Air quality prediction method and device

Abstract

The present application discloses an air quality prediction method and device. Among them, the air quality prediction method includes: constructing a directed graph structure associated with multiple monitoring objects; in the directed graph structure, determining the node attribute information of each node, wherein the node attribute information at least includes the pollutant emission information of the monitoring object corresponding to the corresponding node; determining the trajectory point ratio of at least one incoming edge pointing to the same successor node in the directed graph structure, and determining the edge attribute information of at least one incoming edge based on the trajectory point ratio; constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; inputting the graph structure data into the graph neural network model for message transmission to obtain the corresponding graph structure features; and based on the graph structure features, performing air quality prediction related to air pollutants for the target object among the multiple monitoring objects. The accuracy of air quality prediction is improved through the present disclosure.

Description

Technical Field

[0001] The present application relates to the technical field of air quality detection, and in particular to an air quality prediction method and device. Background Art

[0002] In the field of existing air quality prediction technology, deep learning-based methods typically train models based on meteorological data and geographic information, without considering pollution emissions. During the backward trajectory data processing phase, the inability to deeply integrate pollution emission data with trajectory data makes it impossible to fully restore the traceability path and transmission evolution of pollutants. This makes it difficult to quantify the impact of pollution sources in different regions on the air quality of the target area. Ultimately, this leads to significant deviations in prediction results, making it impossible to meet the demand for accurate air quality forecasts.

[0003] With respect to the technical problem of inaccurate air quality prediction in the above-mentioned prior art, no effective solution has been proposed so far. Summary of the Invention

[0004] The embodiments of the present disclosure provide an air quality prediction method and apparatus to at least solve the technical problem of inaccurate air quality prediction in the prior art.

[0005] According to one aspect of an embodiment of the present disclosure, an air quality prediction method is provided, comprising constructing a directed graph structure associated with a plurality of monitoring objects, wherein each of the plurality of monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects, and the directed edges of the directed graph structure are established according to the circulation of air pollutants between the monitoring objects; determining node attribute information of each node in the directed graph structure, wherein the node attribute information includes at least pollutant emission information of the monitoring object corresponding to the corresponding node; determining the proportion of trajectory points of at least one incoming edge pointing to the same successor node in the directed graph structure, and determining edge attribute information of the at least one incoming edge based on the trajectory point proportion, wherein the trajectory point proportion is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of a first monitoring object corresponding to the same successor node between the at least one incoming edge; constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; inputting the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and performing air quality prediction related to the air pollutants for a target object among the plurality of monitoring objects based on the graph structure features.

[0006] According to another aspect of an embodiment of the present disclosure, a storage medium is further provided, the storage medium including a stored program, wherein when the program is run, a processor executes any one of the above methods.

[0007] According to another aspect of the embodiment of the present disclosure, an air quality prediction device is also provided, including: a directed graph structure construction module, used to construct a directed graph structure associated with multiple monitoring objects, wherein each of the multiple monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects respectively, and the directed edges of the directed graph structure are established according to the circulation of air pollutants between the monitoring objects; a determination module, used to determine the node attribute information of each node in the directed graph structure, wherein the node attribute information at least includes the pollutant emission information of the monitoring object corresponding to the corresponding node; and to determine the node pointing to the same successor node in the directed graph structure. The invention relates to a method for determining the ratio of trajectory points of at least one incoming edge, and determining the edge attribute information of at least one incoming edge according to the ratio of trajectory points, wherein the ratio of trajectory points is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of the first monitoring object corresponding to the same successor node between the at least one incoming edge; a graph structure data construction module, for constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; a processing module, for inputting the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and based on the graph structure features, performing air quality prediction related to the air pollutants for the target object among the multiple monitoring objects.

[0008] According to another aspect of the embodiment of the present disclosure, an air quality prediction device is also provided, comprising: a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: constructing a directed graph structure associated with multiple monitoring objects, wherein each of the multiple monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects, and the directed edges of the directed graph structure are established according to the circulation of air pollutants between the monitoring objects; in the directed graph structure, determining node attribute information of each node, wherein the node attribute information at least includes pollutant emission information of the monitoring object corresponding to the corresponding node; Determine the proportion of trajectory points of at least one incoming edge pointing to the same successor node in the directed graph structure, and determine the edge attribute information of the at least one incoming edge based on the trajectory point proportion, wherein the trajectory point proportion is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of the first monitoring object corresponding to the same successor node among the at least one incoming edge; construct graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; input the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and based on the graph structure features, perform air quality prediction related to the air pollutants for the target object among the multiple monitoring objects.

[0009] In the embodiment of the present disclosure, by associating multiple monitoring objects with a directed graph structure, each monitoring object corresponds to a node, and directed edges are established based on the circulation of air pollutants between the monitoring objects, clearly showing the propagation relationship of pollutants between the various monitoring objects, and using the pollutant emission information of the monitoring objects as the node attributes of the nodes, the model can take into account the situation of pollution emission information and more comprehensively reflect the influencing factors of air quality. By calculating the proportion of trajectory points of at least one incoming edge pointing to the same successor node, the edge attribute information is determined, which helps to quantify the distribution of pollutants on different transmission paths, thereby more accurately describing the source and transmission process of pollutants, and can comprehensively reflect the relationship between monitoring objects and the relevant characteristics of pollutants. Based on the extracted graph structure features, the air quality of the target object is predicted, and the impact of multiple factors on air quality is comprehensively considered, thereby achieving accurate prediction of air quality. This solves the technical problem of inaccurate air quality prediction in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of this application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure. In the drawings:

[0011] Figure 1is a hardware structure block diagram of a computing device for implementing the method according to embodiment 1 of the present disclosure;

[0012] Figure 2A A schematic diagram of an air pollution monitoring system for implementing the air quality prediction method described in Example 1 is shown;

[0013] Figure 2B Shown Figure 2A Module architecture diagram of the monitoring platform;

[0014] Figure 2C A schematic diagram of a deep learning model for air quality prediction in Example 1 of the present disclosure is shown;

[0015] Figure 3 A flowchart of a method for implementing air quality prediction according to the first aspect of embodiment 1 of the present disclosure is shown;

[0016] Figure 4A A schematic diagram showing the directed graph structure described in Example 1 of the present disclosure is shown;

[0017] Figure 4B A schematic diagram showing the distribution of monitoring objects corresponding to some nodes in the directed graph structure is shown;

[0018] Figure 5A A schematic diagram of a geographical grid corresponding to a monitoring object is shown when the monitoring object is an air monitoring station;

[0019] Figure 5B A schematic diagram of a geographical grid corresponding to a monitored object is shown when the monitored object is a city;

[0020] Figure 6 A schematic diagram showing how to use a graph neural network in a deep learning model to pass messages to graph structured data.

[0021] Figure 7 A schematic diagram of air quality prediction using the GRU neural network in a deep learning model is shown;

[0022] Figure 8 A schematic diagram showing the distribution of air monitoring stations in and around the city being monitored;

[0023] Figure 9 A schematic diagram of an air quality prediction device according to embodiment 2 of the present disclosure is shown;

[0024] Figure 10 A schematic diagram of an air quality prediction device according to Example 3 of the present disclosure is shown. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present disclosure.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] First, some nouns or terms that appear in the description of the embodiments of the present disclosure are subject to the following explanations:

[0028] Monitoring object: a facility used to monitor air pollutants. In some examples of the present disclosure, a monitoring object can be understood as an air monitoring station or a city that can monitor air pollutants.

[0029] Directed graph structure: A directed graph consists of a set of vertices and a set of directed edges, with each directed edge connecting a pair of ordered vertices. In some examples of this disclosure, the vertices in the directed graph structure can be understood as nodes. Each monitored object corresponds to its own node in the directed graph structure. The existence of directed edges between nodes in the directed graph structure is determined based on the flow of air pollutants between the monitored objects.

[0030] Backward trajectory point: a trajectory point on the propagation path of an air pollutant determined by backward trajectory analysis, determined based on the propagation path of the air pollutant. In some examples of the present disclosure, edge attribute information of each directed edge in a directed graph structure can be determined based on the backward trajectory point.

[0031] Successor node: In some examples of the present disclosure, a successor node can be understood as a node pointed to by a directed edge in a directed graph structure.

[0032] Predecessor node: In some examples of the present disclosure, the predecessor node can be understood as a source node where a directed edge starts in a directed graph structure.

[0033] Specified time: In some examples of this disclosure, the specified time can be understood as a future time relative to the current time, or a historical time relative to the current time. When the specified time is a future time, the air quality prediction method of this application can be used in air quality forecasting scenarios; when the specified time is a historical time, the air quality prediction method of this application can be used in scenarios where historical pollutant information is supplemented or verified.

[0034] Example 1

[0035] According to this embodiment, an embodiment of a method for air quality prediction is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1 FIG1 shows a hardware structure block diagram of a computing device for implementing a method for air quality prediction. Figure 1 As shown, a computing device may include one or more processors (the processor may include, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA) or other processing device), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, transmission device, and input / output interface are connected to the processor via a bus. In addition, it may also include: a display, a keyboard, and a cursor control device connected to the input / output interface. Those skilled in the art will understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0037] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be fully or partially integrated into any of the other components of the computing device. As discussed in the embodiments of the present disclosure, the data processing circuitry functions as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0038] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the air quality prediction in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the air quality prediction method of the above-mentioned application. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0039] The transmission device is used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by a communications provider of the computing device. In one embodiment, the transmission device includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0040] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computing device.

[0041] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing device described above.

[0042] Figure 2AFIG2 shows a schematic diagram of an air pollution monitoring system for implementing the air quality prediction method described in this embodiment. Figure 2A As shown, the air pollution monitoring system includes: multiple monitoring objects S 1~ S m ; and the monitored objects S 1~ S m The air pollution monitoring platform 100 (hereinafter referred to as the "monitoring platform") is connected to the communication. Figure 2A In this paper, multiple air monitoring stations (hereinafter referred to as "stations") are used as monitoring objects. S 1~ S m However, cities can also be used as monitoring objects, so that each city can use air monitoring equipment to monitor the air pollution of the corresponding city. I will not go into details here.

[0043] Figure 2B FIG. 1 shows a module architecture diagram of the monitoring platform 100. Figure 2B As shown, the monitoring platform 100 includes a data acquisition unit 110, a model training unit 120 and a daily forecast unit 130. The data acquisition unit 110 is used to collect air quality data, meteorological data, and transmission trajectory data required by the model training unit 120 and the daily forecast unit 130. The model training unit 120 is used to construct a training sample set for a deep learning model including a graph neural network based on the data collected by the data acquisition unit 110, and use the training sample set to train the deep learning module. The daily forecast unit 130 is used to predict the air quality (such as the concentration of pollutants) corresponding to the specified target monitoring object at a specified time (such as a future time of the current time or a time in the past history) based on the data collected by the data acquisition unit 110.

[0044] exist Figure 2B In the embodiment, the data collection unit 110 includes an air quality data downloading module 111 , an atmospheric reanalysis data downloading module 112 , a backward trajectory data downloading module 113 and a weather forecast data downloading module 114 .

[0045] The air quality data download module 111 is used to obtain hourly air quality data monitored by the monitoring object. The air quality data includes, for example, PM 2.5 、PM 10 , O3, NO2, SO2 and CO, etc.

[0046] The atmospheric reanalysis data download module 112 is configured to obtain meteorological information and a gridded inventory related to atmospheric pollutant emissions based on the operating frequency of the training unit 120 (e.g., monthly or weekly). The gridded inventory indicates pollutant emission values ​​within each geographic grid covering the geographic area as grid values ​​for each geographic grid. The operating frequency of the atmospheric reanalysis data download module 112 depends on the operating frequency of the model training unit 120.

[0047] The backward trajectory data download module 113 is used to obtain historical trajectory point data of air pollutants and backward trajectory data point data of air pollutants at a specified time. For example, GDAS1 (Global Data Assimilation System, 1-degree) global data assimilation system product data can be obtained from the National Centers for Environmental Prediction (NCEP) of the United States and used as historical trajectory point data and NCEP GFS (Global Forecast System) forecast data, and NCEP GFS data can be used as backward trajectory point data of air pollutants at future times.

[0048] The weather forecast data download module 114 is used to obtain weather forecast data for a future simulation period corresponding to the monitored object. For example, the operating frequency of the weather forecast data download module depends on the updating frequency of the weather forecast data.

[0049] The model training unit 120 includes a first meteorological data processing module 121, a first air quality data processing module 122, a first backward trajectory data processing module 123, a first graph structure generation module 124 and a deep learning model training module 125. Figure 2C As shown, the deep learning models used in this application include graph neural network GNN and recurrent neural network GRU.

[0050] Before running the model training unit 120, the data acquisition unit 110 first obtains a list of basic station and city information (e.g., name, latitude and longitude, code, etc.). The time range of the training dataset for the model training is obtained, measured in years. The meteorological elements used by the model training unit 120, the specific types of predicted atmospheric pollution factors, the specific duration of the backpropagation of the transmission trajectory, the gridded list, and the digital elevation data storage path are all modifiable information and are defined in the corresponding configuration file. This allows the inputs to the various modules of the model training unit 120 to be controlled based on the data in the configuration file and in combination with the data collected by the data acquisition unit 110.

[0051] The first air quality data processing module 122 is responsible for dividing the data acquired by the data acquisition unit 110 according to a specified time. The historical pollutant information corresponding to the historical time before the specified time (for example, the historical O3 concentration data for the station 24 or 48 hours before the specified time) can be used as part of the node attribute information of the directed graph structure; and the data at the specified time can be used as the training set label data for the model training unit 120. In other words, the first air quality data processing module 122 is responsible for processing and generating the training set label data corresponding to the city or station and the historical pollutant information as part of the node attribute information required by the deep learning model training module 125.

[0052] The first air quality data processing module 122 operates as follows: first, it obtains the code of the city or station for which forecasts are required based on the basic information list of cities and stations in the configuration file; then, it determines the training time range and the atmospheric pollution factor (e.g., O3) to be simulated based on the configuration file; then, it filters the data obtained from the air quality data download module to obtain air quality data (e.g., O3 concentration data) for the target city or station. The data is then divided into label data and historical pollutant information.

[0053] The first meteorological data processing module 121 extracts the meteorological factors corresponding to the specified time according to the configuration file, and then constructs node attribute information with the pollutant history information of the air quality data processing module.

[0054] The first meteorological data processing module 121 is responsible for processing and generating some node attribute information required by the deep learning model training module 125. The operation process of the first meteorological data processing module 121 can be understood as follows: based on the meteorological elements to be extracted, the time range, and the basic information list of cities and stations in the configuration file, the historical meteorological reanalysis data downloaded by the atmospheric reanalysis data download module 112 is extracted by performing meteorological element, time, and grid value extraction (where the grid value corresponds to pollutant emission information within the geographic grid, such as O3 emissions), obtaining the historical meteorological data and pollutant emission information for the corresponding cities and stations, and using them as some node attribute information in the directed graph structure.

[0055] The first backward trajectory data processing module 123 simulates the backward propagation trajectory of the node, so as to determine the edge attribute information of the directed edge in the directed structure graph according to the backward trajectory points.

[0056] The first graph structure generation module 124 considers the horizontal distance between nodes and the terrain height to construct the connection characteristics of nodes and edges; combines the atmospheric emission inventory data and time characteristic data with the aforementioned pollutant historical information and meteorological factors to form node attribute information; and composes graph structure data based on the node attribute information and edge attribute information.

[0057] The deep learning model training module 125 inputs the graph structure data into the deep learning model, calculates the loss function based on the prediction results output by the deep learning model and the training set labels, and then backpropagates to update the parameters to finally obtain the trained model parameters.

[0058] The daily forecast unit 130 includes a second meteorological data processing module 131 , a second air quality data processing module 132 , a second backward trajectory data processing module 133 , a second graph structure generating module 134 and an air quality forecast module 135 .

[0059] Based on the data acquired by the data acquisition unit 110, the second air quality data processing module 132 obtains historical pollutant information corresponding to historical moments prior to the specified time (e.g., historical O3 concentration data for the site 24 or 48 hours prior to the specified time) as partial node attribute information for the nodes of the directed graph structure. Specifically, the second air quality data processing module 132 is responsible for processing and generating the pollutant historical information as partial node attribute information required by the air quality forecasting module 135 for subsequent air quality prediction. For more details, please refer to the first air quality data processing module 122.

[0060] The second meteorological data processing module 131 extracts the meteorological factors at the specified time according to the configuration file, and then combines them with the pollutant history information from the air quality data processing module to form partial node attribute information for subsequent air quality prediction. For details, please refer to the first meteorological data processing module 121.

[0061] The second backward trajectory data processing module 133 simulates the backward propagation trajectory of the node and determines the edge attribute information of the directed edge in the directed structure graph according to the trajectory for subsequent air quality prediction.

[0062] The second graph structure generation module 134 considers the horizontal distance between nodes and the terrain height to construct the connection characteristics of nodes and edges; combines the atmospheric emission inventory data and time characteristic data with the aforementioned pollutant historical information and meteorological factors to form node attribute information; and based on the node attribute information and edge attribute information, together constitutes the graph structure data.

[0063] The air quality forecast module 135 inputs the graph structure data into the trained deep learning model, aggregates and updates the spatial information through the graph neural network, and then inputs it into the recurrent neural network GRU to update the time information, thereby predicting the air quality of the target object in the monitoring object and the air pollutants (for example, predicting the O3 concentration at a specified time).

[0064] Under the above operating environment, according to the first aspect of this embodiment, an air quality prediction method is provided. Figure 2A The monitoring platform 100 shown in FIG. Figure 3 A flow chart showing a method for implementing air quality prediction is shown. Figure 3 As shown, the method includes:

[0065] S302: Constructing a directed graph structure associated with a plurality of monitoring objects, wherein each of the plurality of monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects, and the directed edges of the directed graph structure are established based on the mobility of air pollutants between the monitoring objects;

[0066] S304: Determine node attribute information of each node in the directed graph structure, wherein the node attribute information at least includes pollutant emission information of the monitoring object corresponding to the corresponding node;

[0067] S306: Determine, in the directed graph structure, a trajectory point ratio of at least one incoming edge pointing to the same successor node, and determine edge attribute information of the at least one incoming edge based on the trajectory point ratio, wherein the trajectory point ratio is used to indicate a distribution ratio of multiple backward trajectory points of the air pollutant of the first monitoring object corresponding to the same successor node between the at least one incoming edge;

[0068] S308: Constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information;

[0069] S310: Inputting the graph structure data into the graph neural network model for message transmission to obtain corresponding graph structure features; and

[0070] S312: Based on the graph structure features, air quality prediction related to air pollutants is performed for the target object among the multiple monitoring objects.

[0071] Specifically, after obtaining the basic information (name, longitude and latitude, code, etc.) of the national sites and cities, the monitoring platform 100 establishes a S 1~ S m Associated directed graph structure (S302). Figure 4A A schematic diagram of a directed graph structure is shown. The directed graph structure includesS 1~ S m Corresponding nodes N 1~ N m . And further reference Figure 4A As shown, each edge in the directed graph structure E 1~ E L is a directed edge. And each edge E 1~ E L According to each monitoring object S 1~ S m The mobility of air pollutants (e.g., O3) between Figure 4A In the node N 3 and nodes N 4, no edge is built between them, which means that the nodes N 3 and nodes N 4. The connections between nodes can be represented by an adjacency matrix. The specific method for establishing a directed graph structure will be explained in detail below.

[0072] Then, the monitoring platform 100 determines the data of each node through the second meteorological data processing module 131 and the second air quality data processing module 132. N 1~ N m Corresponding node attribute information An 1~ An m ,in An 1~ An m ∈R 39*1 (S304). Node attribute information An 1~ An m At least include the corresponding node N 1~ N m Corresponding monitoring objects S 1~ S m Pollutant emission information, such as monitoring objects S 1~ S m The O3 emissions in the area. As for the specific determination of node attribute information An 1~ An m The method will be described in detail below.

[0073] Then, the monitoring platform 100 determines the proportion of trajectory points of at least one incoming edge pointing to the same successor node in the directed graph structure. Figure 4A As shown, the successor node in this application refers to a node specified by a directed edge. E 1, the successor node is N 1; For directed edges E 2, the successor node is N 4; Similarly, for directed edges E L For N m . And for these successor nodes, the directed edges pointing to the node are in-edges. For example, the directed edge E 1 is the successor node N 1 incoming edge; directed edge E 2 is the successor node N 4's incoming edge; and so on, directed edges E L Is the successor node N m Thus, continue to refer to Figure 4A As shown, the node N 1 as an example, the node N 1 is a successor node with multiple incoming edges E 1. E 3 and E 5. That is, for the same successor node, there can be multiple incoming edges. It should be noted that even for the same node, with respect to different directed edges, it can be both a successor node and a predecessor node. For example, node N 1 relative to directed edges E 5 is the successor node, but relative to the directed edge E 6 is the predecessor node, and so on.

[0074] In addition, the second backward trajectory data processing module 133 of the monitoring platform 100 can simulate the backward propagation trajectory of O3 corresponding to the monitored object. Figure 4B Further shows some nodes in the directed graph N 1~ N 4 Corresponding monitoring objects S 1~ S 4 distribution diagram. Figure 4B The black dots shown in the figure are the monitored objects. S 1 (i.e. the first monitoring object) corresponds to O3 relative to the specified time t Multiple backward trajectory points (for example, in this embodiment, the specified time tThe second backward trajectory data processing module 133 further determines the backward trajectory points 24 or 48 moments before the start of the second backward trajectory data processing module 133. Figure 4B The multiple backward trajectory points shown are at the node N Multiple incoming edges of 1 E 1. E 3 and E 5. And according to the determined allocation ratio, determine the input edge E 1. E 3 and E 5 edge attribute information Ae 1. Ae 3 and Ae 5 (S306). And by analogy, the monitoring platform 100 can determine the direction of each directed edge based on each successor node in the same way. E 1~ E L Edge attribute information Ae 1~ Ae L .

[0075] Then, the second graph structure generation module 134 of the monitoring platform 100 generates the node attribute information of each node according to the determined node attribute information. An 1~ An m and edge attribute information of each directed edge Ae 1~ Ae L , build with Figure 4A Graph structure data corresponding to the directed graph structure described in (S308).

[0076] Then, the air quality forecast module 135 of the monitoring platform 100 inputs the graph structure data into the graph neural network of the deep learning model to obtain graph structure features corresponding to the graph structure data (S310). The specific process will be described in detail below.

[0077] Finally, the air quality forecast module 135 of the monitoring platform 100 performs a multi-monitoring operation based on the graph structure feature. S 1~ S m For example, the target object is predicted to be S 1At a specified time t The specific prediction process will be described in detail below.

[0078] According to the embodiment of the present disclosure, a directed graph structure is established based on the circulation of air pollutants between the monitored objects, which intuitively and accurately shows the degree of correlation between the propagation directions of air pollutants between different monitored objects. In addition, pollutant emission information is added to the node attribute information of the directed graph structure, which improves the accuracy of air quality prediction compared to the method of using meteorological data and geographic information to predict air quality. In addition, the proportion of input edge trajectory points pointing to the same successor node is determined and the edge attribute information is determined accordingly. Since the proportion of backward trajectory points reflects the transmission trajectory of air pollutants to the monitored object, the weight of the impact of the transmission path of air pollutants on the air quality of the monitored object is clarified. Constructing graph structure data based on node attributes and edge attributes, and inputting it into the graph neural network model for message passing, the model can fully learn the spatiotemporal characteristics in the air quality data. As a result, the graph neural network model can better capture the intrinsic connection between air quality data between different monitored objects and at different time points, thereby improving the ability to predict the trend of air quality changes. This solves the technical problem of inaccurate air quality prediction in the prior art.

[0079] Optionally, the operation of constructing a directed graph structure associated with multiple monitoring objects includes: when a first distance between two monitoring objects is less than a first threshold and there is no terrain with an altitude greater than a second threshold between the two monitoring objects, constructing a directed edge between the nodes corresponding to the two monitoring objects.

[0080] Specifically, during the construction of the directed graph structure, the second graph structure generation module 134 obtains the digital elevation stored in the path and the list of basic information of sites and cities according to the configuration file, thereby determining whether directed edges can be constructed between the nodes of the directed graph structure.

[0081] Specifically, the second graph structure generating module 134 determines whether the distance between two monitoring objects (ie, the first distance) is less than a first threshold value according to the station and city basic information list. Figure 4B As shown, for example, the monitoring object can be determined S 1~ S 4 (ie, the first distance), and determines whether it is less than the first threshold.

[0082] Then, in the monitoring object S 1~ S 4 is less than the first threshold, it is determined whether there is a terrain with an altitude greater than the second threshold between the two monitoring objects based on the digital elevation data. If there is no terrain with an altitude greater than the second threshold between the two monitoring objects, a directed edge is constructed between the nodes corresponding to the two monitoring objects. For example, Figure 4A As shown, due to the monitoring object S 1~ S3 is less than the first threshold, and there is no terrain with an altitude greater than the second threshold between the monitoring objects, so the node N 1~ N 3 can build directed edges between each other. For example, monitoring objects S 3 and S 4 may be greater than the first threshold, or the monitoring object S 3 and S 4 There is a terrain with an altitude greater than the second threshold, so that the node N 3 and N 4 cannot construct directed edges between them.

[0083] According to the embodiment of the present disclosure, digital elevation data is used to construct directed edges by comprehensively considering the distances and terrain factors between different monitoring objects, thereby accurately simulating the transmission paths of air pollutants, and further accurately reflecting the flow of pollutants between different monitoring objects, thereby improving the accuracy of air quality predictions.

[0084] When it is determined that there are directed edges in the directed graph structure, edge attribute information of each directed edge is further determined.

[0085] Optionally, the operation of determining the proportion of trajectory points of each incoming edge pointing to the same successor node in a directed graph structure includes: determining multiple backward trajectory points of air pollutants with respect to the first monitoring object at a specified future time; determining the farthest backward trajectory point with the farthest distance from the first monitoring object among the multiple backward trajectory points, and determining the second distance between the first monitoring object and the farthest backward trajectory point; determining an adjacent predecessor node among the predecessor nodes of the successor node, wherein the distance between the second monitoring object corresponding to the adjacent predecessor node and the first monitoring object is less than the second distance, and determining a first incoming edge corresponding to the adjacent predecessor node among each incoming edge; and setting the proportion of trajectory points of the second incoming edges other than the first incoming edge in each incoming edge to zero, and determining the proportion of trajectory points corresponding to the first incoming edge based on the multiple backward trajectory points.

[0086] Specifically, Figure 4A The node that is the successor node in N 1 as an example to illustrate how to determine the proportion of trajectory points of each incoming edge pointing to the same successor node.

[0087] First, the second graph structure generating module 134 determines the node N 1Corresponding monitoring object S 1 (i.e. the first monitored object) relative to the specified time t Multiple backward trajectory points of Figure 4BAs shown. The multiple backward trajectory points can be determined by, for example, the second transmission trajectory data processing module 133 using the hysplit model to simulate the backward propagation trajectory of the air pollutant (such as O3) corresponding to the monitoring object. For example, the multiple backward trajectory points can be from the specified time t Calculated from the monitoring object S 1 corresponds to the 24 hourly trajectory points of O3 backpropagated for 24 hours.

[0088] Then, further reference Figure 4B As shown, the second graph structure generating module 134 determines the points corresponding to the monitored object in the multiple backward trajectory points. S 1. The farthest backward trajectory point (for example, the trajectory point at the 24th hour of backward propagation). In addition, the monitoring object can be determined. S 1 and the distance between the farthest backward trajectory point (i.e., the second distance).

[0089] Then the second graph structure generating module 134 is used to monitor the object S 1 is the center of the circle, with the farthest backward trajectory point and the monitored object S 1 is the radius, which determines the monitoring object S 1 is the circular area with the center (see Figure 4B Then, in the circle area with the node N 1's predecessor node N 2~ N 4 Corresponding monitoring objects S 2~ S 4. Determine the monitoring object within the circular area S 2 and S 3 (i.e. the second monitoring object). And further determine the monitoring object S 2 and S 3 corresponding nodes N 2 and N 3 (i.e., adjacent predecessor node).

[0090] Then, the second graph structure generation module 134 further generates the graph structure from the input edge. E 1. E 3 and E Determine the slave node in 5 N 2 and N 3 points to the node N 1's incoming edge E 5 and E 3 (i.e. the first incoming edge).

[0091] Then, since the monitoring object S 4 and monitoring objects S 1 is greater than the second distance mentioned above, so it is closer to the incoming edge E1 (i.e., the second input edge) is set to 0. Figure 4B The backward trajectory points shown determine the incoming edge E 5 and E 3 (i.e. the first incoming edge) of the trajectory point ratio. E 5 and E 3 (i.e., the first incoming edge) trajectory point ratio method is described in detail below.

[0092] Thus, according to the embodiment of the present disclosure, by determining that the first monitoring object is at a specified time in the future t The backward trajectory points of air pollutants and the farthest backward trajectory point are found to clarify the possible farthest source location of air pollutants, and then accurately identify the range of air pollutants that affect the air quality of the first monitored object.

[0093] Optionally, the operation of determining the proportion of trajectory points corresponding to the first input edge based on multiple backward trajectory points includes: determining the connecting lines between each second monitored object and the first monitored object; from the first input edge, determining the input edge corresponding to the connecting line with the smallest distance between the backward trajectory points as the input edge corresponding to the backward trajectory point; and determining the corresponding trajectory point proportion based on the number of backward trajectory points corresponding to the first input edge.

[0094] Specifically, further reference Figure 4A and Figure 4B , after determining the slave node N 2 and N 3 points to the node N 1's incoming edge E 5 and E 3 (i.e., the first input edge), the second graph structure generation module 134 determines the monitoring object S 2 and S 3 and monitoring objects S The connection between 1 L 12 and L 13 (See Figure 4B Indicated by the dashed line). L 12 With the incoming edge E 5 corresponding, connecting L 13 With the incoming edge E 3 correspondences.

[0095] Then, see further Figure 4B As shown, the second graph structure generating module 134 calculates the relationship between each backward trajectory point and the connecting line L 12 and L 13Specifically, the distance between the backward trajectory points and the connecting line is calculated separately. L 12 and L 13 Then, the incoming edge corresponding to the line with the smallest distance is taken as the incoming edge corresponding to the backward trajectory point. For example, for a backward trajectory point, when it reaches the line L 12 The distance is less than the connecting line L 13 If the distance is less than , then the edge is determined E 5 is the incoming edge corresponding to the backward trajectory point. For another example, for another backward trajectory point, when it reaches the line L 13 The distance is less than the connecting line L 12 If the distance is less than , then the edge is determined E 3 is the incoming edge corresponding to the backward trajectory point. Similarly, for Figure 4B For each backward trajectory point shown in , confirm the corresponding incoming edge.

[0096] Then, the second graph structure generation module 134 determines the incoming edge E 5 and entering edge E 3 corresponding to the number of backward trajectory points, and further determine the number of backward trajectory points corresponding to the input edge E 5 and entering edge E 3 corresponds to the trajectory point ratio.

[0097] As for the entry E 1. Due to the monitoring object S 4 and monitoring objects S 1 is greater than the second distance mentioned above, so it is closer to the incoming edge E The trajectory point ratio corresponding to 1 is set to 0.

[0098] Therefore, according to the disclosed embodiments, by determining the connection between each second monitored object and the first monitored object, and then finding the incoming edge corresponding to the connection with the smallest distance to the backward trajectory point, the key transmission path affecting the air quality of the first monitored object can be accurately located. Furthermore, the trajectory point ratio is determined based on the number of backward trajectory points corresponding to each incoming edge, thereby quantifying the weight of different transmission paths in the pollution transmission process.

[0099] Optionally, the operation of determining the edge attribute information of each incoming edge based on the trajectory point ratio includes: determining the distance information between the monitoring objects corresponding to the predecessor node and the successor node of each incoming edge; and determining the edge attribute information of each incoming edge based on the distance information and the trajectory point ratio.

[0100] Specifically, still refer to Figure 4Aand Figure 4B As shown, the second graph structure generating module 134 determines the node N 1 corresponding to the incoming edge E 1. E 3 and E 5. After that, further determine the monitoring object S 1 and S 2 as the distance between the edge E 5 corresponding distance information, and determine the monitoring object S 1 and S 3 as the distance between the edges E 3 corresponding distance information, and determine the monitoring object S 1 and S 4 as the distance between the edges E 1 corresponds to the distance information.

[0101] Then, the second graph structure generation module 134 generates the graph structure according to the input edge. E 1. E 3 and E 5 Corresponding distance information and trajectory point ratio to determine the edge E 1. E 3 and E 5 edge attribute information Ae 1. Ae 3 and Ae 5 .

[0102] Therefore, according to the embodiment of the present disclosure, by determining the distance information between the monitoring objects corresponding to the predecessor node and the successor node of the incoming edge, and combining the trajectory point ratio to determine the edge attribute information, it is possible to more accurately describe the characteristics of the directed edges in the directed graph. The distance information reflects the spatial proximity of different monitoring objects, while the trajectory point ratio reflects the distribution of pollutants on different transmission paths. The combination of the two allows the edge attribute information to more comprehensively and meticulously reflect the difficulty and possibility of pollutants being transmitted from one monitoring object to another.

[0103] Optionally, the operation of determining the node attribute information of each node includes: obtaining historical pollutant information related to air pollutants of the monitored object at historical moments before the specified moment and meteorological information of the monitored object at the specified moment; determining the pollutant emission information of the monitored object corresponding to the specified moment; and determining the node attribute information of the node corresponding to the monitored object based on the pollutant historical information, pollutant emission information and meteorological information.

[0104] Specifically, Figure 4A The nodes shown in N 1 and Figure 4B Monitoring objects in S1 is used as an example. In the process of constructing directed graph data, the second graph structure generating module 134 obtains the node from the second air quality data processing module 132. N 1Corresponding monitoring object S 1 pollutant history information. For example, in this embodiment, the specified time t Previous 24 moments and monitored objects S The O3 concentration information corresponding to 1 is recorded as the pollutant history information. an 1,1 ~ an 1,24 That is, node attribute information An The first 24 elements of 1 an 1,1 ~ an 1,24 Historical information of pollutants.

[0105] Then, the second graph structure generating module 134 obtains the node N 1Corresponding monitoring object S 1At a specified time t In this embodiment, 10 different meteorological factors can be taken as meteorological information, such as wind speed, temperature, humidity, precipitation, etc. Thus, the second graph structure generation module 134 will monitor the object S 1At a specified time t The weather information is recorded as an 1,25 ~ an 1,34 , representing node attribute information An 1's 25th to 34th elements.

[0106] Then, the second graph structure generating module 134 obtains the node from the second meteorological data processing module 131. N 1Corresponding monitoring object S 1At a specified time t The O3 emissions (i.e., pollutant emission information) are recorded as an 1,35 , representing node attribute information An The 35th element of 1.

[0107] Finally, the second graph structure generating module 134 obtains the data corresponding to the specified time from the second meteorological data processing module 131. t The corresponding year, month, week, and hour are recorded as an 1,36 ~ an 1,39 , representing node attribute information An The 36th to 39th elements of 1.

[0108] Thus, through the above operations, the second graph structure generating module 134 constructs N 1 corresponding node attribute information An 1∈R 39*1 For other nodes N 2~ N m , you can refer to the above operations to build the corresponding node attribute information.

[0109] In addition, this embodiment schematically shows that the node attribute information has 39 elements, but the specific number of elements can be adjusted according to actual conditions as long as it complies with the type of node attribute information defined above.

[0110] According to the embodiment of the present disclosure, the node attribute information is determined by comprehensively acquiring the historical pollutant information, pollutant emission information and meteorological information of the monitored object, which comprehensively covers the key factors affecting air quality. The historical pollution status of the monitored object is determined based on the historical pollutant information, thereby reflecting the changing trend of air quality at historical moments. In addition, since the air quality data corresponding to different monitored objects are different, and the factors affecting the air quality of different monitored objects are also different, the node attribute information of the node is determined by the historical pollutant information, pollutant emission information and meteorological information, so that the model can subsequently accurately predict the air quality of different monitored objects according to the specific conditions of different monitored objects.

[0111] Optionally, when the monitored object is an air monitoring station, the operation of determining the pollutant emission information of the station at a specified time includes: determining multiple geographic grids corresponding to the air monitoring station, wherein the geographic grid corresponding to the air monitoring station is located in the area covered by the multiple geographic grids; and averaging the pollutant emission information corresponding to the multiple geographic grids at the specified time as the pollutant emission information corresponding to the air monitoring station at the specified time.

[0112] Specifically, the second graph structure generating module 134 determines the specified time t When pollutant emission information is obtained, the grid inventory data is obtained from the storage path according to the configuration file, and the geographic grid corresponding to the monitored object is determined according to the geographic location information of the monitored object.

[0113] Specifically, the monitoring object S 1 as an example. Figure 5A As shown, in the monitoring object S 1 is a site, the second graph structure generating module 134 determines the monitoring object S 1 The corresponding geographic grid includes monitoring objects S 1 geographic grid and eight geographic grids surrounding the geographic grid.

[0114] Then, the second graph structure generating module 134 determines the gridded list at the specified time. t , the grid value corresponding to each geographic grid, wherein the grid value is used to indicate the pollutant emission information (such as O3 emission) within the corresponding geographic grid. In addition, the second graph structure generation module 134 calculates the average value of the grid values ​​of each geographic grid as the station at the specified time. t Pollutant emission information (such as O3 emissions).

[0115] Optionally, when the monitored object is a city, the operation of determining the pollutant emission information of the city at a specified time includes: determining multiple geographic grids corresponding to the area of ​​the city; and averaging the pollutant emission information corresponding to the multiple geographic grids and the specified time as the pollutant emission information corresponding to the city at the specified time.

[0116] Specifically, the monitoring object S 1 as an example. Figure 5B As shown, in the case where the monitoring object S1 is a city, the city area may include multiple geographical grids. Figure 5B As shown, the second graph structure generating module 134 determines the monitoring object S 1 The corresponding geographic grid includes the geographic grid within the city shp area.

[0117] Then, the second graph structure generating module 134 determines the gridded list at the specified time. t , the grid value corresponding to each geographic grid, wherein the grid value is used to indicate the pollutant emission information (such as O3 emission) within the corresponding geographic grid. In addition, the second graph structure generation module 134 calculates the average value of the grid values ​​of each geographic grid as the average value of the city at the specified time. t Pollutant emission information (such as O3 emissions).

[0118] Optionally, the graph structure data is input into the graph neural network model for message passing to obtain the corresponding graph structure feature operations, including: performing message passing on the directed edges of the graph structure data to determine the edge features corresponding to the directed edges; and performing message passing on the successor node based on the node attributes of the successor node in the graph structure data and the edge features of the incoming edges corresponding to the successor node to determine the node features of the successor node.

[0119] Specifically, Figure 6 Shown Figure 2C Schematic diagram of the graph neural network GNN shown in . Reference Figure 2CAs shown, according to this embodiment, the air quality forecast module 135 inputs the graph structure data constructed according to the above content into the deep learning model, thereby inputting the graph structure data into the graph neural network GNN.

[0120] Then, further reference Figure 6 As shown, a first perceptron and a second perceptron are deployed in the graph neural network GNN. The first perceptron is used to perform message passing operations on edge attribute information in the graph structure data. Preferably, the first perceptron can be a three-layer linear perceptron. The second perceptron is used to perform message passing operations on node attribute information in the graph structure data. Preferably, the second perceptron can be a one-layer linear perceptron.

[0121] Thus, reference Figure 6 As shown, the graph neural network GNN first performs a message passing operation on the edge attribute information of the graph structure data through the first perceptron to generate corresponding edge features; then the graph neural network GNN performs a message passing operation on the node attribute information of the graph structure data through the second perceptron to generate corresponding node features.

[0122] In this way, the graph neural network GNN generates nodes that are respectively connected to each other in the directed graph structure. N 1~ N m The corresponding node features and the directed edges E 1~ E L Corresponding edge features. Thus, the graph structure features are constructed.

[0123] According to the embodiment of the present disclosure, message passing is performed on the directed edges of the graph structure data, so that the information contained in the directed edges can be deeply mined and the edge features corresponding to the directed edges can be accurately determined. In the air quality prediction scenario, the directed edges represent the circulation relationship of air pollutants between different monitoring objects, and the edge features reflect the intensity, direction and other characteristics of this circulation. Through message passing, information such as the distance between the monitoring objects and the pollutant transmission rate can be integrated into the edge features, so that the model can more accurately capture the transmission pattern of pollutants between different areas, thereby more carefully characterizing the properties of the directed edges and providing a richer information basis for subsequent analysis. Message passing is performed based on the node attributes of the successor node and the edge features of the corresponding incoming edge, and then the node features of the successor node are determined. This fully considers the node's own attributes and the information of the directed edges connected to it, so that the node features can more comprehensively and accurately reflect the status of the monitoring object in the entire system. Compared with determining the node features based on only a single piece of information, it can more realistically reflect its actual situation during the air quality change process, thereby improving the accuracy and reliability of the node features.

[0124] Optionally, message passing is performed on the directed edges of the graph structure data to determine the edge features corresponding to the directed edges, including: determining the associated successor node and the associated predecessor node associated with the directed edge; determining the pollutant information corresponding to the associated successor node at the previous moment before the specified moment, and the pollutant information corresponding to the associated predecessor node at the previous moment; inputting the node attribute information of the associated successor node, the pollutant information corresponding to the associated successor node at the previous moment, the node attribute information of the associated predecessor node, the pollutant information corresponding to the associated predecessor node at the previous moment, and the edge attribute information of the directed edge into a pre-trained first linear perceptron to generate edge features corresponding to the directed edge.

[0125] Specifically, the directed edges in the graph structure data E 1 is used as an example. When performing message passing on the directed edge, the graph neural network GNN first determines the directed edge E 1 associated successor node N 1 and associated predecessor node N 4.

[0126] Then, the graph neural network GNN determines the following information: associated successor nodes N Node attribute information of 1 An 1 and at the specified time t The last moment t -1 O3 concentration; associated predecessor node N Node attribute information of 4 An 4 and at the specified time t The last moment t -1 O3 concentration; and directed edges E 1's edge attribute information.

[0127] Then, the graph neural network GNN fuses the above information (for example, splicing it) and inputs it into the first perceptron to generate the directed edge E 1 corresponds to the edge feature, which corresponds to the directed edge in the graph structure feature FE 1.

[0128] Thus, for each directed edge in the graph structure data E 1~ E L , the corresponding edge features can be determined according to the above operations, corresponding to the directed edges in the graph structure features FE 1~ FE L .

[0129] According to the embodiment of the present disclosure, the predecessor and successor nodes associated with the directed edge and their pollutant information at the previous moment are accurately determined, and the node attributes and edge attribute information are combined to achieve a deep fusion of the spatial dimension and the temporal dimension. In the air quality monitoring scenario, the transmission and diffusion of pollutants depends not only on the spatial distance between the monitoring points, but also on the concentration state at the previous moment. By fusing this information to generate edge features, the transmission law of pollutants in the spatiotemporal dimension can be more realistically portrayed, thereby enabling the model to more accurately capture the pollution propagation path and reduce prediction errors caused by feature loss or deviation.

[0130] Optionally, based on the node attributes of the successor node in the graph structure data and the edge features of the incoming edges corresponding to the successor node, a message is passed to the successor node to determine the node features of the successor node, including: taking the average of the edge features of the incoming edges corresponding to the successor node to obtain the aggregated edge features; and inputting the aggregated edge features and the node attributes of the successor node into the second perceptron to generate the node features corresponding to the successor node.

[0131] Specifically, the node N 1 as an example, to determine the successor node N 1. The operation of node features is described below. Figure 4A Knowable and Node N 1 The corresponding input edge includes the input edge E 1. E 3 and E 5. Determine the entry edge based on the above content E 1. E 3 and E 5 corresponding edge features, and then determine the input edge E 1. E 3 and E 5, and then get the average value of the edge features corresponding to the node N 1 The aggregate edge feature of multiple incoming edges.

[0132] Then, the node corresponding to N 1's aggregated edge features and nodes N Node attribute information of 1 An 1 is input into the second perceptron to obtain node N Node features of 1 fn 1, corresponding to the nodes in the graph structure feature FN 1.

[0133] Thus, for the nodes in the graph structure data N 1~ N m Adoption and Node N 1. In the same way, determine the corresponding node features fn 1~ fnm , corresponding to the nodes in the graph structure feature FN 1~ FN m .

[0134] According to the embodiment of the present disclosure, the edge features of the incoming edges corresponding to the successor node are averaged to obtain the aggregated edge features. The information of multiple incoming edges connected to the successor node can be effectively integrated, avoiding the one-sidedness of a single incoming edge feature. In the air quality monitoring scenario, a monitoring object (successor node) may be affected by multiple different predecessor nodes, and the edge features of each incoming edge reflect the impact characteristics of different pollution sources on the monitoring point. By averaging, the role of each pollutant information can be comprehensively considered, and the information such as pollution transmission intensity and path characteristics carried by multiple incoming edges can be integrated to form a more representative aggregated edge feature, which fully reflects the comprehensive impact on the successor node. The aggregated edge features and the node attributes of the successor node are input into the second perceptron to generate node features, which further improves the integrity and accuracy of the node features.

[0135] Optionally, based on the graph structure features, an operation of air quality prediction related to air pollutants is performed on a target object among multiple monitoring objects, including: determining a first target node corresponding to the target object in the graph structure data and determining a second target node corresponding to the target object in the graph structure features; determining the pollutant information corresponding to the target object at the previous moment; and determining the pollutant information corresponding to the target object at the specified moment using a pre-set GRU-based prediction model based on the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, and the node features of the second target node.

[0136] Specifically, Figure 7 A schematic diagram of the recurrent neural network GRU is further shown, referring to Figure 7 As shown in Figure 2, the recurrent neural network GRU includes input units, GRU models, and MLP linear layers. S 1 is described as an example of a target object.

[0137] Therefore, after the deep learning model uses the graph neural network to obtain the graph structure features corresponding to the graph structure data according to the above operations, for example, it can determine the target object in the graph structure data. S 1 corresponds to the first target node N 1. And determine the target object in the graph structure features S 1 corresponds to the second target node FN 1.

[0138] The deep learning model can then S 1At a specified time tThe last moment t -1 O3 concentration, first target node N Node attribute information of 1 an 1 and the second target node FN Node features of 1 fn 1 is input to the input unit.

[0139] Thus, the GRU neural network can be used to S 1At a specified time t The last moment t -1 O3 concentration, first target node N Node attribute information of 1 an 1 and the second target node FN Node features of 1 fn 1. Determine your target audience S 1At a specified time t O3 concentration (pollutant information).

[0140] According to the embodiment of the present disclosure, accurate positioning of a specific monitoring object is achieved by determining the target node corresponding to the target object in the graph structure features. In a complex air quality monitoring network, there are multiple monitoring objects. Accurately locking the target node can avoid interference from irrelevant information and focus subsequent analysis on key objects. By determining the pollutant information of the target node at the previous moment and combining the node attributes with the node characteristics, a deep fusion of time dimension (historical pollutant concentration) and space dimension information is achieved. Air quality changes have spatiotemporal dynamic characteristics. Historical pollutant concentrations reflect the trend of pollution evolution. Node attributes reflect the characteristics of the target object itself, and node characteristics incorporate the influence of the surrounding environment. Compared with the prediction method that only considers single-dimensional information, it can more realistically simulate the pollution diffusion process and significantly improve the accuracy and reliability of the prediction.

[0141] Optionally, based on the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, and the node characteristics of the second target node, a pre-set GRU-based prediction model is used to determine the pollutant information corresponding to the target object at the specified moment, including: determining the first hidden state information corresponding to the target node at the previous moment, wherein the first hidden state information is generated by the GRU model of the prediction model; inputting the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, the node characteristics of the second target node, and the first hidden state information into the GRU model to determine the second hidden state information corresponding to the target node at the specified moment; and inputting the second hidden state information into the linear layer to determine the pollutant information corresponding to the target object at the specified moment.

[0142] Specifically, refer to Figure 7As shown, the input unit also receives the GRU model at the previous moment t -1 calculated hidden state H t-1 (ie, the first hidden state information), and the target object S 1At a specified time t The last moment t -1 O3 concentration, first target node N Node attribute information of 1 an 1 and the second target node FN Node features of 1 fn 1 and hidden state H t-1 Input to the GRU model, so that the GRU model calculates the time according to the input information t The corresponding hidden state H t (ie the second hidden state information).

[0143] Hidden State H t It is input into the MLP linear layer of the recurrent neural network GRU to predict the target object S 1O3 concentration at a specified time (i.e. pollutant information).

[0144] According to the embodiment of the present disclosure, the first hidden state information generated by the GRU model is obtained. This information condenses multi-dimensional information such as the pollutant change trend of the target node in the past, the impact of the surrounding environment, etc. It is input into the GRU model together with the pollutant information, node attributes and node characteristics of the current node, so that the GRU model can deeply remember the historical pollution evolution process and dynamically update the hidden state (i.e., the second hidden state information) according to the current input information. Avoid prediction bias caused by ignoring historical trends. Air quality data has significant nonlinear dynamic change characteristics, and the GRU model shows unique advantages in processing this type of data. The GRU model can fully explore the complex laws of the target node pollutant information in the time series.

[0145] Optionally, when the target object is a city, the method also includes: obtaining pollutant information corresponding to the air monitoring stations around the city as the monitoring object at a specified time; averaging the pollutant information of the air monitoring stations to obtain average pollutant information; and weighted summing the predicted pollutant information and the average pollutant information to determine the pollutant information of the city at the specified time.

[0146] Specifically, refer to Figure 8As shown, when the target object is a city, after the air quality forecast module 135 predicts the predicted O3 concentration (predicted pollutant information) of the city at a specified time according to the above method, the O3 concentration of the stations around the city at the specified time is obtained, wherein the O3 concentration of the station at the specified time can also be predicted by the above method.

[0147] The air quality forecast module 135 then averages the O3 concentrations at each station at a specified time to obtain an average O3 concentration (ie, average pollutant information).

[0148] Then, linear regression can be used to determine the final pollutant information of the city at a specified time:

[0149] y=a*x1+b*x2

[0150] x1 and x2 are the average O3 concentration calculated by the station mean and the O3 concentration directly predicted by the city, respectively, and y is the city forecast concentration value obtained by the final regression calculation.

[0151] Thus, through the above method, this application simultaneously forecasts the air quality of the site and the city, and performs a secondary forecast for the city through linear regression to improve the accuracy of the city forecast.

[0152] This embodiment not only considers pollutant emissions but also fully utilizes backward trajectory data through methods such as backward trajectory point ratios. This comprehensively covers multiple factors influencing air quality, avoiding the limitations of focusing solely on single factors such as meteorological data and geographic information. This allows the model to more realistically reflect the actual state of air quality changes. The directed graph structure and its edge attribute information accurately describe the circulation and propagation paths of air pollutants between monitored objects, quantifying the distribution ratio of pollutants along different paths. This helps accurately determine the actual impact of different regions on the air quality of the target area, thereby improving prediction accuracy.

[0153] Take the hourly O3 concentration forecast for national stations from 00:00 to 23:00 on April 1, 2025 as an example:

[0154] (1) Data acquisition unit 110

[0155] The air quality data download module 111 downloads hourly O3 data from national stations from 00:00 on December 31, 2019, to 23:00 on March 31, 2025, denoted as Data A. The atmospheric reanalysis data download module 112 downloads ERA5 data from 00:00 on January 1, 2020, to 23:00 on March 31, 2025, denoted as Data B. The backward trajectory data download module 113 downloads GDAS1 data from 01:00 on December 31, 2019, to 23:00 on March 31, 2025, denoted as Data C; and downloads Hysplit-GFS weather forecast data from 01:00 on March 31, 2025, to 23:00 on April 1, denoted as Data D. Assuming a 24-hour backward propagation period, the trajectory to be propagated at 00:00 on April 1 requires 24 data points from 01:00 on March 31 to 00:00 on April 1. The weather forecast data download module 114 downloads the GFS weather data from 00:00 to 23:00 on April 1, 2025, which is recorded as data E.

[0156] (2) Model training unit 120

[0157] Simulation period: 00:00 on January 1, 2020 - 23:00 on March 31, 2025.

[0158] The label data and pollutant historical information extracted by the first air quality data processing module 122 are both O3 concentration data. The label data is Data A; the pollutant historical information is the preceding 24 hours of data, namely, O3 hourly data from 00:00 on December 31, 2019, to 23:00 on March 30, 2025. The reason for the 24-hour lead time is to account for the influence of the node on the same time period of the previous day. The pollutant historical information is the first node attribute information of the node.

[0159] The first meteorological data processing module 121 extracts meteorological reanalysis data on the site. Assuming there are 10 factors, the 2nd to 11th node attribute information of the node are the 10 factors.

[0160] The first backward trajectory data processing module 123 inputs data C to simulate the backward trajectory of the station every hour for 24 hours, calculates the proportion of trajectory points on the edges within the trajectory circle, and obtains the weight feature of the edge associated with each station as the first edge attribute information of the directed edge.

[0161] The first graph structure generation module 124 first connects nodes to form edges based on inter-site distance (less than 200 km) and inter-site altitude (less than 1000 km), forming an adjacency matrix. The inter-site distance corresponding to the node is used as the second edge attribute of the edge. Based on the gridded emission inventory, the corresponding emission amount of the station is extracted as the 12th feature of the node. The emission inventory here uses the MEIC inventory. Direct emission statistics exist for the other five factors except O3, which is replaced by VOCs data (given that national control stations are mainly distributed in urban areas, mainly VOCs control areas). The year, month, week, and hour are also added as the 13th to 16th node attributes. Therefore, the above graph structure consists of the adjacency matrix, nodes (16 features), and edges (2 features).

[0162] The deep learning model training module 125 divides the graph data into a training set (January 1, 2020, to December 31, 2023), a validation set (January 1, 2024, to December 31, 2024), and a test set (January 1, 2025, to March 31, 2025). The training set is used to learn model parameters, the validation set is used for hyperparameter tuning, and the test set is used to evaluate model generalization. Hyperparameters in the validation set are externally defined (in a configuration file) and do not require model learning. They can be set based on empirical experience or determined after several experimental simulations. Therefore, we will focus on simulations of the training set. For example, for the prediction of node A at 01:0 ... The measured O3 concentration at node A at time 01 is used as a loss function with the predicted result, and then backpropagation is performed to update the weight parameters to finally obtain the trained model parameters.

[0163] (3) Forecast unit 130

[0164] Forecast period: 00:00, April 1, 2025 - 23:00, April 1, 2025. The node characteristics are consistent with those of the training unit.

[0165] The second air quality data processing module 132 uses the data extracted from the training unit from 00:00 on March 31, 2025 to 23:00 on March 31, 2025, as the historical O3 concentration information of the site during the forecast period, which is also the first node attribute information of the node. The second meteorological data processing module 131 extracts the meteorological forecast data gfs on the site, and the 10 factors consistent with the aforementioned factors, so the second to 11 node attribute information of the node are these 10 factors. The second backward trajectory data processing module 133 inputs data D to simulate the 24-hour backward trajectory of the site every hour, and calculates the proportion of trajectory points on the edges within the trajectory circle to obtain the weight information of the incoming edge associated with each site as the first edge attribute information of the incoming edge.

[0166] The second graph structure generation module 134 first connects nodes to form edges based on inter-station distance (less than 200 km) and inter-station altitude (less than 1000 km), forming an adjacency matrix. The inter-station distance is used as the second feature of the edge. Based on the gridded emissions inventory, the emissions corresponding to the station are extracted as the 12th feature of the node. The year, month, week, and hour are also added as the 13th to 16th features of the node. (These training units have already been completed and do not need to be repeated.) Thus, the above graph structure consists of the adjacency matrix, nodes (16 features), and edges (2 features).

[0167] Air quality forecast module 135 (still taking the forecast of node A at 01:00 as an example): The O3 concentration of node A at 00:00 is used as one input information, and the O3 concentration of node B connected to node A at 00:00 is used as another input information. Then, it is spliced ​​with the 01:00 node attribute information of nodes A and B themselves and the 01:00 edge attribute information of the edge between A and B. Then, it is input into the three-layer perceptron for the edge. The parameters are provided by the training unit. The 01:00 edge feature of the edge between node A and node B with node A as the target is obtained. According to the above method, the 01:00 edge feature of the incoming edge connecting other nodes to node A is obtained. Then, the 01:00 edge feature of the edge is aggregated (averaged) and input into the one-layer perceptron for the node. The parameters are provided by the training unit. Then, the updated 01:00 node feature of node A is obtained. The O3 concentration of node A at time 00, the node attribute information of node A at time 01, the updated node features of node A at time 01, and the hidden state at time 00 are input into the GRU model. The parameters are provided by the training unit to obtain the hidden state at time 01. Then, it is output through a linear layer with parameters provided by the training unit to obtain the prediction result.

[0168] Take the hourly O3 concentration forecast for Wuxi, a national city, from 00:00 to 23:00 on April 1, 2025, as an example:

[0169] In this system, both site and city forecasts are treated as node forecasts. A single system is used to construct separate graph structures for training and forecasting. The following example uses cities nationwide as nodes to construct a graph structure, obtaining forecast results for Wuxi. Alternatively, regions encompassing Wuxi, such as Jiangsu Province and the Yangtze River Delta, can be selected for this purpose, taking into account factors such as transmission and computational costs.

[0170] (1) Data acquisition unit 110

[0171] The time period is the same as that of the aforementioned sites. Air quality data needs to be downloaded at the national city level. The rest of the data is the same and can be used universally.

[0172] (2) Model training unit 120

[0173] Simulation period: the same as the above-mentioned site.

[0174] The first air quality data processing module 122 is the same as the above, except that it extracts data from national stations.

[0175] The first meteorological data processing module 121 extracts the meteorological data within the grid where the longitude and latitude of the station are located, while the city extracts the mean value within the shp area.

[0176] The first backward trajectory data processing module 123 is the same as the above, except that the city serves as the receiving point of the backward trajectory.

[0177] The first graph structure generation module 124 similarly connects nodes to form edges based on inter-site distance (less than 200 km) and inter-city altitude (less than 1000 km), forming an adjacency matrix. The inter-node distance is used as the second feature of the edge. Based on the gridded emissions inventory, the average emissions within the city's shp area are extracted as the 12th node attribute information. The emissions inventory here uses the MEIC inventory. Direct emission statistics exist for the other five factors except O3, which is replaced by VOCs data (urban areas are primarily VOCs control zones). The year, month, week, and hour are also added as the 13th to 16th node attributes. Thus, the graph structure described above consists of the adjacency matrix, nodes (16 node attributes), and edges (2 node attributes).

[0178] Deep learning model training module 125, forecast for Wuxi at 01:00: The O3 concentration of Wuxi at 00:00 is used as one input information, and the O3 concentration of Suzhou at 00:00, which is connected to Wuxi, is used as another input information. Then, the O3 concentration is concatenated with the node attribute information of Wuxi and Suzhou at 01:00, and the edge attribute information of the edge between Wuxi and Suzhou at 01:00 is input into the three-layer perceptron of the edge to obtain the edge features of Wuxi at 01:00 for the edge between Wuxi and Suzhou with Wuxi as the target. The edge features of other cities connected to Wuxi at 01:00 are obtained according to the above method. The edge features of the edges at 01:00 are aggregated (averaged) and input into the one-layer perceptron of the Wuxi node to obtain the updated node features of Wuxi at 01:00. The O3 concentration in Wuxi at 00:00, the node attribute information of Wuxi at 01:00, the updated node features of Wuxi at 01:00, and the hidden state at 00:00 are input into the GRU to obtain the hidden state at 01:00. This is then output through a linear layer to obtain the predicted result. The loss function is applied to the measured O3 concentration in Wuxi at 01:00 and the predicted result, and backpropagation is used to update the weight parameters to obtain the trained model parameters.

[0179] (3) Daily Forecast Unit 130

[0180] Forecast period: Same as the above-mentioned site.

[0181] The second air quality data processing module 132 is the same as the above, except that it extracts data from national stations.

[0182] The second meteorological data processing module 131 extracts the gfs meteorological data within the grid where the longitude and latitude of the station are located, while the city extracts the mean value within the shp area.

[0183] The second backward trajectory data processing module 133 is the same as the above, except that the city serves as the receiving point of the backward trajectory.

[0184] The second graph structure generation module is composed of the adjacency matrix, city node features (16 features), and edges (2 features) in the previous training unit to form the graph structure.

[0185] The air quality forecast module 135 takes the forecast of Wuxi at 01:00 as an example: the O3 concentration of Wuxi at 00:00 is used as one input information, and the O3 concentration of Suzhou at 00:00, which is connected to Wuxi, is used as another input information. Then, the O3 concentration is spliced ​​with the 01:00 node attribute information of Wuxi and Suzhou themselves, and the 01:00 edge attribute information of the edge between Wuxi and Suzhou. Then, the information is input into the three-layer perceptron for the edge. The parameters are provided by the training unit 110. The 01:00 edge features of the edge between Wuxi and Suzhou with Wuxi as the target are obtained. The 01:00 edge features of the edges connecting other cities to Wuxi are obtained according to the above method. Then, the 01:00 edge features of the edges are aggregated (averaged) and input into the one-layer perceptron of the Wuxi node. The parameters are provided by the training unit to obtain the updated 01:00 node features of Wuxi. The O3 concentration in Wuxi at 00:00, the node attribute information of Wuxi at 01:00, the updated node features of Wuxi at 01:00, and the hidden state at 00:00 are input into the GRU model. The parameters are provided by the training unit to obtain the hidden state at 01:00. Then, the model is output through a linear layer with the parameters provided by the training unit to obtain the direct prediction result of Wuxi at 01:00.

[0186] (4) Regression calculation

[0187] Further calculations were made for the forecast of Wuxi City. The O3 forecast results from the nine national monitoring stations in Wuxi City from 00:00 on April 1, 2025 to 23:00 on April 1, 2025, were obtained in Case 1. The hourly average was calculated to obtain the hourly concentration value of Wuxi City (the concentration value of Wuxi City at 00:00 was equal to the average concentration value of the nine stations at 00:00). This was then regressed with the concentration value of Wuxi City directly predicted in Case 2, namely:

[0188] y=a*x1+b*x2,

[0189] Where a and b are weights (obtained based on data fitting, and their values ​​will change dynamically as data continues to increase), x1 and x2 are the city concentrations calculated based on the site mean and the city’s direct forecast concentration, respectively, and y is the city’s forecast concentration value obtained by the final regression calculation.

[0190] In addition, reference Figure 1 As shown, according to a third aspect of this embodiment, a storage medium is provided, wherein the storage medium includes a stored program, wherein when the program is run, a processor executes any one of the above methods.

[0191] Therefore, in this embodiment, by associating multiple monitoring objects with a directed graph structure, each monitoring object corresponds to a node, and directed edges are established based on the circulation of air pollutants between the monitoring objects, clearly showing the propagation relationship of pollutants between the various monitoring objects. The pollutant emission information of the monitoring objects is used as the node attribute of the node, so that the model can take into account the situation of pollution emission information and more comprehensively reflect the influencing factors of air quality. By calculating the proportion of trajectory points of at least one incoming edge pointing to the same successor node to determine the edge attribute information, it helps to quantify the distribution of pollutants on different transmission paths, thereby more accurately describing the source and transmission process of pollutants, and being able to fully reflect the relationship between monitoring objects and the relevant characteristics of pollutants. Based on the extracted graph structure features, the air quality of the target object is predicted, and the influence of multiple factors on air quality is comprehensively considered, thereby achieving accurate prediction of air quality. This solves the technical problem of inaccurate air quality prediction in the prior art.

[0192] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0193] Through the description of the above embodiments, those skilled in the art will clearly understand that the methods according to the above embodiments can be implemented using software plus the necessary general-purpose hardware platform. Of course, hardware can also be used, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, or optical disk) and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0194] Example 2

[0195] Figure 9 FIG. 1 shows an air quality prediction device 900 according to this embodiment, which corresponds to the method according to the first aspect of embodiment 1. Figure 9As shown, the device 900 includes: a directed graph structure construction module 910, which is used to construct a directed graph structure associated with multiple monitoring objects, wherein each of the multiple monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects, and the directed edges of the directed graph structure are established according to the circulation of air pollutants between the monitoring objects; a node attribute determination module 920, which is used to determine the node attribute information of each node in the directed graph structure, wherein the node attribute information at least includes the pollutant emission information of the monitoring object corresponding to the corresponding node; a trajectory point ratio determination module 930, which is used to determine at least one entry point pointing to the same successor node in the directed graph structure. The trajectory point ratio of the edge, and determine the edge attribute information of at least one incoming edge according to the trajectory point ratio, wherein the trajectory point ratio is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of the first monitoring object corresponding to the same successor node between at least one incoming edge; a graph structure data construction module 940, used to construct graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; a message passing module 950, used to input the graph structure data into the graph neural network model for message passing to obtain corresponding graph structure features; and an air quality prediction module 960, used to perform air quality prediction related to air pollutants for target objects among multiple monitoring objects based on the graph structure features.

[0196] Therefore, in this embodiment, by associating multiple monitoring objects with a directed graph structure, each monitoring object corresponds to a node, and directed edges are established based on the circulation of air pollutants between the monitoring objects, clearly showing the propagation relationship of pollutants between the various monitoring objects. The pollutant emission information of the monitoring objects is used as the node attribute of the node, so that the model can take into account the situation of pollution emission information and more comprehensively reflect the influencing factors of air quality. By calculating the proportion of trajectory points of at least one incoming edge pointing to the same successor node to determine the edge attribute information, it helps to quantify the distribution of pollutants on different transmission paths, thereby more accurately describing the source and transmission process of pollutants, and being able to fully reflect the relationship between monitoring objects and the relevant characteristics of pollutants. Based on the extracted graph structure features, the air quality of the target object is predicted, and the influence of multiple factors on air quality is comprehensively considered, thereby achieving accurate prediction of air quality. This solves the technical problem of inaccurate air quality prediction in the prior art.

[0197] Example 3

[0198] Figure 10 FIG1 shows an air quality prediction device 1000 according to the first aspect of this embodiment, which corresponds to the method according to the first aspect of embodiment 1. Figure 10As shown, the device 1000 includes: a processor 1010; and a memory 1020, connected to the processor 1010, for providing the processor 1010 with instructions for processing the following processing steps: constructing a directed graph structure associated with multiple monitoring objects, wherein each of the multiple monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure correspond to different monitoring objects, and the directed edges of the directed graph structure are established according to the circulation of air pollutants between the monitoring objects; in the directed graph structure, determining node attribute information of each node, wherein the node attribute information at least includes pollutant emission information of the monitoring object corresponding to the corresponding node; in the Determine the proportion of trajectory points of at least one incoming edge pointing to the same successor node in a directed graph structure, and determine the edge attribute information of the at least one incoming edge based on the trajectory point proportion, wherein the trajectory point proportion is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of the first monitoring object corresponding to the same successor node among the at least one incoming edge; construct graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; input the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and based on the graph structure features, perform air quality prediction related to the air pollutants for the target object among the multiple monitoring objects.

[0199] In the embodiment of the present disclosure, by associating multiple monitoring objects with a directed graph structure, each monitoring object corresponds to a node, and directed edges are established based on the circulation of air pollutants between the monitoring objects, clearly showing the propagation relationship of pollutants between the various monitoring objects, and using the pollutant emission information of the monitoring objects as the node attributes of the nodes, the model can take into account the situation of pollution emission information and more comprehensively reflect the influencing factors of air quality. By calculating the proportion of trajectory points of at least one incoming edge pointing to the same successor node, the edge attribute information is determined, which helps to quantify the distribution of pollutants on different transmission paths, thereby more accurately describing the source and transmission process of pollutants, and can comprehensively reflect the relationship between monitoring objects and the relevant characteristics of pollutants. Based on the extracted graph structure features, the air quality of the target object is predicted, and the impact of multiple factors on air quality is comprehensively considered, thereby achieving accurate prediction of air quality. This solves the technical problem of inaccurate air quality prediction in the prior art.

[0200] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0201] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0202] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0203] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0204] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0205] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), a mobile hard drive, a magnetic disk, or an optical disk.

[0206] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An air quality prediction method, characterized in that: include: Constructing a directed graph structure associated with a plurality of monitoring objects, wherein each of the plurality of monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure respectively correspond to different monitoring objects, and the directed edges of the directed graph structure are established based on the mobility of air pollutants between the monitoring objects; In the directed graph structure, determining node attribute information of each node, wherein the node attribute information at least includes pollutant emission information of the monitoring object corresponding to the corresponding node; Determining, in the directed graph structure, a trajectory point ratio of each incoming edge pointing to the same successor node, and determining edge attribute information of each incoming edge based on the trajectory point ratio, wherein the trajectory point ratio is used to indicate a distribution ratio of multiple backward trajectory points of air pollutants of a first monitoring object corresponding to the same successor node among the respective incoming edges, wherein the multiple backward trajectory points are determined by simulating backward propagation trajectories of air pollutants corresponding to the monitoring object using a hysplit model; Constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; Inputting the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and Based on the graph structure features, air quality prediction related to the air pollutants is performed on the target object among the multiple monitoring objects, and wherein, The operation of determining the proportion of trajectory points of each incoming edge pointing to the same successor node in the directed graph structure includes: determining, based on a specified time, the plurality of backward trajectory points corresponding to the air pollutants of the first monitoring object; Determining a farthest backward trajectory point, which is farthest from the first monitored object, among the plurality of backward trajectory points, and determining a second distance between the first monitored object and the farthest backward trajectory point; Determine an adjacent predecessor node among the predecessor nodes of the successor node, wherein a distance between a second monitoring object corresponding to the adjacent predecessor node and the first monitoring object is less than the second distance, and determine a first incoming edge corresponding to the adjacent predecessor node among the incoming edges; and Setting the trajectory point ratio of the second incoming edges other than the first incoming edge among the incoming edges to zero, and determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points, wherein the operation of determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points includes: Determine the connecting lines between each second monitored object and the first monitored object; determine, from the first input edges, the input edge corresponding to the connecting line with the smallest distance between the backward trajectory points as the input edge corresponding to the backward trajectory point; and determine the corresponding trajectory point ratio based on the number of backward trajectory points corresponding to the first input edge.

2. The method according to claim 1, characterized in that The operations for constructing a directed graph structure associated with multiple monitoring objects include: When a first distance between two monitored objects is less than a first threshold and there is no terrain with an altitude greater than a second threshold between the two monitored objects, a directed edge is constructed between the nodes corresponding to the two monitored objects.

3. The method according to claim 1, characterized in that The operation of determining the node attribute information of each node includes: obtaining the pollutant history information related to the air pollutant of the monitored object at the historical moment before the specified moment and the meteorological information of the monitored object at the specified moment; determining the pollutant emission information of the monitored object corresponding to the specified moment; and determining the node attribute information of the node corresponding to the monitored object based on the pollutant history information, the pollutant emission information and the meteorological information, and wherein In a case where the monitored object is an air monitoring station, the operation of determining the pollutant emission information of the monitored object corresponding to the specified time includes: determining multiple geographical grids corresponding to the air monitoring station, wherein the geographical grids corresponding to the air monitoring station are located within the area covered by the multiple geographical grids; and averaging the pollutant emission information corresponding to the multiple geographical grids at the specified time as the pollutant emission information corresponding to the air monitoring station at the specified time. Alternatively, in a case where the monitored object is a city, the operation of determining the pollutant emission information corresponding to the monitored object at the specified time includes: Determine a plurality of geographic grids corresponding to the area of ​​the city; and average the pollutant emission information corresponding to the plurality of geographic grids at the designated time as the pollutant emission information corresponding to the city at the designated time.

4. The method according to claim 3, characterized in that The operation of determining the node attribute information of the node corresponding to the monitoring object according to the pollutant historical information, the pollutant emission information, and the meteorological information includes: The pollutant history information, the pollutant emission information, and the meteorological information are spliced ​​with the time information corresponding to the designated moment as node attribute information of the node corresponding to the monitoring object.

5. The method according to claim 1, wherein The operation of determining the edge attribute information of each input edge according to the trajectory point ratio includes: Determine the distance information between the monitoring objects corresponding to the predecessor node and the successor node of each incoming edge; and determine the edge attribute information of each incoming edge according to the distance information and the trajectory point ratio.

6. The method according to claim 1, characterized in that Inputting the graph structure data into the graph neural network model for message transmission to obtain the corresponding graph structure features includes: Performing message passing on the directed edges of the graph structure data to determine edge features corresponding to the directed edges; and According to the node attributes of the successor node in the graph structure data and the edge characteristics of the incoming edge corresponding to the successor node, the message is passed to the successor node to determine the node characteristics of the successor node, and wherein The operations of performing message passing on the directed edges of the graph structure data and determining the edge features corresponding to the directed edges include: determining an associated successor node and an associated predecessor node associated with the directed edge; determining the pollutant information corresponding to the associated successor node at the previous moment before a specified moment, and the pollutant information corresponding to the associated predecessor node at the previous moment; and inputting the node attribute information of the associated successor node, the pollutant information corresponding to the associated successor node at the previous moment, the node attribute information of the associated predecessor node, the pollutant information corresponding to the associated predecessor node at the previous moment, and the edge attribute information of the directed edge into a pre-trained first linear perceptron to generate edge features corresponding to the directed edge, and wherein According to the node attributes of the successor node in the graph structure data and the edge features of the incoming edges corresponding to the successor node, the message transmission is performed on the successor node to determine the node features of the successor node, including: averaging the edge features of the incoming edges corresponding to the successor node to obtain an aggregated edge feature; and inputting the aggregated edge feature and the node attributes of the successor node into a second perceptron to generate a node feature corresponding to the successor node, and wherein Based on the graph structure features, an operation of air quality prediction related to the air pollutants is performed on the target object among the multiple monitored objects, including: determining a first target node corresponding to the target object in the graph structure data and determining a second target node corresponding to the target object in the graph structure features; determining the pollutant information corresponding to the target object at the previous moment; and determining the pollutant information corresponding to the target object at the specified moment using a pre-set GRU-based prediction model based on the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, and the node characteristics of the second target node.

7. The method according to claim 6, characterized in that The operation of determining the pollutant information corresponding to the target object at the specified moment by using a preset GRU-based prediction model based on the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, and the node feature of the second target node includes: Determining first hidden state information of the target node corresponding to the previous moment, wherein the first hidden state information is generated by a GRU model of the prediction model; Inputting the pollutant information corresponding to the target object at the previous moment, the node attribute information of the first target node, the node feature of the second target node, and the first hidden state information into the GRU model to determine the second hidden state information corresponding to the target node at the specified moment; and The second hidden state information is input into the linear layer of the prediction model to determine the predicted pollutant information corresponding to the target object at the specified time.

8. The method according to claim 7, characterized in that When the target object is a city, it also includes: Obtaining pollutant information corresponding to the air monitoring stations around the city as monitoring objects at the specified time; averaging the pollutant information of the air monitoring sites to obtain average pollutant information; and The predicted pollutant information and the average pollutant information are weightedly summed to determine the pollutant information of the city at the specified time.

9. An air command prediction device, characterized in that: include: a directed graph structure construction module, configured to construct a directed graph structure associated with a plurality of monitoring objects, wherein each of the plurality of monitoring objects is configured to monitor air pollutants, the nodes of the directed graph structure respectively corresponding to different monitoring objects, and the directed edges of the directed graph structure are established based on the mobility of air pollutants between the monitoring objects; a node attribute determination module, configured to determine node attribute information of each node in the directed graph structure, wherein the node attribute information at least includes pollutant emission information of a monitoring object corresponding to the corresponding node; A trajectory point ratio determination module is used to determine the trajectory point ratio of at least one incoming edge pointing to the same successor node in the directed graph structure, and determine the edge attribute information of the at least one incoming edge based on the trajectory point ratio, wherein the trajectory point ratio is used to indicate the distribution ratio of multiple backward trajectory points of air pollutants of the first monitoring object corresponding to the same successor node among the at least one incoming edge, wherein the multiple backward trajectory points are determined by using a hysplit model to simulate the backward propagation trajectory of air pollutants corresponding to the monitoring object, and wherein the operation of determining the trajectory point ratio of each incoming edge pointing to the same successor node in the directed graph structure includes: determining the multiple backward trajectory points corresponding to the air pollutants of the first monitoring object based on a specified time; determining the farthest backward trajectory point with the farthest distance from the first monitoring object among the multiple backward trajectory points, and determining the distance between the first monitoring object and the farthest backward trajectory point. a second distance between trajectory points; determining an adjacent predecessor node in the predecessor node of the successor node, wherein the distance between the second monitoring object corresponding to the adjacent predecessor node and the first monitoring object is less than the second distance, and determining a first incoming edge corresponding to the adjacent predecessor node in each incoming edge; and setting the trajectory point ratio of the second incoming edges other than the first incoming edge in each incoming edge to zero, and determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points, and wherein the operation of determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points includes: determining the connection between each second monitoring object and the first monitoring object; from the first incoming edges, determining the incoming edge corresponding to the connection with the smallest distance between the backward trajectory point and the backward trajectory point as the incoming edge corresponding to the backward trajectory point; and determining the corresponding trajectory point ratio according to the number of backward trajectory points corresponding to the first incoming edge; A graph structure data construction module, configured to construct graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; A message passing module, configured to input the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and An air quality prediction module is used to predict the air quality related to the air pollutants for a target object among the multiple monitored objects based on the graph structure features.

10. An air quality prediction device, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Constructing a directed graph structure associated with a plurality of monitoring objects, wherein each of the plurality of monitoring objects is used to monitor air pollutants, the nodes of the directed graph structure respectively correspond to different monitoring objects, and the directed edges of the directed graph structure are established based on the mobility of air pollutants between the monitoring objects; In the directed graph structure, determining node attribute information of each node, wherein the node attribute information at least includes pollutant emission information of the monitoring object corresponding to the corresponding node; Determining, in the directed graph structure, a trajectory point ratio of at least one incoming edge pointing to the same successor node, and determining edge attribute information of the at least one incoming edge based on the trajectory point ratio, wherein the trajectory point ratio is used to indicate a distribution ratio of multiple backward trajectory points of air pollutants of a first monitoring object corresponding to the same successor node among the at least one incoming edge, wherein the multiple backward trajectory points are determined by simulating backward propagation trajectories of air pollutants corresponding to the monitoring object using a hysplit model; Constructing graph structure data corresponding to the directed graph structure based on the node attribute information and the edge attribute information; Inputting the graph structure data into a graph neural network model for message passing to obtain corresponding graph structure features; and Based on the graph structure features, air quality prediction related to the air pollutants is performed on the target object among the multiple monitoring objects, and wherein, The operation of determining the proportion of trajectory points of each incoming edge pointing to the same successor node in the directed graph structure includes: determining, based on a specified time, the plurality of backward trajectory points corresponding to the air pollutants of the first monitoring object; Determining a farthest backward trajectory point, which is farthest from the first monitored object, among the plurality of backward trajectory points, and determining a second distance between the first monitored object and the farthest backward trajectory point; Determine an adjacent predecessor node among the predecessor nodes of the successor node, wherein a distance between a second monitoring object corresponding to the adjacent predecessor node and the first monitoring object is less than the second distance, and determine a first incoming edge corresponding to the adjacent predecessor node among the incoming edges; and Setting the trajectory point ratio of the second incoming edges other than the first incoming edge among the incoming edges to zero, and determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points, wherein the operation of determining the trajectory point ratio corresponding to the first incoming edge based on the multiple backward trajectory points includes: Determine the connecting lines between each second monitored object and the first monitored object; determine, from the first input edges, the input edge corresponding to the connecting line with the smallest distance between the backward trajectory points as the input edge corresponding to the backward trajectory point; and determine the corresponding trajectory point ratio based on the number of backward trajectory points corresponding to the first input edge.

Images