A PM2.5 concentration prediction method, system, terminal device and medium

Abstract

The application is suitable for the field of spatiotemporal big data mining technology, and provides a PM2.5 concentration prediction method, system, terminal device and medium. The method obtains PM2.5 concentration data, meteorological data, land use data and DEM data; calculates the three-dimensional terrain distance between grids, and constructs a PM2.5 spatial transmission graph; sequentially constructs a static covariant matrix, a PM2.5 concentration matrix and a dynamic covariant matrix; embeds the constructed characteristic attribute variable into the PM2.5 spatial transmission graph to obtain a high-dimensional feature vector; performs time sequence embedding on the high-dimensional feature vector to obtain a dense vector; constructs an initial PM2.5 concentration prediction model, and trains the model by using the dense vector and a pre-set precision loss function; and uses the trained PM2.5 concentration prediction model to predict the PM2.5 concentration. The application can improve the accuracy of PM2.5 concentration prediction.

Description

Technical Field

[0001] This application belongs to the field of spatiotemporal big data mining technology, and in particular relates to a PM2.5 concentration prediction method, system, terminal equipment and medium. Background Technology

[0002] In recent years, with the continuous improvement of urbanization and industrialization levels, traffic emissions, industrial emissions, and energy emissions have been increasing, leading to increasingly serious air pollution. PM2.5 (Particulate Matter 2.5, referring to particulate matter with a diameter of 2.5 micrometers or less) is the most significant air pollutant, seriously threatening people's green travel, livable environment, and physical and mental health. Therefore, timely and accurate prediction of PM2.5 concentrations is of great early warning significance for the rational planning of industrial production, daily travel plans, and large-scale air pollution prevention and control.

[0003] Existing PM2.5 prediction methods are mainly divided into two categories: model-driven and data-driven.

[0004] Model-driven methods primarily model the physicochemical mechanisms and evolution of PM2.5 formation, transport, and deposition using numerical equations or simulations. They predict future PM2.5 concentrations by solving simultaneous equations or through simulation. While these methods typically produce reliable prediction trends, their accuracy heavily relies on complex meteorological parameters, such as atmospheric boundaries, which are difficult to obtain directly. Furthermore, at high spatiotemporal resolutions, these methods are computationally complex and extremely slow, making it difficult to guarantee the stability of prediction accuracy.

[0005] Data-driven approaches primarily use models such as statistical regression, machine learning, or deep learning to fit the mapping relationship between historical PM2.5 observation concentrations and themselves or other variables, and then use the learned mapping relationship to predict future PM2.5 concentrations. However, considering that the spatiotemporal processes of PM2.5 generation, transport, and diffusion in the real world are influenced by the nonlinear combination of multiple variables, such as topography, land use type, and meteorological factors (wind speed, rainfall, temperature, air pressure, etc.), classical statistical regression methods often struggle to capture such complex nonlinear spatiotemporal relationships, resulting in poor fitting performance. Machine learning methods utilize strategies such as support vector machines and regression trees to improve the fitting accuracy of regression learning; however, these methods are essentially still extreme learning of linear relationships and cannot model the spatiotemporal propagation process between multiple variables and PM2.5. With the rapid development and integration of neural network science and computer science, deep learning methods have achieved good results in capturing complex nonlinear relationships between variables. Among them, temporal deep convolutional networks such as Long Short-Term Memory (LSTM) can effectively capture the temporal dependencies of variables from historical time series, while spatial deep convolutional networks such as Graph Convolutional Networks (GCN) can effectively capture the spatial dependencies between variables from observation data between sites. However, existing deep learning methods compress the real world into a two-dimensional planar space, using only planar distances such as Euclidean distance to construct spatial convolutional networks, ignoring the actual situation of PM2.5 emission, propagation, and dissipation in three-dimensional space. In addition, most deep learning models only model the coupling relationship between dynamic meteorological elements such as wind speed, rainfall, temperature, and air pressure and PM2.5. In fact, static geographical elements such as land use type and dynamic meteorological elements simultaneously affect the spatiotemporal processes of PM2.5 generation, transmission, and diffusion.

[0006] In summary, existing deep learning methods have a strong ability to capture multivariate nonlinear spatiotemporal dependencies. However, existing deep learning-based PM2.5 concentration prediction methods do not fully consider the transmission and diffusion of PM2.5 in three-dimensional space, thus severing the spatiotemporal coupling between static geographical elements and dynamic meteorological elements. Summary of the Invention

[0007] This application provides a PM2.5 concentration prediction method, system, terminal equipment, and medium, which can solve the problem of low accuracy in current PM2.5 concentration prediction.

[0008] In a first aspect, this application provides a method for predicting PM2.5 concentration, comprising:

[0009] Historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data were obtained for each air quality monitoring station in the study area.

[0010] The three-dimensional topographic distance between adjacent grids in the DEM data is calculated, and a PM2.5 spatial transport map corresponding to the study area is constructed based on the three-dimensional topographic distance. The three-dimensional topographic distance represents the Euclidean distance between adjacent grids in the DEM data after correction for elevation error. The PM2.5 spatial transport map includes multiple nodes, with each node corresponding to an air quality monitoring station, and each air quality monitoring station corresponding to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station.

[0011] Based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, static covariate matrices, PM2.5 concentration matrices, and dynamic covariate matrices were constructed sequentially. The static covariate matrix represents the static characteristic attributes of each monitoring station during the study period, including the area of ​​four land use types: cultivated land, forest land, water bodies, and construction land. The dynamic covariate matrix represents the dynamic characteristic attributes of each monitoring station during the study period, including observed data on temperature, air pressure, wind speed, and rainfall.

[0012] Based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, characteristic attribute variables are constructed and embedded into the PM2.5 spatial transport map to obtain high-dimensional feature vectors. Among them, there is a one-to-one correspondence between the characteristic attribute variables and the air quality monitoring stations. The characteristic attribute variables are used to describe the characteristic attributes of the air quality monitoring stations, and the high-dimensional feature vectors are used to describe the deep spatial dependencies of each node in the PM2.5 spatial transport map.

[0013] Temporal embedding is performed on the high-dimensional feature vectors to obtain dense vectors; the dense vectors are used to describe the deep spatial dependencies and continuous temporal dependencies of each node in the PM2.5 spatial transport graph.

[0014] An initial PM2.5 concentration prediction model was constructed, and the PM2.5 concentration prediction model was trained using dense vectors and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model.

[0015] Using the trained PM2.5 concentration prediction model, the PM2.5 concentration at each air quality monitoring station in the study area was predicted.

[0016] Optionally, the three-dimensional topographic distance between adjacent grids in the DEM data is calculated, and based on the three-dimensional topographic distance, a PM2.5 spatial transport map corresponding to the study area is constructed, including:

[0017] Through calculation formula

[0018]

[0019] Get grid R i,j With grid R w,v Three-dimensional terrain distance between Among them, grid R i,j With grid R w,v For adjacent grids in DEM data, Represents grid R w,v The position of the projected coordinate system. Represents grid R i,j The position of the projected coordinate system is given by c, where c represents the correction factor for the elevation distance relative to the planar distance, and c∈[1,10]. Represents grid R i,j With grid R w,v The elevation difference between them Represents grid R w,v DEM elevation, Represents grid R i,j DEM elevation; grid R i,j With grid R w,v The subscript values ​​are: 0≤i,w<number of grid rows, 0≤j,v<number of grid columns;

[0020] Calculate the minimum 3D terrain distance between each grid and other grids. Use this minimum 3D terrain distance as the PM2.5 transmission distance between grids. Calculate the corresponding reciprocal normalized value of the PM2.5 transmission distance. Treat each grid as node V and the reciprocal normalized value as the edge length E between nodes, thus obtaining the PM2.5 spatial transmission map G = {V, E}, where e i,j ∈E represents the inverse normalized value of the distance from the i-th air quality monitoring station to the j-th air quality monitoring station.

[0021] Optionally, based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix are constructed sequentially, including:

[0022] From historical PM2.5 concentration data, PM2.5 concentration data from all air quality monitoring stations for a historical period are extracted, and the PM2.5 concentration data is reconstructed into a one-dimensional column vector X. t ∈R n*1 We obtain the PM2.5 concentration matrix, where n represents the number of air quality monitoring stations;

[0023] Land use type area data for each air quality monitoring station was extracted from historical land use data, and a four-dimensional column vector was constructed based on the land use type area data to obtain the static covariate matrix S.t ∈R n*4 Land use types include arable land, forest land, water bodies, and buildings;

[0024] From historical meteorological data, temperature, air pressure, wind speed, and rainfall for the first m hours of a historical time period t are extracted for each air quality monitoring station to obtain a dynamic covariate matrix. in, These represent the temperature, air pressure, wind speed, and rainfall observation data from all air quality monitoring stations during the historical period t.

[0025] Optionally, feature attribute variables are constructed based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, and these feature attribute variables are embedded into the PM2.5 spatial transport map to obtain a high-dimensional feature vector, including:

[0026] Based on the preset timestamps, the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix are reconstructed into feature attribute variables H0 = [X]. t S t D t ], thus obtaining the PM2.5 spatial transport map G = (V, E, H0); where X t S represents the PM2.5 concentration matrix corresponding to timestamp t. t D represents the static covariate matrix corresponding to the timestamp t. t This represents the dynamic covariate matrix corresponding to the timestamp t;

[0027] Through calculation formula

[0028]

[0029]

[0030] Obtain the high-dimensional feature vector H k Among them, H k H represents the output of the k-th layer graph convolutional network, k∈[1,3]. When k=1, H k-1 =H0, σ(·) represents the activation function, H k-1 W represents the output of the (k-1)th layer graph convolutional network. k-1 This represents the weight vector to be learned.

[0031] Optionally, temporal embedding is performed on the high-dimensional feature vectors to obtain dense vectors, including:

[0032] Through calculation formula

[0033]

[0034]

[0035]

[0036]

[0037] Obtain the dense vector at historical time t Where tanh(·) represents the activation function, U, W, W (z) U (z) W (r) U (r) Both represent the parameter matrix to be learned, H t This indicates that the graph embedding vector Hk = 3 at historical time t. This represents the embedding vector at historical time t-1, when t-1 = 0. It is a zero-valued vector matrix. This represents the initial embedding vector at historical time t. Indicates r t and Hadama multiplication between them (1-z) t )and Hadama multiplication between them Indicate z t and Hadama multiplication between them The initial embedding vector representing historical time t Dense vectors of historical time t-1 polymerization.

[0038] Optionally, the initial PM2.5 concentration prediction model includes a two-layer decoder;

[0039] The initial expression for the PM2.5 concentration prediction model is as follows:

[0040]

[0041] in, This represents the predicted PM2.5 concentration at time 'at', where W1 and W2 represent the learnable parameters of the first and second layer decoders, respectively, and b1 and b2 both represent bias terms. This represents the dense vector at time at.

[0042] Optionally, the expression for the accuracy loss function is as follows:

[0043]

[0044] Among them, y mThis represents the actual PM2.5 concentration value at the m-th air quality monitoring station, obtained from historical PM2.5 concentration data. Let Y represent the predicted PM2.5 concentration at the m-th air quality monitoring station, where m = 1, 2, ..., Num, and Num represents the total number of air quality monitoring stations in the study area. Y represents the matrix of actual PM2.5 concentration values. This represents the PM2.5 concentration prediction matrix, where λ represents the hyperparameter, and L... reg This indicates the bias term.

[0045] Secondly, this application provides a PM2.5 concentration prediction system, comprising:

[0046] The data acquisition module is used to acquire historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data for each air quality monitoring station within the study area.

[0047] The spatial transport map construction module is used to calculate the three-dimensional terrain distance between adjacent grids in the DEM data, and construct the PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance. The three-dimensional terrain distance represents the Euclidean distance between adjacent grids in the DEM data after correction for elevation error. The PM2.5 spatial transport map includes multiple nodes, with each node corresponding to an air quality monitoring station, and each air quality monitoring station corresponding to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station.

[0048] The static and dynamic covariate construction module is used to construct static covariate matrices, PM2.5 concentration matrices, and dynamic covariate matrices sequentially based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, respectively.

[0049] The spatial feature extraction module is used to construct feature attribute variables based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, and embed the feature attribute variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector. There is a one-to-one correspondence between the feature attribute variables and the air quality monitoring stations. The feature attribute variables are used to describe the feature attributes of the air quality monitoring stations, and the high-dimensional feature vector is used to describe the deep spatial dependencies of each node in the PM2.5 spatial transport map.

[0050] The temporal feature extraction module is used to perform temporal embedding on high-dimensional feature vectors to obtain dense vectors; the dense vectors are used to describe the deep spatial dependencies and continuous temporal dependencies of each node in the PM2.5 spatial transport graph.

[0051] The prediction model training module is used to build an initial PM2.5 concentration prediction model and train the PM2.5 concentration prediction model using dense vectors and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model.

[0052] The prediction module is used to predict the PM2.5 concentration at each air quality monitoring station in the study area using the trained PM2.5 concentration prediction model.

[0053] Thirdly, this application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the PM2.5 concentration prediction method described above.

[0054] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described PM2.5 concentration prediction method.

[0055] The above-mentioned solution in this application has the following beneficial effects:

[0056] The PM2.5 concentration prediction method provided in this application corrects the Euclidean distance between adjacent grids using elevation errors to obtain the three-dimensional topographic distance between adjacent grids. This distance is then used to construct a PM2.5 spatial transport map for the study area, considering the transport and diffusion process of PM2.5 in three-dimensional space. This addresses the limitation of traditional PM2.5 concentration prediction methods that only consider two-dimensional planar Euclidean distances, thus improving the accuracy of PM2.5 concentration prediction. Based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, static covariate matrices, PM2.5 concentration matrices, and dynamic covariate matrices are constructed sequentially. Feature attribute variables are then constructed based on these matrices, and PM2.5 concentration is predicted based on these feature attribute variables. This fully considers the coupling effect of static geographical factors and dynamic meteorological data, further improving the accuracy of PM2.5 concentration prediction. By embedding feature attribute variables into the PM2.5 spatial transport map and temporally embedding high-dimensional feature vectors, the method comprehensively considers the temporal dependence of PM2.5 concentration on dynamic meteorological factors and its spatial dependence on static geographical factors, further improving the accuracy of PM2.5 concentration prediction.

[0057] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart illustrating a PM2.5 concentration prediction method provided in an embodiment of this application;

[0060] Figure 2 This is a schematic diagram of the structure of a PM2.5 concentration prediction model provided in an embodiment of this application;

[0061] Figure 3 A comparison chart showing the effectiveness of a PM2.5 concentration prediction method provided in one embodiment of this application;

[0062] Figure 4 This is a schematic diagram of the structure of a PM2.5 concentration prediction system provided in an embodiment of this application;

[0063] Figure 5 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0064] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0065] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0066] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0067] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0068] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0069] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0070] To address the issue of low accuracy in current PM2.5 concentration prediction, this application provides a PM2.5 concentration prediction method, system, terminal equipment, and medium. The method utilizes elevation errors to correct the Euclidean distance between adjacent grids, obtaining the three-dimensional topographic distance between them. This distance is then used to construct a PM2.5 spatial transport map for the study area, considering the transport and diffusion process of PM2.5 in three-dimensional space. This overcomes the limitation of traditional PM2.5 concentration prediction methods that only consider two-dimensional plane Euclidean distance, thus improving the accuracy of PM2.5 concentration prediction. The method is based on historical land use data and historical PM2.5 concentration data. Based on historical meteorological data, a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix are constructed sequentially. Feature attribute variables are then constructed based on these matrices, and PM2.5 concentration is predicted based on these feature attribute variables. This fully considers the coupling effect of static geographical factors and dynamic meteorological data, which is beneficial to improving the accuracy of PM2.5 concentration prediction. By embedding the feature attribute variables into the PM2.5 spatial transport map and temporally embedding the high-dimensional feature vectors, the temporal dependence of PM2.5 concentration on dynamic meteorological factors and its spatial dependence on static geographical factors are comprehensively considered, thus improving the accuracy of PM2.5 concentration prediction.

[0071] The PM2.5 concentration prediction method provided in this application is illustrated below.

[0072] like Figure 1 As shown, this PM2.5 concentration prediction method specifically includes the following steps:

[0073] Step 11: Obtain historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data for each air quality monitoring station in the study area.

[0074] For example, in one embodiment of this application, the Yangtze River Delta region of China is used as the study area, and historical PM2.5 concentration data, historical meteorological data, historical land use data, and digital elevation model (DEM) data recorded by various air quality monitoring stations in the region from March 1, 2017 to February 28, 2018 are used. The temporal resolution (data recording interval) of the historical PM2.5 concentration data is 1 hour.

[0075] Considering the possibility of accidental errors in data collection, the data needs to be preprocessed after it is obtained to ensure its validity and accuracy, as shown in steps I to IV.

[0076] Step 1: Cleaning historical PM2.5 concentration data.

[0077] Specifically, the historical PM2.5 concentration data of each air quality monitoring station were transformed into a normal distribution. Then, the Rayleigh Criterion was used to remove PM2.5 concentration values ​​that were less than μ-3σ or greater than μ+3σ. It should be noted that, as discussed later, μ here represents the mean of the normal distribution, and σ represents the average difference between each data point and the mean.

[0078] In addition, to address the potential missing PM2.5 concentration data for certain historical moments, it is necessary to imputate the missing values ​​in the historical PM2.5 concentration data. Specifically, the missing values ​​are repaired using the average PM2.5 concentration of the three nearest neighboring data points before and after the missing value.

[0079] Step II: Cleaning historical meteorological data.

[0080] In the embodiments of this application, the aforementioned historical meteorological data includes temperature, air pressure, wind speed, and rainfall observation data recorded by air quality monitoring stations over past time periods.

[0081] Specifically, based on the available range of meteorological data, impossible meteorological observation values ​​are deleted. For missing data values, refer to the process in step I. Furthermore, for cases where the air quality monitoring station and the meteorological monitoring station are not the same station, inverse distance weighted interpolation is used to calculate the temperature, air pressure, wind speed, and rainfall observation data for each air quality monitoring station; for cases where the air quality monitoring station and the meteorological monitoring station are the same station, the temperature, air pressure, wind speed, and rainfall observation data from the same station are directly used.

[0082] Step III: Cleaning historical land use data.

[0083] Specifically, a buffer zone with a radius of 10 kilometers (KM) is generated with the air quality monitoring station as the center, and the area of ​​land use types within the buffer zone is calculated. In one embodiment of this application, the aforementioned land use types include cultivated land, forest land, water bodies, and construction land.

[0084] Step IV: Multi-source data normalization.

[0085] To prevent the influence of different dimensions among multi-source data on PM2.5 concentration prediction, a calculation formula is used.

[0086]

[0087] The data cleaned in steps I-III above are normalized, where y represents the result after normalization, and x represents the data before normalization. min x represents the maximum value of the data. min This represents the minimum value of the data.

[0088] Step 12: Calculate the three-dimensional terrain distance between adjacent grids in the DEM data, and construct the PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance.

[0089] Among them, the three-dimensional terrain distance represents the Euclidean distance between adjacent grids in the DEM data after elevation error correction. The PM2.5 spatial transport map includes multiple nodes, each corresponding to an air quality monitoring station. Each air quality monitoring station corresponds to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station.

[0090] Step 13: Construct a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix in sequence based on historical land use data, historical PM2.5 concentration data, and historical meteorological data.

[0091] The static covariate matrix represents the static characteristic attributes of each monitoring station during the study period, including the area of ​​four land use types: cultivated land, forest land, water bodies, and construction land.

[0092] The dynamic covariate matrix represents the dynamic characteristics of each monitoring station during the study period. It includes observational data for temperature, air pressure, wind speed, and rainfall.

[0093] Step 14: Based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, construct feature attribute variables and embed the feature attribute variables into the PM2.5 spatial transport map to obtain high-dimensional feature vectors.

[0094] Among them, there is a one-to-one correspondence between the feature attribute variables and the air quality monitoring stations. The feature attribute variables are used to describe the feature attributes of the air quality monitoring stations, and the high-dimensional feature vectors are used to describe the deep spatial dependencies of each node in the PM2.5 spatial transport map.

[0095] Step 15: Perform temporal embedding on the high-dimensional feature vector to obtain a dense vector.

[0096] Dense vectors are used to describe the deep spatial dependencies and continuous temporal dependencies of each node in the PM2.5 spatial transport graph.

[0097] Specifically, through calculation formula

[0098]

[0099]

[0100]

[0101]

[0102] Obtain the dense vector at historical time t Where tanh(·) represents the activation function, U, W, W (z) U (z) W (r) U (r) Both represent the parameter matrix to be learned, H t This indicates that the graph embedding vector Hk = 3 at historical time t. This represents the embedding vector at historical time t-1, when t-1 = 0. It is a zero-valued vector matrix. This represents the initial embedding vector at historical time t. Indicates r t and Hadama multiplication between them (1-z) t)and Hadama multiplication between them Indicates zt and Hadama multiplication between them The initial embedding vector representing historical time t Dense vectors of historical time t-1 polymerization.

[0103] Step 16: Construct an initial PM2.5 concentration prediction model, and train the PM2.5 concentration prediction model using dense vectors and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model.

[0104] The aforementioned initial PM2.5 concentration prediction model includes two decoder layers, specifically... Figure 2 As shown.

[0105] The expression corresponding to the initial PM2.5 concentration prediction model is as follows:

[0106]

[0107] in, This represents the predicted PM2.5 concentration at time 'at', where W1 and W2 represent the learnable parameters of the first and second layer decoders, respectively, and b1 and b2 both represent bias terms. This represents the dense vector at time at.

[0108] The expression for the above accuracy loss function is as follows:

[0109]

[0110] Among them, y m This represents the actual PM2.5 concentration value at the m-th air quality monitoring station, obtained from historical PM2.5 concentration data. Let Y represent the predicted PM2.5 concentration at the m-th air quality monitoring station, where m = 1, 2, ..., Num, and Num represents the total number of air quality monitoring stations in the study area. Y represents the matrix of actual PM2.5 concentration values. This represents the PM2.5 concentration prediction matrix, where λ represents the hyperparameter, and L... reg This indicates the bias term.

[0111] Specifically, using the accuracy loss value as the criterion, the Adaptive Moment Estimator (Adam) is used to update each parameter to be learned (details are explained in each step) until the accuracy loss value is less than or equal to the preset accuracy loss threshold. The corresponding values ​​of each parameter to be learned at this time are retained to obtain the trained PM2.5 concentration prediction model.

[0112] Step 17: Using the trained PM2.5 concentration prediction model, predict the PM2.5 concentration at each air quality monitoring station in the study area.

[0113] For example, real-time PM2.5 concentration data, real-time meteorological data, real-time land use data, and DEM data are collected from air quality monitoring stations. By performing steps 11-15 above, dense vectors of air quality monitoring stations are obtained, and these vectors are input into the trained PM2.5 concentration prediction model to obtain the predicted PM2.5 concentration values ​​of the air quality monitoring stations.

[0114] It should be noted that after obtaining the predicted PM2.5 concentration, this predicted value can provide guidance and early warning for industrial production, daily travel, and air pollution prevention and control within the study area.

[0115] The following is an illustrative description of the process in step 12, which involves calculating the three-dimensional terrain distance between adjacent grids in the DEM data and constructing a PM2.5 spatial transport map of the study area based on this distance, including steps 12.1 to 12.2:

[0116] Step 12.1, using the calculation formula

[0117]

[0118] Get grid R i，j With grid R w，v Three-dimensional terrain distance between

[0119] Among them, grid R i，j With grid R w，v For adjacent grids in DEM data, Represents grid R w，v The position of the projected coordinate system. Represents grid R i，j The position of the projected coordinate system is given by c, where c represents the correction factor for the elevation distance relative to the planar distance, and c∈[1,10]. Represents grid R i，j With grid R w，v The elevation difference between them Represents grid R w，v DEM elevation, Represents grid R i，j DEM elevation, grid R i，j With grid R w，v The subscript values ​​are: 0≤i, w<number of grid rows, 0≤j, v<number of grid columns.

[0120] Step 12.2: Calculate the minimum three-dimensional terrain distance between each grid and other grids. Use the minimum three-dimensional terrain distance as the PM2.5 transmission distance between grids. Calculate the reciprocal normalized value corresponding to the PM2.5 transmission distance. Use the grid as a node and the reciprocal normalized value as the side length between the node and other nodes to obtain the PM2.5 spatial transmission map.

[0121] For example, using the reciprocal normalized value as the edge length between nodes includes: calculating the 3D terrain distance between all neighboring grids according to step 12.1, and then, for any station s i Grid R si Starting from point S, any other station s j Grid R sj As the endpoint, based on the 3D terrain distance between neighboring grids, a greedy algorithm is used to solve for grid R. si To grid R sj The shortest three-dimensional terrain distance is used as the station s i To station s j PM 2.5 Transmission distance; repeat the above process until PM between all stations is completed. 2.5 Transmission distance calculation; finally, the reciprocal normalized value of the PM2.5 spatial transmission distance between stations is taken as the side length E between nodes.

[0122] For example, the PM2.5 spatial transport map can be formally expressed as G = {V, E}, where V represents a node in the PM2.5 spatial transport map, which corresponds one-to-one with an air quality monitoring station, and E represents the side length in the PM2.5 spatial transport map.

[0123] The following is an illustrative example of the process in step 13, which involves constructing the static covariate matrix, PM2.5 concentration matrix, and dynamic covariate matrix sequentially based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, including steps 13.1 to 13.3:

[0124] Step 13.1: Extract land use type area data for each air quality monitoring station from historical land use data, and construct a four-dimensional column vector based on the land use type area data to obtain a static covariate matrix.

[0125] For example, the above four-dimensional column vector can be formally represented as S t ∈R n*4 .

[0126] Step 13.2: Extract PM2.5 concentration data from all air quality monitoring stations for a historical period from historical PM2.5 concentration data, and reconstruct the PM2.5 concentration data into a one-dimensional column vector to obtain the PM2.5 concentration matrix.

[0127] For example, the above one-dimensional column vector can be formally represented as X t ∈R n*1 .

[0128] Step 13.3: Extract the temperature, air pressure, wind speed, and rainfall data for the first m hours of a historical time period t from historical meteorological data for each air quality monitoring station, and obtain the dynamic covariate matrix. in, These represent the temperature, air pressure, wind speed, and rainfall observation data from all air quality monitoring stations during the historical period t.

[0129] For example, the above dynamic covariate matrix can be formally represented as: in, Let represent the temperature, air pressure, wind speed, and rainfall observation data of all air quality monitoring stations at hour t, respectively.

[0130] The following is an illustrative example of step 14, which involves constructing characteristic attribute variables based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, and embedding these characteristic attribute variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector. This includes steps 14.1 to 14.2:

[0131] Step 14.1: Reconstruct the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix into feature attribute variables H0 = [X] according to the preset timestamps. t S t D t ], thus obtaining the PM2.5 spatial transport map G=(V,E,H0); .

[0132] Among them, X t S represents the PM2.5 concentration matrix corresponding to timestamp t. t D represents the static covariate matrix corresponding to the timestamp t. t This represents the dynamic covariate matrix corresponding to the timestamp t.

[0133] Step 14.2, using the calculation formula

[0134]

[0135]

[0136] The high-dimensional feature vector Hk is obtained. It is an embedding vector matrix, where each row vector represents the embedding vector of each monitoring station.

[0137] Among them, H k H represents the output of the k-th layer graph convolutional network, k∈[1,3]. When k=1, H k-1 =H0, σ(·) represents the activation function, H k-1 W represents the output of the (k-1)th layer graph convolutional network. k-1 This represents the weight vector to be learned.

[0138] Figure 3 In one embodiment of this application, a comparison chart between the predicted PM2.5 concentration and the actual PM2.5 concentration obtained after executing the PM2.5 concentration prediction method provided in this application is shown. It is easy to see that the PM2.5 concentration prediction method provided in this application has excellent accuracy, further verifying the effectiveness of the PM2.5 concentration prediction method provided in this application.

[0139] As can be seen from the above steps: The PM2.5 concentration prediction method provided in this application uses elevation error to correct the Euclidean distance between adjacent grids, obtaining the three-dimensional terrain distance between adjacent grids, and constructs a PM2.5 spatial transport map corresponding to the study area. This considers the transport and diffusion process of PM2.5 in three-dimensional space, solving the one-sidedness of traditional PM2.5 concentration prediction methods that only consider two-dimensional plane Euclidean distance, thus improving the accuracy of PM2.5 concentration prediction. Based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix are constructed sequentially. Feature attribute variables are constructed based on the above matrices, and PM2.5 concentration is predicted based on the feature attribute variables. This fully considers the coupling effect of static geographical factors and dynamic meteorological data, which is conducive to improving the accuracy of PM2.5 concentration prediction. By embedding the feature attribute variables into the PM2.5 spatial transport map and performing temporal embedding on the high-dimensional feature vector, the temporal dependence of PM2.5 concentration on dynamic meteorological factors and the spatial dependence on static geographical factors are comprehensively considered, improving the accuracy of PM2.5 concentration prediction.

[0140] The PM2.5 concentration prediction system provided in this application is described below as an example.

[0141] like Figure 4 As shown, the PM2.5 concentration prediction system 400 includes:

[0142] The data acquisition module 401 is used to acquire historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data for each air quality monitoring station in the study area;

[0143] The spatial transport map construction module 402 is used to calculate the three-dimensional terrain distance between adjacent grids in the DEM data, and construct the PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance. The three-dimensional terrain distance represents the Euclidean distance between adjacent grids in the DEM data after correction for elevation error. The PM2.5 spatial transport map includes multiple nodes, each corresponding to an air quality monitoring station. Each air quality monitoring station corresponds to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station.

[0144] The static and dynamic covariate construction module 403 is used to construct a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix sequentially based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, respectively.

[0145] The spatial feature extraction module 404 is used to construct feature attribute variables based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, and embed the feature attribute variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector. Among them, there is a one-to-one correspondence between the feature attribute variables and the air quality monitoring stations. The feature attribute variables are used to describe the feature attributes of the air quality monitoring stations, and the high-dimensional feature vector is used to describe the deep spatial dependencies of each node in the PM2.5 spatial transport map.

[0146] The temporal feature extraction module 405 is used to perform temporal embedding on high-dimensional feature vectors to obtain dense vectors; the dense vectors are used to describe the deep spatial dependencies and continuous temporal dependencies of each node in the PM2.5 spatial transport graph.

[0147] The prediction model training module 406 is used to construct an initial PM2.5 concentration prediction model and train the PM2.5 concentration prediction model using dense vectors and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model.

[0148] The prediction module 407 is used to predict the PM2.5 concentration at each air quality monitoring station in the study area using the trained PM2.5 concentration prediction model.

[0149] It should be noted that the information interaction and execution process between the above modules / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0151] like Figure 5 As shown, embodiments of this application provide a terminal device, such as... Figure 5 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 5 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0152] Specifically, when the processor D100 executes the computer program D102, it acquires historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data from each air quality monitoring station within the study area. It calculates the three-dimensional terrain distance between adjacent grids in the DEM data and constructs a PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance. It then constructs a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix sequentially based on the historical land use data, historical PM2.5 concentration data, and historical meteorological data. Based on the PM2.5 concentration matrix, static covariate matrix, and dynamic covariate matrix, it constructs feature attribute variables and embeds these variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector. This high-dimensional feature vector is then time-series embedded to obtain a dense vector, which is used to construct an initial PM2.5 concentration prediction model. The PM2.5 concentration prediction model is trained using the dense vector and a pre-set accuracy loss function to obtain a trained PM2.5 concentration prediction model. Finally, the trained PM2.5 concentration prediction model is used to predict the PM2.5 concentration at each air quality monitoring station within the study area. Specifically, this method corrects the Euclidean distance between adjacent grids using elevation errors to obtain the three-dimensional topographic distance between them. This distance is then used to construct a PM2.5 spatial transport map for the study area, considering the three-dimensional transport and diffusion process of PM2.5. This addresses the limitation of traditional PM2.5 concentration prediction methods that only consider two-dimensional Euclidean distances, thus improving the accuracy of PM2.5 concentration prediction. Based on historical land use data, historical PM2.5 concentration data, and historical meteorological data, static covariate matrices, PM2.5 concentration matrices, and dynamic covariate matrices are constructed sequentially. Feature attribute variables are then constructed based on these matrices, and PM2.5 concentration is predicted based on these variable features. This fully considers the coupling effect of static geographical factors and dynamic meteorological data, further improving the accuracy of PM2.5 concentration prediction. By embedding feature attribute variables into the PM2.5 spatial transport map and temporally embedding high-dimensional feature vectors, the temporal dependence of PM2.5 concentration on dynamic meteorological factors and its spatial dependence on static geographical factors are comprehensively considered, further improving the accuracy of PM2.5 concentration prediction.

[0153] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0154] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0155] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0156] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0157] 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, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of one-dimensional source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a PM2.5 concentration prediction system / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, such as a USB flash drive, a portable hard drive, a magnetic disk, or an optical disk.

[0158] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0159] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0160] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0161] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0162] The PM2.5 concentration prediction method provided in this application has the following advantages:

[0163] (1) Compared with the model-driven method, which is simple and controllable in computation and solution, it provides a prediction framework that takes into account both dynamic factors such as meteorological factors and static factors such as land use on the spatiotemporal nonlinear dependence of PM2.5 evolution, and is suitable for practical applications of predicting PM2.5 concentration by coordinating dynamic and static covariates with fine spatiotemporal resolution.

[0164] (2) Compared with classic data-driven methods such as statistical regression and machine learning, which can fully take into account the complex nonlinear dependencies between multivariate variables and their independent variables, the PM2.5 concentration prediction method provided in this application innovatively utilizes three-dimensional spatial transmission distance to model the PM2.5 transmission accessibility constraint between air quality observation stations, while coupling the driving force of static factors such as land use on PM2.5 generation and the biochemical transmission effect of dynamic factors such as meteorological factors on PM2.5, and develops a PM2.5 concentration prediction method based on geographic spatiotemporal map convolution, which has stronger interpretability and scalability.

[0165] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting PM2.5 concentration, characterized in that, include: Historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data were obtained for each air quality monitoring station in the study area. The three-dimensional terrain distance between adjacent grids in the DEM data is calculated, and a PM2.5 spatial transport map corresponding to the study area is constructed based on the three-dimensional terrain distance. The three-dimensional terrain distance represents the Euclidean distance between adjacent grids in the DEM data after elevation error correction. The PM2.5 spatial transport map includes multiple nodes, each corresponding to an air quality monitoring station. Each air quality monitoring station corresponds to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station. The calculation of the three-dimensional terrain distance between adjacent grids in the DEM data, and the construction of a PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance, includes: Through calculation formula Get the grid With grid Three-dimensional terrain distance between Among them, grid With grid For adjacent grids in the DEM data, Represents a grid The position of the projected coordinate system. Represents a grid The position of the projected coordinate system. This represents the correction factor for elevation distance relative to horizontal distance. , Represents a grid With grid The elevation difference between them , Represents a grid DEM elevation, Represents a grid DEM elevation, grid With grid The subscript values ​​are: 0≤ <Number of grid rows, 0≤ <number of grid columns; Calculate the minimum 3D terrain distance between each grid and other grids, and use this minimum 3D terrain distance as the PM2.5 transmission distance between that grid and other grids. Calculate the reciprocal normalized value corresponding to this PM2.5 transmission distance, and treat that grid as a node. The reciprocal normalized value is used as the edge length between the node and other nodes. The PM2.5 spatial transport map was obtained. ,in Indicates the first i The number of air quality monitoring stations to the first j The inverse normalized value of the distance between each air quality monitoring station; Based on the historical land use data, the historical PM2.5 concentration data, and the historical meteorological data, a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix are constructed sequentially. The static covariate matrix represents the static characteristic attributes of each monitoring station during the study period, and the dynamic covariate matrix represents the dynamic characteristic attributes of each monitoring station during the study period. The process involves constructing a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix sequentially based on the historical land use data, the historical PM2.5 concentration data, and the historical meteorological data, including: From the historical PM2.5 concentration data, extract the PM2.5 concentration data of all air quality monitoring stations for a historical period t, and reconstruct the PM2.5 concentration data into a one-dimensional column vector. The PM2.5 concentration matrix was obtained, where n Indicates the number of air quality monitoring stations; Land use type area data for each air quality monitoring station is extracted from the historical land use data, and a four-dimensional column vector is constructed based on the land use type area data to obtain the static covariate matrix. The land use types include arable land, forest land, water bodies, and buildings. From the historical meteorological data, the temperature, air pressure, wind speed, and rainfall of each air quality monitoring station in the preceding m hours of a historical time period t are extracted to obtain the dynamic covariate matrix. ,in, , , , These represent the temperature, air pressure, wind speed, and rainfall observation data from all air quality monitoring stations during the historical time period t. Based on the PM2.5 concentration matrix, the static covariate matrix, and the dynamic covariate matrix, feature attribute variables are constructed, and these feature attribute variables are embedded into the PM2.5 spatial transport map to obtain a high-dimensional feature vector. The feature attribute variables correspond one-to-one with the air quality monitoring stations, and the feature attribute variables are used to describe the characteristic attributes of the air quality monitoring stations. The high-dimensional feature vector is used to describe the deep spatial dependencies corresponding to each node in the PM2.5 spatial transport map. Temporal embedding is performed on the high-dimensional feature vector to obtain a dense vector; the dense vector is used to describe the deep spatial dependencies and continuous temporal dependencies corresponding to each node in the PM2.5 spatial transmission graph; An initial PM2.5 concentration prediction model is constructed, and the PM2.5 concentration prediction model is trained using the dense vector and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model. Using the trained PM2.5 concentration prediction model, the PM2.5 concentration at each air quality monitoring station in the study area is predicted.

2. The PM2.5 concentration prediction method according to claim 1, characterized in that, The step involves constructing feature attribute variables based on the PM2.5 concentration matrix, the static covariate matrix, and the dynamic covariate matrix, and embedding these feature attribute variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector, including: According to the preset timestamps, the PM2.5 concentration matrix, the static covariate matrix, and the dynamic covariate matrix are reconstructed into feature attribute variables. This leads to the PM2.5 spatial transport map. ;in, Represents timestamp The corresponding PM2.5 concentration matrix, Represents timestamp The corresponding static covariate matrix, Represents timestamp The corresponding dynamic covariate matrix; Through calculation formula The high-dimensional feature vector is obtained. ;in, Indicates the first The output of the layered graph convolutional network ,when hour, , This represents the activation function. Indicates the first The output of the layered graph convolutional network This represents the weight vector to be learned. express The degree matrix.

3. The PM2.5 concentration prediction method according to claim 2, characterized in that, The step of temporally embedding the high-dimensional feature vector to obtain a dense vector includes: Through calculation formula At a historical moment Dense vectors ;in, This represents the activation function. Each represents a parameter matrix to be learned. Indicating a historical moment Graph embedding vectors , Indicating a historical moment The embedding vector, when hour, It is a zero-valued vector matrix. Representing historical moments Initial embedding vector, express and Hadama multiplication between them express and Hadama multiplication between them express and Hadama multiplication between them Representing historical moments Initial embedding vector and historical moments Dense vectors polymerization.

4. The PM2.5 concentration prediction method according to claim 1, characterized in that, The initial PM2.5 concentration prediction model includes two layers of decoders; The expression for the initial PM2.5 concentration prediction model is as follows: in, Indicates at time The predicted PM2.5 concentration Let represent the learnable parameters of the first-layer decoder and the learnable parameters of the second-layer decoder, respectively. Both represent bias terms. Indicates at time Dense vectors.

5. The PM2.5 concentration prediction method according to claim 1, characterized in that, The expression for the accuracy loss function is as follows: in, This indicates the number of times the historical PM2.5 concentration data was obtained. The actual PM2.5 concentration values ​​of each air quality monitoring station. Indicates the first Predicted PM2.5 concentration values ​​from several air quality monitoring stations. , This indicates the total number of air quality monitoring stations within the study area. This represents a matrix of true PM2.5 concentration values. This represents a matrix of predicted PM2.5 concentrations. Indicates hyperparameters, This indicates the bias term.

6. A PM2.5 concentration prediction system, used to execute the PM2.5 concentration prediction method according to any one of claims 1 to 5, characterized in that, include: The data acquisition module is used to acquire historical PM2.5 concentration data, historical meteorological data, historical land use data, and DEM data for each air quality monitoring station within the study area. A spatial transport map construction module is used to calculate the three-dimensional terrain distance between adjacent grids in the DEM data, and construct a PM2.5 spatial transport map corresponding to the study area based on the three-dimensional terrain distance. The three-dimensional terrain distance represents the Euclidean distance between adjacent grids in the DEM data after elevation error correction. The PM2.5 spatial transport map includes multiple nodes, each corresponding to an air quality monitoring station. Each air quality monitoring station corresponds to a grid in the DEM data. The PM2.5 spatial transport map is used to describe the PM2.5 transport between each air quality monitoring station. The static and dynamic covariate construction module is used to construct a static covariate matrix, a PM2.5 concentration matrix, and a dynamic covariate matrix sequentially based on the historical land use data, the historical PM2.5 concentration data, and the historical meteorological data, respectively; wherein, the static covariate matrix represents the static characteristic attributes of each monitoring station during the study period, and the dynamic covariate matrix represents the dynamic characteristic attributes of each monitoring station during the study period; The spatial feature extraction module is used to construct feature attribute variables based on the PM2.5 concentration matrix, the static covariate matrix, and the dynamic covariate matrix, and embed the feature attribute variables into the PM2.5 spatial transport map to obtain a high-dimensional feature vector; wherein, there is a one-to-one correspondence between the feature attribute variables and the air quality monitoring stations, the feature attribute variables are used to describe the feature attributes of the air quality monitoring stations, and the high-dimensional feature vector is used to describe the deep spatial dependencies corresponding to each node in the PM2.5 spatial transport map; The temporal feature extraction module is used to perform temporal embedding on the high-dimensional feature vector to obtain a dense vector; the dense vector is used to describe the deep spatial dependencies and continuous temporal dependencies corresponding to each node in the PM2.5 spatial transmission graph. The prediction model training module is used to construct an initial PM2.5 concentration prediction model and train the PM2.5 concentration prediction model using the dense vector and a pre-set accuracy loss function to obtain the trained PM2.5 concentration prediction model. The prediction module is used to predict the PM2.5 concentration at each air quality monitoring station in the study area using the trained PM2.5 concentration prediction model.

7. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the PM2.5 concentration prediction method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the PM2.5 concentration prediction method as described in any one of claims 1 to 5.

Images