Regional pollutant concentration prediction method and device and storage medium

Abstract

The invention relates to a regional pollutant concentration prediction method and device and a storage medium, and is applied to the technical field of pollution monitoring. The method specifically comprises the following steps: acquiring historical pollutant concentration monitoring data and meteorological data of each single site, and predicting future fine particulate matter concentration and pollutant chemical components of the single site based on an EnvMulti-KAN model fused by KAN and Transform architectures; according to the prediction result of each single station in the target area, particle diffusion and convergence are simulated by adopting a multi-station fine particle diffusion model in combination with geographic information and interaction among the single stations, the pollutant area concentration is constructed, the area fine particle concentration is accurately predicted, and the view limitation of the single station is broken; according to the method, the single-site model and the multi-site model are cooperated, the concentration of the regional fine particulate matter is accurately predicted, the main components of the pollutants are synchronously and accurately predicted, and a demonstration is provided for atmospheric chemistry theoretical research.

Description

Technical Field

[0001] The present invention belongs to the technical field of pollution monitoring, and in particular relates to a method, a device and a storage medium for predicting regional pollutant concentration. Background Art

[0002] At present, fine particulate matter from atmospheric pollution (such as PM2.5) has caused serious harm to the environment and human health and has attracted widespread attention. In response to atmospheric fine particulate matter pollution, relevant monitoring technologies have made certain progress. For example, high-precision particulate matter mass concentration monitoring instruments continue to emerge, and particulate matter size distribution measurement equipment based on principles such as laser scattering is also becoming increasingly popular. However, the component analysis of fine particulate matter from atmospheric pollution faces many challenges. Existing analysis methods often require complex sampling processes and laboratory analysis. For example, after using filter membrane sampling, the organic matter in the particulate matter is analyzed using gas chromatography-mass spectrometry (GC-MS) and other technologies. Although this method is accurate, it has a time lag and cannot achieve instant analysis.

[0003] Existing technologies usually focus on using a single sensor technology to infer the composition of fine particulate matter in atmospheric pollution. For example, optical sensors are used to detect the absorption and scattering characteristics of particulate matter to light of a specific wavelength. A simple discrimination model is established based on the known response patterns of different substances to light. Taking the detection of heavy metal elements as an example, when light is irradiated on particulate matter containing heavy metals, an absorption peak of a specific wavelength will be generated. After the sensor captures the signal, it determines whether the particulate matter contains heavy metals and the approximate content based on preset thresholds and characteristic spectra. However, this method can only identify a limited number of substances with obvious characteristics. For complex organic mixtures and the coexistence of multiple trace elements, it is difficult to accurately distinguish the components due to mutual interference of optical signals, and it is even more impossible to fully reconstruct the overall composition of fine particulate matter. In addition, it is easily affected by factors such as ambient light noise and particle agglomeration, and has poor stability. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a regional pollutant concentration prediction method, device and storage medium to solve the problem that in the prior art, a single sensor technology is used to infer the composition of fine particulate matter in atmospheric pollution, and only a limited number of substances with obvious characteristics can be identified. Moreover, due to the mutual interference of optical signals, it is difficult to accurately distinguish the components, and it is even more impossible to fully reconstruct the overall picture of the composition of fine particulate matter. Moreover, it is easily affected by factors such as ambient light noise and particle agglomeration, and has poor stability.

[0005] According to a first aspect of an embodiment of the present invention, a method for predicting regional pollutant concentration is provided, the method comprising:

[0006] Obtain the pollutant concentration monitoring data and meteorological data of each single site in the target area during the historical preset time period. The pollutant concentration monitoring data includes the concentrations of various pollutants and the chemical composition of each pollutant;

[0007] Take the pollutant concentration monitoring data and meteorological data of each single site in the historical preset time period as input data and input them into the pre-trained EnvMulti-KAN model respectively. Through the EnvMulti-KAN model, predict the concentration of fine particulate matter and the composition of fine particulate matter at each single site in the future for a period of time;

[0008] Obtain the geographical feature data and real-time meteorological data of each single site in the target area;

[0009] Input the concentration of fine particulate matter, the composition of fine particulate matter, geographical feature data, and real-time meteorological data at all single sites in the future for a period of time into the pre-trained multi-site fine particulate matter diffusion model;

[0010] Output the predicted value of the pollutant concentration in the target area, as well as the contribution ratios of meteorological factors, geographical factors, and each single site to the concentrations of various pollutants in the target area through the multi-site fine particulate matter diffusion model.

[0011] Preferably,

[0012] The EnvMulti-KAN model includes: a multi-scale feature fusion module, a dynamic environment interaction module, and a multi-task prediction decoder;

[0013] The prediction of the concentration of fine particulate matter and the composition of fine particulate matter at each single site in the future for a period of time through the EnvMulti-KAN model includes:

[0014] The multi-scale feature fusion module uses a fine-grained spline function to obtain the non-linear relationship between meteorological factors and pollutant concentrations in the input data. The multi-scale feature fusion module also uses a coarse-grained spline function to obtain the long-term trend of meteorological factors and pollutant concentrations;

[0015] The multi-scale feature fusion module is also used to fuse the non-linear relationship and long-term trend of meteorological factors and pollutant concentrations to obtain multi-scale spatio-temporal encoded features;

[0016] The dynamic environment interaction module constructs a component correlation graph of each pollutant component according to the multi-scale spatio-temporal encoded features. The dynamic environment interaction module also generates a pollution component modulation coefficient according to the multi-scale spatio-temporal encoded features;

[0017] The multi-task prediction decoder outputs the concentration of fine particulate matter and the composition of fine particulate matter at a single site within a future period of time according to the component correlation diagram of each pollutant component and the pollution component modulation coefficient.

[0018] Preferably,

[0019] The multi-site fine particulate matter diffusion model includes: a graph construction module, a spatio-temporal cross memory module, a physical constraint module, and a multi-task decoding module;

[0020] The output of the pollutant concentration prediction value of the target area by the multi-site fine particulate matter diffusion model, as well as the meteorological factors, geographical factors, and the contribution ratio of each single site to the concentration of each pollutant in the target area respectively, includes:

[0021] The graph construction module constructs a regional feature graph composed of multiple sites according to the dimensional features in the input data, and the dimensional features in the input data include the concentration of fine particulate matter, the composition of fine particulate matter, geographical features, and real-time meteorology;

[0022] The regional feature graph composed of multiple sites and the pollutant concentration monitoring data of all single sites in the historical preset period are input into the spatio-temporal cross memory module together. The spatio-temporal cross memory module corrects the regional feature graph composed of multiple sites through the pollutant concentration monitoring data in the historical preset period, and at the same time obtains the spatio-temporal dependence relationship between the pollutant concentration monitoring data and the regional feature graph composed of multiple sites through the spatio-temporal memory unit, and obtains the node features of the corrected regional feature graph;

[0023] The node features of the corrected regional feature graph are input into the physical constraint module. A variety of physical laws are preset in the physical constraint module to perform physical law constraints on the input node features of the corrected regional feature graph, and output node features that meet the physical laws;

[0024] The node features that meet the physical laws are input into the multi-task decoding, and the pollutant concentration prediction value of the target area is output, as well as the meteorological factors, geographical factors, and the contribution ratio of each single site to the concentration of each pollutant in the target area are output.

[0025] Preferably, it further includes:

[0026] Before the pollutant concentration monitoring data and meteorological data of each single site in the historical preset period are respectively input into the pre-trained EnvMulti-KAN model as input data, interference factor identification is performed, and abnormal data in the pollutant concentration monitoring data and meteorological data are removed according to the interference factor identification result.

[0027] Preferably,

[0028] The interference factor identification includes: abnormal meteorological disaster identification and temporary large pollution source intrusion identification;

[0029] The abnormal meteorological disaster identification includes: according to a plurality of preset abnormal meteorological disaster determination conditions, judging whether any abnormal meteorological disaster occurs based on the meteorological data in the historical preset time period;

[0030] The temporary large pollution source intrusion identification includes: presetting temporary large pollution source intrusion determination conditions, and judging whether a temporary large pollution source intrusion occurs based on the pollutant concentration monitoring data in the historical preset time period;

[0031] Removing abnormal data in the pollutant concentration monitoring data and meteorological data according to the interference factor identification result includes:

[0032] If it is identified that an abnormal meteorological disaster occurs, the pollutant concentration monitoring data and meteorological data in the historical preset time period and the pollutant concentration monitoring data and meteorological data in the time period when the abnormal meteorological disaster occurs are deleted;

[0033] Obtain the historical same time period when the abnormal meteorological disaster occurs and the pollutant concentration monitoring data and meteorological data without abnormal meteorological disasters, and fill in the deleted pollutant concentration monitoring data and meteorological data;

[0034] If it is identified that a temporary large pollution source intrusion occurs, the pollutant concentration monitoring data in the historical preset time period and in the time period when the temporary large pollution source intrusion occurs are deleted;

[0035] Obtain the historical same time period when the temporary large pollution source intrusion occurs and the pollutant concentration monitoring data without temporary large pollution source intrusion, and fill in the deleted pollutant concentration monitoring data.

[0036] Preferably, it further includes:

[0037] Preprocess the outputs of the EnvMulti-KAN model and the multi-site fine particulate matter diffusion model respectively, so that the outputs of the EnvMulti-KAN model and the multi-site fine particulate matter diffusion model are consistent in time step and spatial resolution;

[0038] Fuse the outputs of the EnvMulti-KAN model and the multi-site fine particulate matter diffusion model with consistent spatio-temporal dimensions to construct spatio-temporal four-dimensional data of time, space, pollutant concentration and chemical composition of pollutants;

[0039] Based on the four-dimensional spatio-temporal data, a regional fine particulate matter component distribution map is generated using a GIS platform, and the regional fine particulate matter component distribution map is used to dynamically present the concentration changes and spatial distributions of pollutant components at each site;

[0040] A concentration heat map is generated based on the regional fine particulate matter component distribution map, and the diffusion path of pollutants is obtained according to the concentration heat map;

[0041] Based on the concentration heat map, the transmission path of pollutants is obtained by combining the wind rose diagram in the meteorological data and the geographical feature map;

[0042] The output of the EnvMulti-KAN model and the output of the multi-site fine particulate matter diffusion model are factorized by a positive definite matrix to obtain the contribution ratios of different pollution source types, and the pollution source with the greatest influence is obtained based on the contribution ratios of different pollution source types.

[0043] According to the second aspect of the embodiments of the present invention, a regional pollutant concentration prediction device is provided, and the device includes:

[0044] Historical data acquisition module: used to acquire the pollutant concentration monitoring data and meteorological data of each single site in the target area within a historical preset time period, and the pollutant concentration monitoring data includes the concentrations of various pollutants and the chemical composition components of various pollutants;

[0045] Single-site prediction module: used to input the pollutant concentration monitoring data and meteorological data of each single site within a historical preset time period as input data into the pre-trained EnvMulti-KAN model respectively, and predict the fine particulate matter concentration and fine particulate matter composition components of each single site within a future period of time through the EnvMulti-KAN model;

[0046] Real-time data acquisition module: used to acquire the geographical feature data and real-time meteorological data of each single site in the target area;

[0047] Multi-site input module: used to input the fine particulate matter concentration and fine particulate matter composition components, geographical feature data and real-time meteorological data of all single sites within a future period of time into the pre-trained multi-site fine particulate matter diffusion model together;

[0048] Regional prediction module: used to output the pollutant concentration prediction value of the target area through the multi-site fine particulate matter diffusion model, as well as the contribution ratios of meteorological factors, geographical factors and each single site to the concentrations of various pollutants in the target area.

[0049] According to a third aspect of an embodiment of the present invention, a storage medium is provided. The storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented.

[0050] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0051] In this application, historical pollutant concentration monitoring data and meteorological data of each single site are obtained, and the EnvMulti-KAN model based on the fusion of KAN and Transformer architectures is used to predict the future fine particulate matter concentration and pollutant chemical composition components of a single site; further, according to the prediction results of each single site in the target area, a multi-site fine particulate matter diffusion model is combined with geographical information and the interaction between each single site to simulate the diffusion and convergence of particles, construct the regional concentration of pollutants, and accurately predict the regional fine particulate matter concentration, breaking the limitation of the single-site perspective; in this application, by making the single-site model and the multi-site model cooperate, the regional fine particulate matter concentration is accurately predicted, and the main components of pollutants are accurately predicted synchronously, providing empirical evidence for the research of atmospheric chemistry theory.

[0052] It should be understood that the above general description and subsequent detailed description are only exemplary and explanatory, and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0054] Figure 1 is a flowchart showing a method for predicting regional pollutant concentration according to an exemplary embodiment;

[0055] Figure 2 is a system diagram showing a device for predicting regional pollutant concentration according to another exemplary embodiment;

[0056] In the drawings: 1 - historical data acquisition module, 2 - single-site prediction module, 3 - real-time data acquisition module, 4 - multi-site input module, 5 - regional prediction module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are only examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0058] Embodiment 1

[0059] Figure 1 is a schematic flowchart of a method for predicting regional pollutant concentration shown according to an exemplary embodiment, as Figure 1 shown, the method includes:

[0060] S1. Obtain the pollutant concentration monitoring data and meteorological data of each single site in the target area within a historical preset time period, where the pollutant concentration monitoring data includes the concentrations of various pollutants and the chemical composition components of each pollutant;

[0061] S2. Input the pollutant concentration monitoring data and meteorological data of each single site in the historical preset time period as input data into the pre-trained EnvMulti-KAN model respectively, and predict the concentration of fine particulate matter and the composition components of fine particulate matter within a future period of time for the single site through the EnvMulti-KAN model;

[0062] S3. Obtain the geographical feature data and real-time meteorological data of each single site in the target area;

[0063] S4. Input the concentration of fine particulate matter and the composition components of fine particulate matter, geographical feature data, and real-time meteorological data within a future period of time of all single sites into the pre-trained multi-site fine particulate matter diffusion model together;

[0064] S5. Output the predicted value of the pollutant concentration in the target area, as well as the contribution ratios of meteorological factors, geographical factors, and each single site to the concentrations of various pollutants in the target area through the multi-site fine particulate matter diffusion model;

[0065] It is understandable that in this application, high-resolution data of each single site in the target area over the past three years is obtained, including pollutant concentration monitoring data, chemical composition of pollutants, and meteorological data, etc. The obtained historical data is input into the EnvMulti-KAN model. The EnvMulti-KAN model is based on the Kolmogorov-Arnold Networks (KAN) and Transformer architectures, and includes: a multi-scale feature fusion module, a dynamic environment interaction module, and a multi-task prediction decoder. Among them, the multi-scale feature fusion module uses a fine-grained spline function to capture the non-linear relationship between meteorological factors and pollutant concentrations in the input data, and uses a coarse-grained spline function to obtain the long-term trends of meteorological factors and pollutant concentrations, and fuses the two to obtain multi-scale spatio-temporal encoded features (hour / day / week-level embeddings). The multi-scale spatio-temporal encoded features are input into the dynamic environment interaction module, which is used to input meteorological variables (temperature, humidity, wind speed) and pollutant concentrations into the dynamic KAN branch to generate non-linear modulation coefficients of meteorological conditions on pollution components, and at the same time construct a correlation graph of each pollution component; the non-linear modulation coefficients and the correlation graph of each pollution component are input into the multi-task prediction decoder to output the concentration of fine particulate matter and the composition of fine particulate matter within a future period of time at a single site;

[0066] The target area is usually composed of the areas of each single site. Through the above EnvMulti-KAN model, the prediction results of each single site have been obtained. Next, the geographical data and real-time meteorological data of each single site are obtained, and the prediction results of each single site described above are combined and input into the multi-site fine particulate matter diffusion model. Among them, the multi-site fine particulate matter diffusion model includes: a graph construction module, a spatio-temporal cross memory module, a physical constraint module, and a multi-task decoding module; the graph construction module constructs a regional feature graph composed of multiple sites according to the features of each dimension in the input data, and the features of each dimension include the concentration of fine particulate matter, the composition of fine particulate matter, geographical features, and real-time meteorology; the input of the spatio-temporal cross memory module includes two parts, one is the regional feature graph output by the graph construction module, and the other is the pollutant concentration monitoring data of the single site in the historical preset period. The spatio-temporal cross memory module corrects the regional feature graph composed of multiple sites through the pollutant concentration monitoring data in the historical preset period, and at the same time obtains the spatio-temporal dependence relationship between the two through the spatio-temporal memory unit to obtain the node features of the corrected regional feature graph; the node features of the corrected regional feature graph are input into the physical constraint module, and there are various physical laws preset in the physical constraint module, such as: Newton's three laws, the law of conservation of momentum, etc., and the node features that meet the physical laws are output; the node features that meet the physical laws are input into the multi-task decoding to output the predicted value of the pollutant concentration in the target area, and the contribution ratio of each factor (output meteorological factor, geographical factor, and each single site) to the pollutant concentration in the target area;

[0067] To further improve the prediction accuracy, this embodiment also discloses an external factor removal method, namely abnormal meteorological disaster identification and temporary large pollution source intrusion identification, including:

[0068] Define the types and determination conditions of abnormal meteorological disasters:

[0069] (1) Sandstorm: wind speed continuously ≥ 15 m / s, visibility ≤ 1 km, PM10 concentration (instantaneous ≥ 500 μg / m 3 );

[0070] (2) Extreme precipitation: 24-hour precipitation ≥ 100 mm (refer to the red rainstorm warning standard of the meteorological bureau);

[0071] (3) Persistent high temperature and inversion: the height of the inversion layer ≤ 200 meters for 3 consecutive days and the daily maximum temperature ≥ 35°C;

[0072] (4) Stable weather: wind speed ≤ 1 m / s for more than 48 hours, boundary layer height ≤ 500 meters;

[0073] Time window verification: the abnormal state needs to last for at least 2 hours (to avoid misjudgment due to instantaneous interference);

[0074] According to historical meteorological data, determine whether any period meets any of the above abnormal meteorological disasters. Since abnormal meteorological disasters will affect pollutant concentration, after determining that any abnormal meteorological disaster has occurred, it is necessary to delete the pollutant monitoring data and meteorological data of the abnormal meteorological disaster occurrence period in the historical input data at the same time, and fill it with the pollutant monitoring data and meteorological data of the historical same period when the abnormal meteorological disaster occurred and no abnormal meteorological disaster occurred. For example: if extreme precipitation occurred from September 15th to September 16th in the historical data of 2016, then delete the meteorological data and pollutant monitoring data from September 15th to September 16th in the input data, and then obtain the meteorological data and pollutant monitoring data from September 15th to September 16th in 2015. If no abnormal meteorological disaster occurred from September 15th to September 16th in 2015, then use the pollutant monitoring data and meteorological data from September 15th to September 16th in 2015 to fill the pollutant monitoring data and meteorological data from September 15th to September 16th in the deleted historical data of 2016. If abnormal meteorological disasters also occurred from September 15th to September 16th in 2015, then continue to obtain the pollutant monitoring data and meteorological data from September 15th to September 16th in 2014;

[0075] Define the intrusion of temporary large pollution sources and determination conditions:

[0076] (1) Industrial accident: The concentration of SO or NOx increases by more than 300% within 1 hour, and there is a factory coordinate match in the adjacent area;

[0077] (2) Traffic emergency: The CO concentration increases sharply (hourly average ≥ 5 ppm) accompanied by abnormal traffic monitoring data;

[0078] (3) Open burning: The proportion of potassium ions (K+) in PM2.5 suddenly increases (> 10%) and matches the satellite fire point monitoring;

[0079] Through spatio-temporal correlation analysis, confirm the spatio-temporal causal relationship between the pollution source and the monitoring data. First, through time matching, when the monitoring station detects an abnormal increase in pollutant concentration, trace back the time axis to find possible pollution events. Through spatial matching, combine the coordinates of the pollution source and the location of the monitoring station to calculate the pollution diffusion path. For example, if a chemical plant leaks and the stations within 5 kilometers downwind detect an increase in the VOCs concentration 30 minutes later, it is determined as an associated event. Through diffusion model verification, use the Gaussian diffusion model or the Lagrangian particle model to simulate the pollutant transmission path and verify whether the concentration peak conforms to the expected distribution. In terms of technical implementation, set a dynamic time window (such as ±1 hour), match the pollution event with the concentration change, and delimit the potential impact area of the pollution source based on meteorological data (wind direction, wind speed), and only associate the monitoring stations within this area; The data sources include emission inventories, traffic monitoring data, satellite remote sensing fire point data, and meteorological data. The meteorological data verifies the pollutant diffusion conditions. For example, under stable weather conditions, local emissions accumulate. If the PM2.5 at a certain station increases but there is no external transmission, it is attributed to the local source. In terms of technical implementation, use Kriging interpolation to align the satellite fire points, traffic data with the location of the monitoring stations, and mine the frequent item sets of multi-source data based on the Apriori algorithm. For example, if "traffic congestion + sudden increase in CO" occur simultaneously, it is considered that a traffic emergency, a temporary large pollution source, has intruded; Since the suddenly increased CO concentration at this time has a strong interference on the prediction results, it is necessary to delete the CO concentration monitoring data during the period when the temporary large pollution source intrudes in the model input data, and then use the CO concentration monitoring data in the same historical period for substitution. The substitution method is the same as that for abnormal meteorological disaster data.

[0080] This embodiment also provides a post-processing and feedback mechanism, that is, record the time, type, affected stations, and treatment measures in the interference event library for model iteration optimization, and support environmental management personnel to manually confirm or correct the automatic determination results;

[0081] This embodiment also provides further analysis applications based on the single-site output results and multi-site output results, including:

[0082] Unify the time step (e.g., hourly prediction) and spatial resolution (e.g., 1 km grid) of the single-site model and multi-site model outputs through data alignment technology, ensure the consistency of meteorological conditions (e.g., wind direction, terrain obstruction) and pollutant diffusion paths in the time and space dimensions, avoid prediction biases caused by data scale differences, and then deeply fuse the output results of the single-site and multi-site to construct a four-dimensional spatio-temporal data including time, space, pollutant concentration, and more than 20 component characteristics; based on the integrated four-dimensional spatio-temporal data, use the GIS platform to generate a regional fine particulate matter component distribution map, dynamically presenting the concentration changes and spatial distributions of pollutant components (e.g., nitrates, sulfates, organic matter, etc.) at each site;

[0083] Generate a heat map from the regional fine particulate matter component distribution map, mark the high-concentration aggregation areas through the heat map, and simulate the pollutant diffusion path in combination with the streamline map;

[0084] Based on the heat map, combine the wind rose diagram in the meteorological data and the geographical data map to analyze the dominant wind field, use the terrain elevation data to identify diffusion barriers (such as mountains, building clusters), and reverse-trace the pollutant transmission path. For example, if the nitrate concentration peak in a certain area is strongly correlated with the northwest wind direction and there is an industrial park upwind, it can be determined that this industrial area is the main contributor; based on the reconstructed component data, use positive matrix factorization (PMF) to obtain the contribution ratios of different pollution source types, and obtain the most influential pollution source based on the contribution ratios of different pollution source types;

[0085] In this embodiment, after obtaining the prediction results of the single-point and multi-point, the fine particulate matter component data predicted by the single-site and multi-site are integrated. Based on the concentrations of the main components of the fine particulate matter predicted by the single-site and the regional fine particulate matter concentration distribution predicted by the multi-site, combined with the geographical data set and meteorological condition data, a spatial distribution of the fine particulate matter components is established. Using advanced data analysis algorithms, analyze the concentration differences of the predicted fine particulate matter components between different sites and regions, consider the impacts of factors such as atmospheric flow and topography on the transmission and distribution of different components, and reverse-infer the component evolution law of the fine particulate matter during the transmission process, providing all-round, high-timeliness, and high-precision data support for air pollution control from source prevention and control, process monitoring to effect evaluation.

[0086] Embodiment 2:

[0087] Figure 2 It is a system schematic diagram of a regional pollutant concentration prediction device shown according to another exemplary embodiment. The device includes:

[0088] Historical data acquisition module 1: It is used to acquire the pollutant concentration monitoring data and meteorological data of each single site within the target area during the historical preset time period. The pollutant concentration monitoring data includes the concentrations of various pollutants and the chemical composition components of each pollutant;

[0089] Single-site prediction module 2: It is used to input the pollutant concentration monitoring data and meteorological data of each single site during the historical preset time period as input data into the pre-trained EnvMulti-KAN model respectively, and predict the concentration of fine particulate matter and the composition components of fine particulate matter within a future period of time for each single site through the EnvMulti-KAN model;

[0090] Real-time data acquisition module 3: It is used to acquire the geographical feature data and real-time meteorological data of each single site within the target area;

[0091] Multi-site input module 4: It is used to input the concentration of fine particulate matter and the composition components of fine particulate matter, geographical feature data, and real-time meteorological data of all single sites within a future period of time into the pre-trained multi-site fine particulate matter diffusion model together;

[0092] Regional prediction module 5: It is used to output the predicted values of pollutant concentrations in the target area through the multi-site fine particulate matter diffusion model, as well as the contribution ratios of meteorological factors, geographical factors, and each single site to the concentrations of various pollutants in the target area.

[0093] Embodiment 3:

[0094] This embodiment provides a storage medium, which stores a computer program. When the computer program is executed by the main controller, it realizes each step in the above method;

[0095] It can be understood that the above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disc, etc.

[0096] It can be understood that the same or similar parts in the above embodiments can be referred to each other. For the content not detailed in some embodiments, reference can be made to the same or similar content in other embodiments.

[0097] It should be noted that in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "sparsely distributed in small amounts" means at least two.

[0098] Any process or method description depicted in the flowchart or described otherwise herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing specific logical functions or steps of the process. The scope of the preferred embodiments of the present invention includes additional implementations where functions may be executed not in the order shown or discussed, including in a substantially simultaneous manner or in the reverse order according to the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0099] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, a small number of sparsely distributed steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0100] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0101] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0102] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, etc.

[0103] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or a small number of sparsely distributed embodiments or examples.

[0104] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting regional pollutant concentration, characterized in that: The method comprises: Acquire historical pollutant concentration monitoring data and meteorological data for each single site in the target area for a preset time period, wherein the pollutant concentration monitoring data includes the concentration of each pollutant and the chemical composition of each pollutant; The pollutant concentration monitoring data and meteorological data of each single station in the preset historical time period are input into the pre-trained EnvMulti-KAN model as input data, and the fine particle concentration and fine particle composition of the single station in the future period are predicted by the EnvMulti-KAN model; Obtain geographic feature data and real-time meteorological data for each single site in the target area; The fine particle concentrations and fine particle composition, geographic feature data, and real-time meteorological data of all single sites in the future are input into the pre-trained multi-site fine particle diffusion model; The multi-site fine particulate matter diffusion model outputs the predicted value of pollutant concentration in the target area, as well as the contribution of meteorological factors, geographical factors and each single site to the concentration of each pollutant in the target area.

2. The method according to claim 1, characterized in that The EnvMulti-KAN model includes: a multi-scale feature fusion module, a dynamic environment interaction module and a multi-task prediction decoder; The prediction of the fine particulate matter concentration and fine particulate matter composition of a single site in the future period of time by the EnvMulti-KAN model includes: The multi-scale feature fusion module uses a fine-grained spline function to obtain the nonlinear relationship between meteorological factors and pollutant concentrations in the input data, and the multi-scale feature fusion module also uses a coarse-grained spline function to obtain the long-term trend of meteorological factors and pollutant concentrations; The multi-scale feature fusion module is also used to fuse the nonlinear relationship between the meteorological factors and the pollutant concentration and the long-term trend to obtain multi-scale spatiotemporal coding features; The dynamic environment interaction module constructs a component association diagram of each pollutant component according to the multi-scale spatiotemporal coding feature, and the dynamic environment interaction module also generates a pollution component modulation coefficient according to the multi-scale spatiotemporal coding feature; The multi-task prediction decoder outputs the fine particulate matter concentration and fine particulate matter composition of a single site within a future period of time according to the component association diagram of each pollutant component and the pollution component modulation coefficient.

3. The method according to claim 2, characterized in that The multi-site fine particle diffusion model includes: a graph construction module, a spatiotemporal cross memory module, a physical constraint module and a multi-task decoding module; The pollutant concentration prediction value of the target area output by the multi-site fine particulate matter diffusion model, as well as the contribution ratio of meteorological factors, geographical factors and each single site to the concentration of each pollutant in the target area include: The graph construction module constructs a regional feature graph composed of multiple sites according to the dimensional features in the input data, wherein the dimensional features in the input data include fine particulate matter concentration, fine particulate matter composition, geographical features, and real-time meteorology; The regional characteristic map composed of multiple sites and the pollutant concentration monitoring data of all single sites in the historical preset time period are input into the spatiotemporal cross memory module. The spatiotemporal cross memory module corrects the regional characteristic map composed of multiple sites through the pollutant concentration monitoring data of the historical preset time period, and at the same time obtains the spatiotemporal dependency relationship between the pollutant concentration monitoring data of the historical preset time period and the regional characteristic map composed of multiple sites through the spatiotemporal memory unit to obtain the node features of the corrected regional characteristic map; Inputting the modified regional feature map node features into a physical constraint module, wherein a plurality of physical laws are preset in the physical constraint module, performing physical law constraints on the input modified regional feature map node features, and outputting node features satisfying the physical laws; The node features that satisfy the physical laws are input into the multi-task decoding, and the predicted value of the pollutant concentration in the target area is output, as well as the contribution ratio of meteorological factors, geographical factors and each single site to the concentration of each pollutant in the target area.

4. The method according to claim 3, characterized in that: Also includes: The pollutant concentration monitoring data and meteorological data of each single site in the preset historical time period are input into the pre-trained EnvMulti-KAN model as input data to identify interference factors, and the abnormal data in the pollutant concentration monitoring data and meteorological data are removed according to the interference factor identification results.

5. The method according to claim 2, characterized in that: The interference factor identification includes: abnormal meteorological disaster identification and temporary large-scale pollution source intrusion identification; The abnormal meteorological disaster identification includes: judging whether any abnormal meteorological disaster occurs according to the meteorological data of the historical preset time period according to the preset multiple abnormal meteorological disaster judgment conditions; The temporary large-scale pollution source intrusion identification includes: presetting the temporary large-scale pollution source intrusion determination conditions, and judging whether the temporary large-scale pollution source intrusion occurs according to the pollutant concentration monitoring data of the historical preset time period; The removing of abnormal data in the pollutant concentration monitoring data and meteorological data according to the interference factor identification result comprises: If an abnormal meteorological disaster is identified, the pollutant concentration monitoring data for the historical preset time period and the pollutant concentration monitoring data for the abnormal meteorological disaster period and the meteorological data are deleted; Obtain pollutant concentration monitoring data and meteorological data for the same period in history when abnormal meteorological disasters occurred and when no abnormal meteorological disasters occurred, and fill in the deleted pollutant concentration monitoring data and meteorological data; If a temporary large-scale pollution source intrusion is identified, the pollutant concentration monitoring data of the historical preset time period during the period of the temporary large-scale pollution source intrusion is deleted; Obtain the pollutant concentration monitoring data of the historical same period when temporary large-scale pollution sources intrusion occurred and when no temporary large-scale pollution sources intrusion occurred, and fill in the deleted pollutant concentration monitoring data.

6. The method according to claim 5, characterized in that Also includes: Preprocessing the output of the EnvMulti-KAN model and the output of the multi-site fine particle diffusion model respectively, so that the time step and spatial resolution of the output of the EnvMulti-KAN model and the output of the multi-site fine particle diffusion model are consistent; The output of the EnvMulti-KAN model with consistent spatiotemporal dimensions and the output of the multi-site fine particle diffusion model are integrated to construct spatiotemporal four-dimensional data of time, space, pollutant concentration and chemical composition characteristics of pollutants; Based on the spatiotemporal four-dimensional data, a regional fine particulate matter component distribution map is generated using a GIS platform, and the regional fine particulate matter component distribution map is used to dynamically present the concentration changes and spatial distribution of pollutant components at each station; generating a concentration thermodynamic map according to the regional fine particulate matter component distribution map, and obtaining a diffusion path of pollutants according to the concentration thermodynamic map; Based on the concentration thermodynamic map, the transmission path of pollutants is obtained by combining the wind rose diagram in the meteorological data and the geographical feature map; The output of the EnvMulti-KAN model and the output of the multi-site fine particulate matter diffusion model are decomposed using positive definite matrix factorization to obtain the contribution ratios of different pollution source types, and the pollution source with the greatest influence is obtained based on the contribution ratios of different pollution source types.

7. A regional pollutant concentration prediction device, characterized in that: The device comprises: Historical data acquisition module: used to obtain pollutant concentration monitoring data and meteorological data of each single site in the target area for a preset period of time, wherein the pollutant concentration monitoring data includes the concentration of each pollutant and the chemical composition of each pollutant; Single-site prediction module: used to input the pollutant concentration monitoring data and meteorological data of each single site in the preset historical time period as input data into the pre-trained EnvMulti-KAN model, and predict the fine particle concentration and fine particle composition of the single site in the future period of time through the EnvMulti-KAN model; Real-time data acquisition module: used to obtain geographic feature data and real-time meteorological data of each single site in the target area; Multi-site input module: used to input the fine particle concentration and fine particle composition, geographic feature data and real-time meteorological data of all single sites in the future period into the pre-trained multi-site fine particle diffusion model; Regional prediction module: used to output the predicted value of pollutant concentration in the target area through the multi-site fine particulate matter diffusion model, as well as the contribution of meteorological factors, geographical factors and each single site to the concentration of each pollutant in the target area.

8. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the main controller, each step in the method for predicting regional pollutant concentration as described in any one of claims 1-6 is implemented.

Images