriging model, etc.; and a deep learning model or a multi-model fusion model.

[0113] Furthermore, based on the establishment of a high-precision monitoring network composed of random and fixed monitoring points, in step S5, only the air pollutant data of the fixed and random monitoring points need to be acquired from the remote sensing images. This ensures that representative data from each monitoring unit can be collected through precise selection of monitoring points, avoiding the large amount of data redundancy caused by the uniform, full-coverage data acquisition method in traditional approaches. It also reduces the difficulty of subsequent data processing, significantly improving the processing efficiency and timeliness of the air pollutant diffusion model described in this invention. In reality, during air pollutant concentration monitoring and prediction, the differences in air pollutants over short distances and small areas are often not significant. Therefore, it is not always necessary to achieve full coverage of the entire monitoring area with remote sensing images. Traditional grid methods often result in large data processing volumes, complex processes, and low efficiency due to the large image coverage area and the need for image stitching, which directly affects the cycle and timeliness of air pollution prediction.

[0114] In this application, by setting up fixed and random monitoring points and constructing a model based on the information from these points, the limitations of traditional fixed monitoring methods—such as limited monitoring points, insufficient spatial coverage, difficulty in comprehensively reflecting the dynamic migration process of pollutants in the atmosphere over a large area, and lack of coupling ability with topography and real-time meteorological conditions—are avoided. At the same time, the accuracy of predictions is also taken into account.

[0115] Furthermore, in step S5, after obtaining the predicted air pollution situation in the monitoring area for a future period, the predicted air pollution situation can first be analyzed and processed. The specific process includes the following steps:

[0116] S51, Obtain the pollutant distribution prediction results and determine the pollutant distribution boundary and center;

[0117] S52, starting from the pollutant distribution center and ending at the pollutant distribution boundary line, draw a pollutant distribution vector map C corresponding to the predicted future air pollution situation.

[0118] As some examples of the present invention, the process of determining the pollutant distribution boundary and center based on the pollutant distribution prediction results can be as follows: First, the pollutant concentration in each region is compared with a preset concentration threshold. Regions with pollutant concentrations exceeding the concentration threshold are polluted regions, while those with concentrations below the threshold are non-polluted regions. After determining the polluted regions, the centroid or geometric center of the polluted regions is calculated and used as the pollutant distribution center point of the region.

[0119] Furthermore, step S6 specifically includes:

[0120] S61, Obtain the atmospheric pollution situation prediction results corresponding to the current moment from the output of the atmospheric pollutant diffusion model;

[0121] S62, Obtain the measured results of air pollution at the current moment based on remote sensing image monitoring;

[0122] S63, analyze the predicted air pollution situation and the measured air pollution situation at the current time respectively, determine the pollutant distribution boundary and center, and then draw the pollutant distribution vector map A corresponding to the current air pollution situation and the pollutant distribution vector map B corresponding to the current air pollution situation, with the pollutant distribution center as the starting point and the pollutant distribution boundary line as the ending point.

[0123] S64, Determine the prediction bias of the atmospheric pollutant diffusion model based on the results of pollutant distribution vector maps A and B;

[0124] S65, based on the determined prediction deviation, correct the prediction results of air pollution in the monitoring area for a future period of time to obtain the corrected prediction results of air pollution.

[0125] Preferably, in steps S52 and S63, the drawn pollutant distribution vector map includes multiple one-to-one corresponding sub-vectors with the same angle, specifically as follows: Figure 2 As shown: Component vectors can be drawn along the four directions of east, west, south, and north, starting from the pollutant distribution center and ending at the pollutant distribution boundary line; or, with true north as 0°, component vectors can be drawn at intervals of 30°, 45°, or 60°. Figure 2 This shows the component vectors drawn at 45° intervals. It is understandable that the more component vectors drawn, the more accurate the prediction result will be after correction, but the computational load of the correction process will be greater. Therefore, the number of component vectors should be set appropriately, preferably 6 to 12.

[0126] It should be noted that in this invention, the historical moment corresponds to the previous monitoring cycle, the current moment corresponds to the current monitoring cycle, and a future period corresponds to the next monitoring cycle, with the duration of the interval between each monitoring cycle being equal.

[0127] As examples of the present invention, in step S62, the measured air pollution situation at the current moment obtained based on remote sensing image monitoring can be the measured air pollution situation obtained in real time by the present invention or others through full coverage of remote sensing images. When the present invention outputs the measured air pollution situation at the current moment using remote sensing images obtained from its own monitoring, the pollutant distribution area can be obtained according to the air pollution situation prediction result corresponding to the current moment output by the air pollutant diffusion model in step S61. Then, using the predicted pollutant distribution area as a reference, the remote sensing image coverage area is dynamically adjusted and appropriately expanded to obtain the measured air pollution situation that meets the needs of subsequent use. Specifically, it is advisable to be able to monitor and cover the actual pollutant clouds corresponding to the pollutant clouds in the prediction results. It is understood that the measured air pollution situation only needs to cover the polluted area, and it is not necessary to perform full coverage detection of the monitoring area.

[0128] Furthermore, in step S64, the process of determining the prediction bias of the atmospheric pollutant diffusion model based on the pollutant distribution vector map results A and B is as follows:

[0129] S641, calculate the deviation coefficient α of the pollutant distribution area in the pollutant distribution vector maps A and B, wherein the deviation coefficient α = (1-K)*100%, and K is the overlap ratio, which is K = the overlap area of ​​vector maps A and B / the total area of ​​the pollutant distribution area in map B.

[0130] S642, compare the deviation coefficient α and the preset threshold α 阈 Relative size:

[0131] If α < α 阈 If the deviation coefficient α is small, it indicates that the air pollution prediction result and the actual air pollution result at the current moment are highly consistent. Therefore, the air pollution prediction result for the next period output by the air pollutant diffusion model does not need to be corrected. Step S65 is then executed to directly output the air pollution prediction result of the air pollutant diffusion model.

[0132] If α≥α 阈 If the deviation coefficient α is large, it indicates that the consistency between the predicted air pollution situation at the current moment and the measured air pollution situation is low. It is necessary to correct the predicted air pollution situation for the next period output by the air pollutant diffusion model and continue to execute step S643.

[0133] S643, using the center point of the pollutant distribution area in pollutant distribution vector map A as the starting point and the center point of the pollutant distribution area in pollutant distribution vector map B as the ending point, draw the prediction deviation vector between the center points of the pollutant distribution areas in pollutant distribution vector maps A and B. ;

[0134] Then along the vector The pollutant distribution vector map A is translated from the starting point to the ending point to obtain the translated pollutant distribution vector map A';

[0135] Calculate the differences between each component vector in the pollutant distribution vector map B and the pollutant distribution vector map A' respectively, to obtain a set of component vector deviation vectors corresponding to each angle.

[0136] The prediction deviation vector between the center points calculated above , and the deviation vector This can represent the prediction bias of the atmospheric pollutant diffusion model.

[0137] As some examples of the present invention, the preset threshold α 阈 and subsequent α 阈 The size of α can be set according to the required precision. Generally, the higher the precision requirement, the larger the preset threshold α. 阈 and α 阈 The smaller the value of '.

[0138] Furthermore, step S65 includes:

[0139] When α < α 阈 At that time, the air pollution prediction results of the air pollutant diffusion model are directly output;

[0140] When α≥α 阈 First, the center point of the pollutant distribution vector diagram C is aligned with the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. The corrected component vectors are obtained by adding them together. The boundaries of the pollutant distribution vector map C are then adjusted according to the corrected component vectors to obtain the pollutant distribution vector map C'. The pollutant distribution vector map C' is then output as the final prediction result of air pollution.

[0141] Furthermore, in the process of developing the pollutant distribution vector map C', once the corrected component vectors are determined, the boundary lines of the pollutant distribution vector map C can be adjusted according to the corrected component vectors, and the boundary lines near the adjustment area can be smoothed.

[0142] As an example of the present invention, the above process describes in detail the correction process for a single pollutant cloud. When the pollutant distribution vector diagram contains multiple pollutant clouds, the boundaries and center points of each pollutant cloud can be determined separately according to the above process, and then corrected separately in a similar manner.

[0143] Furthermore, step S65 also includes: when α ≥ α 阈 When determining whether a deep correction is needed for the prediction results, the specific process is as follows:

[0144] First, the pollutant distribution vector map A is aligned with the prediction deviation vector. The pollutant distribution vector map A' is translated to obtain a vector map of pollutant distribution A', such that the center point of the translated vector map A' coincides with the center point of the pollutant distribution vector map B. Then, the deviation coefficient α' of the pollutant distribution area in the vector maps A' and B is calculated, where the deviation coefficient α' = (1-K')*100%, and K' is the overlap ratio, which is equal to the overlap area of ​​vector maps A' and B / the total area of ​​the pollutant distribution area in map B.

[0145] Then compare the deviation coefficient α' with the preset threshold α. 阈 Relative size of ':

[0146] If α' < α 阈 'No depth correction is needed for the pollutant distribution vector map C. In this case, the center point of the pollutant distribution vector map C can be first aligned with the vector...' After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. After adding them, the corrected component vector is obtained. The boundary of the pollutant distribution vector map C is adjusted according to the corrected component vector to obtain the pollutant distribution vector map C'. The pollutant distribution vector map C' is then output as the final prediction result of air pollution.

[0147] If α'≥α 阈 The pollutant distribution vector map C needs to be depth-corrected. The depth correction process is as follows:

[0148] First, the boundary line of the pollutant distribution vector map B is analyzed to obtain the extreme points on the boundary line. The extreme points include the maximum and minimum values. Then, the extreme value vectors of the pollutant distribution vector map B are drawn with the center point of the pollutant distribution vector map B as the starting point and each extreme point on the boundary line as the ending point. The vector set composed of the extreme value vectors is called the extreme value vector set.

[0149] Then, draw the extreme value vectors on the pollutant distribution vector map A, which correspond one-to-one with the angles of the extreme value vectors;

[0150] Next, the differences between the extreme value vectors in pollutant distribution vector map B and pollutant distribution vector map A are calculated respectively, resulting in a set of extreme value deviation vectors corresponding to each extreme value.

[0151] And use the calculated prediction deviation vector between the center points , and the deviation vector and extreme value deviation vector This indicates the prediction bias of the atmospheric pollutant diffusion model;

[0152] In α'≥α 阈 First, draw the extreme value vectors on the pollutant distribution vector map C, which correspond one-to-one with the angles of the extreme value vectors. Then, draw the center point of the pollutant distribution vector map C along the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. After addition, the corrected component vector is obtained. This vector is then used to compare the extreme value vectors in the pollutant distribution vector map C with the corresponding extreme value deviation vectors. The corrected extreme value vector is obtained by adding them together;

[0153] The boundaries of the pollutant distribution vector map C are adjusted according to the corrected component vectors and extreme value vectors to obtain the pollutant distribution vector map C', and the pollutant distribution vector map C' is output as the final prediction result of air pollution.

[0154] As an example of the present invention, the present invention does not limit the remote sensing images used. These images can be remote sensing images obtained by meteorological satellites or environmental monitoring satellites that regularly observe the Earth using multispectral and multi-band sensors, or remote sensing images obtained by using aircraft or drones equipped with sensors such as spectral imagers and lidar.

[0155] Preferably, the area of ​​a single remote sensing image used in this invention should be no less than 8 km². 2 The spatial resolution is not less than 1m / pixel.

[0156] Compared to existing data-driven models (LSTM), the prediction bias correction mechanism based on pollutant distribution vector maps proposed in this invention comprehensively corrects the model prediction results from multiple dimensions, including the pollution distribution center, boundaries, component vectors, and extreme value vectors, and has the following advantages:

[0157] (1) By calculating the deviation vector between the measured and predicted center point of the pollution area at the current time, and shifting the prediction results for the next period accordingly, the offset between the prediction results and the actual distribution in geographic space can be significantly reduced, thus solving the center drift problem caused by prediction errors in traditional models.

[0158] (2) By using the fractional vector and the extreme value vector to correct the boundary contour and key inflection point respectively, the contour of the pollution map is flexibly adjusted, rather than deformed on a single scale, thereby improving the geometric realism and fitting degree of the predicted map boundary.

[0159] (3) Based on the predicted and actual values ​​at the current moment, by introducing the deviation vector and sub-vector and extreme value vector of the center point, the rolling correction and dynamic optimization of the predicted value of the next period are realized. It has the advantages of simple calculation process, no risk of "error amplification", no external conditions, no need for frequent retraining, and good timeliness.

[0160] (4) Introducing the deviation coefficient α / α' mechanism to quantify the prediction error provides a clear and measurable way to assess the difference between the prediction results and the actual observations, making it easier to determine whether to make corrections and the depth and extent of the corrections;

[0161] (5) An error judgment and control mechanism with adjustable thresholds is formed by preset thresholds such as α_threshold and α_threshold', allowing users to flexibly set the tolerance error range according to accuracy requirements, ensuring a balance between resources and accuracy, avoiding blind correction, and improving system efficiency.

[0162] (5) Based on the magnitude of the deviation coefficient α / α', a multi-level and multi-granular adaptive correction strategy is realized, providing the ability to independently correct multiple pollution clouds, adapting to the complex and multi-source situation in real air pollution, and enhancing the adaptability and scalability of the model.

[0163] (6) By using a vector grid with uniformly distributed angles, the model can capture the pollution diffusion deviation in all directions more comprehensively, effectively avoiding directional omissions.

[0164] (7) This invention is a post-correction mechanism that can be independently embedded into the existing platform as a plug-in or post-processing module without changing the original pollution diffusion model structure, and has good compatibility and low modification cost.

[0165] The high-precision air pollution prediction method described in this invention is used to predict air pollution in Ningbo, Zhejiang Province. The process is as follows:

[0166] (1) First, a digital topographic map of the monitoring area was obtained based on publicly available data, and the monitoring area was divided into 187 monitoring units based on publicly available topographic maps, land cover types and vegetation distribution data.

[0167] (2) Then, based on the publicly available information, potential sources of air pollutant emissions in the monitoring area were identified and marked as fixed monitoring points, totaling 768 fixed monitoring points;

[0168] (3) Establish a dynamic assessment mechanism for gas diffusion variation factors, and determine the number of random monitoring points in each monitoring unit according to the magnitude of the gas diffusion variation factors:

[0169] When the gas diffusion variation factor is >0.8, the basic number of random monitoring points is 7;

[0170] When the gas diffusion variation factor is between 0.2 and 0.8, the basic number of random monitoring points is 4;

[0171] When the gas diffusion variation factor is <0.2, the basic number of random monitoring points is 2;

[0172] The sample space independence was verified by Moran's index, achieving a sampling error accuracy control of ≤8%, and the spatial representativeness (|I|≤0.2) was verified by Moran's index.

[0173] The number of random monitoring points is then adjusted based on the area of ​​the monitoring unit:

[0174] When the monitoring unit is greater than 100km 2 At that time, the final number of random monitoring points is: base number * 3;

[0175] When the area of ​​the monitoring unit is between 50 and 100 km² 2 At that time, the final number of random monitoring points is: base number * 1.5;

[0176] When the monitoring unit is less than 50km 2 At that time, the final number of random monitoring points is: base number * 1;

[0177] The final number of randomly selected monitoring points established was 1334.

[0178] (4) Based on the geographical information, historical meteorological information and historical pollutant information of the fixed monitoring points and random monitoring points, construct an atmospheric pollutant diffusion model (Lagrange envelope model) for the monitoring area;

[0179] (5) Based on the remote sensing images of fixed monitoring points and random monitoring points, real-time air pollutant data of fixed monitoring points and random monitoring points are obtained and input into the air pollutant diffusion model to predict the future air pollution situation and obtain the prediction results of the air pollution situation in the monitoring area in the future.

[0180] (6) Based on the current monitoring data and the historical air pollution prediction results, determine the prediction bias of the air pollutant diffusion model, and correct the air pollution prediction results of the monitoring area in the future period according to the determined prediction bias, so as to obtain the corrected air pollution prediction results.

[0181] The predicted results were compared with measured values ​​and traditional gridded, full-coverage modeling and remote sensing data fusion techniques, respectively, to obtain the following results: Figure 2 The results are shown, where the red dashed line represents the measured value of the monitoring point, the black dots represent the corrected predicted value of the monitoring point, and the green triangles represent the predicted values ​​obtained by traditional gridding, full-coverage modeling, and remote sensing data fusion techniques.

[0182] according to Figure 2 It can be seen that the high-precision air pollution prediction method of the present invention obtains remote sensing images of selected areas by sampling monitoring points, constructs a prediction model, and obtains air pollution prediction results with strong timeliness and high accuracy by correcting the predicted values.

[0183] Furthermore, compared to traditional full-coverage analysis and prediction methods, this embodiment effectively reduces the amount of data processing by conducting fixed-point and sampling monitoring of specific monitoring areas. For example, for a monitoring area of ​​10,000 square kilometers, the number of remote sensing images taken can be reduced from nearly 10,000 to about 2,000, greatly reducing the amount of data processing. It also avoids data stitching, which simplifies the data processing process, improves processing efficiency, and reduces the prediction cycle time from 3-8 hours to 30-60 minutes, significantly improving the timeliness of air pollution prediction.

[0184] Furthermore, by combining fixed and random monitoring points, this invention can ensure comprehensive monitoring while highlighting key monitoring areas and improving monitoring accuracy.

[0185] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A high-precision method for predicting air pollution, characterized in that, Including the following steps: S1, the monitoring area is divided into several monitoring units based on land use, land cover type and / or vegetation distribution data; S2, identify potential sources of air pollutant emissions in the monitoring area and mark them as fixed monitoring points; S3. Based on the accuracy requirements for air pollutant prediction, the number of random monitoring points in each monitoring unit is determined according to the area of ​​each monitoring unit and the gas diffusion variation factor. Random monitoring points and fixed monitoring points are coupled to establish a high-precision monitoring network. S4. Construct an atmospheric pollutant diffusion model for the monitoring area based on the geographical information, historical meteorological information, and historical pollutant information of the fixed and random monitoring points. S5. Based on remote sensing images of fixed and random monitoring points, real-time air pollutant data of fixed and random monitoring points are obtained and input into the air pollutant diffusion model to predict future air pollution and obtain the prediction results of air pollution in the monitoring area for a period of time in the future. S6. Based on the current monitoring data and historical air pollution prediction results, determine the prediction bias of the air pollutant diffusion model, and correct the air pollution prediction results of the monitoring area in the future period according to the determined prediction bias to obtain the corrected air pollution prediction results. In step S5, after obtaining the predicted air pollution situation in the monitoring area for a future period, the predicted air pollution situation is first analyzed and processed. The specific process includes the following steps: S51, Obtain the pollutant distribution prediction results and determine the pollutant distribution boundary and center; The process of determining the pollutant distribution boundary and center based on the pollutant distribution prediction results is as follows: First, compare the pollutant concentration of each region with the preset concentration threshold. If the pollutant concentration exceeds the concentration threshold, it is a polluted area; otherwise, it is a non-polluted area. After determining the polluted area, calculate the centroid or geometric center of the polluted area and use it as the pollutant distribution center point of the region. S52, Starting from the pollutant distribution center and ending at the pollutant distribution boundary line, draw a pollutant distribution vector map C corresponding to the future air pollution forecast results; Step S6 specifically includes: S61, Obtain the atmospheric pollution situation prediction results corresponding to the current moment from the output of the atmospheric pollutant diffusion model; S62, Obtain the measured results of air pollution at the current moment based on remote sensing image monitoring; S63, analyze the predicted air pollution situation and the measured air pollution situation at the current time respectively, determine the distribution boundary and center of pollutants, and then draw the pollutant distribution vector map A corresponding to the current air pollution situation and the pollutant distribution vector map B corresponding to the current air pollution situation, with the pollutant distribution center as the starting point and the pollutant distribution boundary line as the ending point. S64, Determine the prediction bias of the atmospheric pollutant diffusion model based on the results of pollutant distribution vector maps A and B; S65, Based on the determined prediction deviation, the prediction results of air pollution in the monitoring area for a future period of time are corrected to obtain the corrected prediction results of air pollution. In steps S52 and S63, the drawn pollutant distribution vector map includes multiple one-to-one corresponding sub-vectors with the same angle. The sub-vectors are drawn at predetermined angular intervals, with the pollutant distribution center as the starting point and the pollutant distribution boundary line as the ending point.

2. The high-precision air pollution prediction method according to claim 1, characterized in that, Step S2 specifically includes: S21, Obtain or construct geographic information of the monitoring area; S22, Identify and classify potential sources of air pollution; S23 identifies potential sources of air pollutant emissions as candidate sites for fixed monitoring points; S24, merge and screen the candidate fixed monitoring points to determine the final fixed monitoring points; S25, Determine the specific location of the fixed monitoring point.

3. The high-precision air pollution prediction method according to claim 1, characterized in that, In step S3, the process of determining the number of random monitoring points in each monitoring unit is as follows: (1) Establish a dynamic assessment mechanism for gas diffusion variation factors, wherein the gas diffusion variation factor of each monitoring unit = standard deviation of historical pollutant content / average value of historical pollutant content; (2) Determine the basic number of random monitoring points for each monitoring unit based on the magnitude of the gas diffusion variation factor; (3) Adjust the basic number of random monitoring points based on the area of ​​each monitoring unit.

4. The high-precision air pollution prediction method according to claim 1, characterized in that, In step S64, the process of determining the prediction bias of the atmospheric pollutant diffusion model based on the pollutant distribution vector map results A and B is as follows: S641, calculate the deviation coefficient α of the pollutant distribution areas in pollutant distribution vector maps A and B, where the deviation coefficient α = (1-K) 100%, where K is the overlap ratio, and the overlap ratio K = the overlap area of ​​vector diagrams A and B / the total area of ​​pollutant distribution areas in diagram B; S642, compare the deviation coefficient α and the preset threshold α 阈 Relative size: If α < α 阈 Instead of correcting the air pollution prediction results for the next period output by the air pollutant diffusion model, step S65 is executed to directly output the air pollution prediction results of the air pollutant diffusion model. If α≥α 阈 The atmospheric pollution forecast results for the next period output by the atmospheric pollutant diffusion model need to be corrected, and step S643 should be continued. S643, using the center point of the pollutant distribution area in pollutant distribution vector map A as the starting point and the center point of the pollutant distribution area in pollutant distribution vector map B as the ending point, draw the prediction deviation vector between the center points of the pollutant distribution areas in pollutant distribution vector maps A and B. ; Then along the vector The pollutant distribution vector map A is translated from the starting point to the ending point to obtain the translated pollutant distribution vector map A'; Calculate the differences between each component vector in the pollutant distribution vector map B and the pollutant distribution vector map A' respectively, to obtain a set of component vector deviation vectors corresponding to each angle. , , , ...; And use the calculated prediction deviation vector between the center points , and the deviation vector , , , ..., represent the prediction bias of the atmospheric pollutant diffusion model.

5. The high-precision air pollution prediction method according to claim 4, characterized in that, Step S65 includes: When α < α 阈 At that time, the air pollution prediction results of the air pollutant diffusion model are directly output; When α≥α 阈 First, the center point of the pollutant distribution vector diagram C is aligned with the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. , , The corrected component vectors are obtained by adding the components together. The boundary of the pollutant distribution vector map C is then adjusted according to the corrected component vectors to obtain the pollutant distribution vector map C'. The pollutant distribution vector map C' is then output as the final prediction result of the air pollution situation.

6. The high-precision air pollution prediction method according to claim 5, characterized in that, Step S65 further includes: when α≥α 阈 Then, determine whether a deep correction is needed for the prediction results: First, map the pollutant distribution vector A along the prediction deviation vector. The pollutant distribution vector map A' is translated so that its center point coincides with the center point of the pollutant distribution vector map B. Then, the deviation coefficient α' between the pollutant distribution areas in vector maps A' and B is calculated, where the deviation coefficient α' = (1 - K'). 100%, where K' is the overlap ratio, and the overlap ratio K' = the overlap area of ​​vector diagrams A' and B / the total area of ​​pollutant distribution areas in diagram B; Then compare the deviation coefficient α' with the preset threshold α. 阈 Relative size of ': If α' < α 阈 ', No depth correction is required for the pollutant distribution vector map C; If α'≥α 阈 ', Depth correction is needed for the pollutant distribution vector map C.

7. The high-precision air pollution prediction method according to claim 6, characterized in that, The depth correction process is as follows: First, the boundary line of the pollutant distribution vector map B is analyzed to obtain the extreme points on the boundary line. Then, the extreme value vectors of the pollutant distribution vector map B are drawn with the center point of the pollutant distribution vector map B as the starting point and each extreme point on the boundary line as the ending point. The vector set composed of the extreme value vectors is called the extreme value vector set. Then, draw the extreme value vectors on the pollutant distribution vector map A, which correspond one-to-one with the angles of the extreme value vectors; Next, the differences between the extreme value vectors in pollutant distribution vector map B and pollutant distribution vector map A are calculated respectively, resulting in a set of extreme value deviation vectors corresponding to each extreme value. , , , ...; In α'≥α 阈 First, draw the extreme value vectors on the pollutant distribution vector map C, which correspond one-to-one with the angles of the extreme value vectors. Then, draw the center point of the pollutant distribution vector map C along the vector... After translation, the component vectors in the pollutant distribution vector map C are compared with the corresponding component vector deviation vectors. , , ..., after adding them together, we obtain the corrected component vector. This vector is then used to compare the extreme value vectors in the pollutant distribution vector map C with the corresponding extreme value deviation vectors. , , The corrected extreme value vector is obtained by adding the following: ... The boundaries of the pollutant distribution vector map C are adjusted according to the corrected component vectors and extreme value vectors to obtain the pollutant distribution vector map C', and the pollutant distribution vector map C' is output as the final prediction result of air pollution.

Images