Flow field data diffusion calculation method, device, equipment and medium

Abstract

The invention discloses a flow field data diffusion calculation method, device, equipment and medium, and relates to the technical field of diffusion monitoring, according to the method, a GNN model is used for replacing CFD to conduct diffusion calculation on pollutant data in an unstructured grid area, the calculation speed can be increased in a multiplied mode, and the pollutant diffusion result of an unstructured grid can be accurately obtained. Under the condition that a training data set is large enough, calculation under a new regional grid is effective, modeling calculation does not need to be conducted again through CFD, and the whole data processing and calculation period is shortened.

Description

Technical Field

[0001] The present invention relates to the field of diffusion monitoring technology, and in particular to a flow field data diffusion calculation method, device, equipment and medium based on graph neural network. Background Art

[0002] Flow field data diffusion refers to the process of changes in physical quantities (such as velocity, temperature, etc.) in the flow field in space and time. In fluid mechanics, flow field data diffusion usually involves two main mechanisms: convection and diffusion.

[0003] CFD model is the core component of computational fluid dynamics (CFD) simulation. CFD model numerically solves the differential equations that control fluid flow, obtains the discrete distribution of the flow field of fluid flow in a continuous area, and approximates the fluid flow. The basic principles of CFD model include model establishment, numerical algorithm and result visualization.

[0004] The patent document with application number 202211562736.9 records a method for analyzing the diffusion of atmospheric pollution caused by substation fires under the influence of urban environment. This method uses a steady-state k-ε two-equation model combined with a component transport equation to simulate the flow field in a typical two-lane street canyon and the atmospheric pollutant concentration field of a substation fire; based on the simulation, a CFD model of pollutant diffusion in a specific area of ​​a typical urban area is established, and the actual parameters are determined to analyze the diffusion of pollutants under different incoming wind directions and wind speeds; different wind directions and wind speed ranges obtained through S3 are selected, and the wind speed field and pollutant concentration field are simulated respectively to obtain the extreme values ​​of pollutant concentration and their areas under different wind direction angles, and then the distribution of pollutants in different areas is analyzed.

[0005] The above technical solution has the following disadvantages:

[0006] 1. CFD models are extremely complex to build without the use of professional software tools.

[0007] 2. The preparation and processing of input data for CFD models is a huge workload, and missing data may cause large deviations in the results.

[0008] 3. The CFD model calculation time is very long, and the desired calculation results cannot be obtained in a short time. Summary of the invention

[0009] In view of the above problems, the present invention provides a flow field data diffusion calculation method, device, equipment and medium for overcoming the above problems or at least partially solving the above problems.

[0010] The present invention provides the following scheme:

[0011] A flow field data diffusion calculation method, comprising:

[0012] Acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration;

[0013] Converting the grid data into a graph structure;

[0014] The nodes and edges in the graph structure are encoded using physical information, position information and node types of the nodes in the grid, the graph structure is coarsened in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and all nodes of the graph structure are classified into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set;

[0015] The features of the coarsened graph structure are summarized and extracted by using a graph neural network block through message passing; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update the features to obtain an updated graph structure;

[0016] In the graph recovery layer, a spatial interpolation method is used to upsample the updated graph structure and a fully connected layer is used to extract features;

[0017] The decoder is used to decode the result output by the graph recovery layer to obtain the actual parameter information of each node, and the concentration value and diffusion condition of the pollutant in the entire grid area.

[0018] Preferably: after the grid data is converted into a graph structure, the encoder uses a multi-layer perceptron to embed the converted graph features into a high-dimensional latent vector.

[0019] Preferably: the graph feature is embedded into a high-dimensional latent vector as follows:

[0020] V i =[pos i ,physical parameters],E ij =[u(pos i ,pos j )]

[0021] v i =MLP v (V i ),e ij =MLP e (E ij )

[0022] Where: pos i is the coordinate of node i, and physical parameters are some scalars, u(posi ,pos j ) represents the distance between node i and node j.

[0023] Preferably: the connectivity relationship of the coarsened graph structure is obtained using the continuation matrix and the adjacency matrix of the graph.

[0024] Preferably: the connectivity relationship is represented by the following formula:

[0025] A c =P T AP

[0026] Where: P represents the extension matrix, A represents the adjacency matrix of the refined graph, and A c Represents the adjacency matrix of the coarsened graph.

[0027] Preferably: the message passing of the graph neural network block is represented by the following formula:

[0028] e ij =MLP e (e′ ij ,v i ,v j ),

[0029] Where: e i ' j and v i ′ represents the output of the previous graph neural network block, N i is the set of adjacent nodes of node i. Preferably, the feature of node i is calculated by the following formula:

[0030] in

[0031] Where: Represents the spatial distance between nodes i and j.

[0032] A flow field data diffusion calculation device, used to execute the flow field data diffusion calculation method, the device comprising:

[0033] A grid data acquisition unit, used to acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration;

[0034] A graph structure conversion unit, used to convert the grid data into a graph structure;

[0035] A coarsening unit is used to encode nodes and edges in the graph structure by using physical information, position information and node types of nodes in the grid, to coarsen the graph structure in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and to classify all nodes of the graph structure into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set;

[0036] A graph structure feature extraction unit, used to summarize and extract features of the coarsened graph structure by using a neural network block through message passing; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update features to obtain an updated graph structure;

[0037] An upsampling unit, configured to upsample the updated graph structure using a spatial interpolation method in a graph recovery layer and extract features using a fully connected layer;

[0038] The decoding unit is used to decode the result output by the graph recovery layer using a decoder to obtain the actual parameter information of each node and the concentration value and diffusion condition of the pollutant in the entire grid area.

[0039] A flow field data diffusion computing device, the device comprising a processor and a memory:

[0040] The memory is used to store program code and transmit the program code to the processor;

[0041] The processor is used to execute the above-mentioned flow field data diffusion calculation method according to the instructions in the program code.

[0042] A computer-readable storage medium is used to store program codes, and the program codes are used to execute the above-mentioned flow field data diffusion calculation method.

[0043] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0044] The embodiment of the present application provides a flow field data diffusion calculation method, device, equipment and medium. The method uses a GNN model instead of CFD to calculate the diffusion of pollutant data in an unstructured grid area, which can multiply the calculation speed and accurately obtain the pollutant diffusion results of the unstructured grid. In the case of a sufficiently large training data set, the calculation is also effective for a new regional grid, and there is no need to use CFD for re-modeling and calculation, which shortens the entire data processing and calculation cycle.

[0045] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 is a flow chart of a flow field data diffusion calculation method provided by an embodiment of the present invention;

[0048] Figure 2 It is a flowchart of a flow field data diffusion calculation method provided by an embodiment of the present invention;

[0049] Figure 3 Schematic diagram of the extension matrix of the coarsening node set detail point set provided by an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of a graph structure provided by an embodiment of the present invention;

[0051] Figure 5 It is a schematic diagram of a process of updating graph coarsening layer features by graph neural network block message passing provided by an embodiment of the present invention;

[0052] Figure 6 is a schematic diagram of a flow field data diffusion calculation device provided by an embodiment of the present invention;

[0053] Figure 7 It is a schematic diagram of a flow field data diffusion calculation device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0055] See also Figure 1 , is a flow field data diffusion calculation method provided by an embodiment of the present invention, such as Figure 1 As shown, the method may include:

[0056] S101: Acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration;

[0057] S102: Convert the grid data into a graph structure; in specific implementation, the embodiment of the present application may provide that after the grid data is converted into a graph structure, the encoder uses a multi-layer perceptron to embed the converted graph features into a high-dimensional latent vector.

[0058] Furthermore, the graph features are embedded into a high-dimensional latent vector as follows:

[0059] V i =[pos i ,physical parameters],E ij =[u(pos i ,pos j )]

[0060] v i =MLP v (V i ),e ij =MLP e (E ij )

[0061] Where: pos i is the coordinate of node i, and physical parameters are some scalars, u(pos i ,pos j ) represents the distance between node i and node j.

[0062] The connectivity relationship of the coarsened graph structure is obtained using the continuation matrix and the adjacency matrix of the graph.

[0063] The connectivity relationship is expressed by the following formula:

[0064] A c =P T AP

[0065] Where: P represents the extension matrix, A represents the adjacency matrix of the refined graph, and A c Represents the adjacency matrix of the coarsened graph.

[0066] S103: Encoding nodes and edges in the graph structure using physical information, location information, and node types of nodes in the grid, performing coarsening processing on the graph structure in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and classifying all nodes of the graph structure into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set;

[0067] S104: Summarize and extract the features of the coarsened graph structure by message passing using the graph neural network block; in the graph neural network block, the node aggregates information from all adjacent nodes, and uses the aggregated information to update the features to obtain the updated graph structure; in specific implementation, the embodiment of the present application can provide that the message passing of the graph neural network block is represented by the following formula:

[0068] e ij =MLP e (e′ ij ,v i ,v j ),

[0069] Where: e i ' j and v i ′ represents the output of the previous graph neural network block, N i is the set of adjacent nodes of node i.

[0070] S105: In the graph recovery layer, the spatial interpolation method is used to upsample the updated graph structure and the fully connected layer is used to extract features. In specific implementation, the embodiment of the present application can provide the feature of node i to be calculated by the following formula:

[0071] in

[0072] Where: Represents the spatial distance between nodes i and j.

[0073] S106: Use a decoder to decode the result output by the graph recovery layer to obtain the actual parameter information of each node, and obtain the concentration value and diffusion status of the pollutant in the entire grid area.

[0074] The flow field data diffusion calculation method provided in the embodiment of the present application performs predictive calculation on the fluid data in a grid area based on GNN, converts the grid result into a graph structure, encodes the physical information, position information and node type of the nodes in the grid to encode the nodes and edges in the graph, and performs coarsening processing in combination with the coarsening layer of the algebraic multigrid algorithm, classifies all nodes into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set; uses the neural network information block to summarize and extract the features of the graph; performs upsampling processing in the graph recovery layer, and finally uses a decoder to obtain the data information of each node.

[0075] The method provided by this application is described in detail below. Figure 2 shown.

[0076] The input data of the method provided in this application is regional unstructured grid data of a certain pollutant at different release concentrations. The data of each grid contains longitude and latitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration. According to the structure of the grid, it is converted into edge features in the graph structure. The grid consists of node coordinates and the topological relationship of the nodes. Therefore, it can be interpreted as a graph.

[0077] Define a graph G = (V, E) to represent a physical system, V is a set of grid nodes, and the edge set E comes from the connection of grid nodes. At the same time, the encoder uses a multi-layer perceptron (MLP) to embed the converted graph features into a high-dimensional latent vector.

[0078] V i =[pos i ,physical parameters],E ij =[u(pos i ,pos j )]

[0079] v i =MLP v (V i ),e ij =MLP e (E ij )

[0080] Among them, pos i are the coordinates of node i, and the physical parameters are some scalar quantities, such as the Mach number and angle of attack (AOA) in the airfoil scenario or the Reynolds number in the cylindrical flow scenario. u(pos i ,pos j ) represents the distance between node i and node j.

[0081] Roughening treatment:

[0082] For graph coarsening, we first use the traditional algebraic multigrid algorithm to divide all nodes into a coarsening node set S and a detail point set V, where the coarsening nodes are a subset of the detail points. The next step is to calculate and obtain the continuation matrix P.

[0083]

[0084] The coarsened node set S corresponds to the columns of the continuation matrix P, while the detail point set V corresponds to the rows. If the edges in the continuation matrix P exist in the edge set of the graph, then p ij =e ij , otherwise p ij = 0. When the fine graph is coarsened, a new adjacency matrix A is constructed c To maintain the connectivity of the coarsened graph, the continuation matrix P and the adjacency matrix of the graph are used to obtain the connectivity relationship of the coarsened graph.

[0085] A c =P T AP

[0086] in Matrix A is the adjacency matrix of the refined graph, A c is the adjacency matrix of the coarsened graph. If A cij In A c If is not 0, then there is an edge E in the coarsened graph. ij , and the edge feature is A cij The connectivity and edge set E of the coarsened graph are obtained through the graph coarsening algorithm of algebraic multigrid. c And the edge feature e c The continuation matrix P of the coarsened node set and the graph structure are as follows: Figure 3 , Figure 4 shown.

[0087] Neural network updates node features.

[0088] The neural network block is used to summarize and extract graph features through message passing. In the graph neural network block, the node aggregates information from all adjacent nodes and uses the aggregated information to update the features. The coarsening layer of the graph is continuously iterated to obtain a more accurate feature description of the node. The formula for the message passing of the graph neural network block is as follows:

[0089] e ij =MLP e (e′ ij ,v i ,v j ), Among them, e i ' j and v i ′ is the output of the previous graph neural network block, N i is the set of neighboring nodes of node i. The process of updating the graph coarsening layer features through message passing of graph neural network blocks is as follows: Figure 5 shown.

[0090] Graph Neural Network (GNN) is a deep learning model for graph structured data that can effectively capture the complex relationships between nodes and their neighbors. The core idea of ​​GNN is to update the representation (embedding) of each node through message passing of neighboring nodes, so that the representation of the node not only depends on its own features, but also contains its local or global graph structure information.

[0091] Graph structure representation: Graphs are composed of nodes and edges. Nodes represent entities in the graph, and edges represent the relationship between nodes. Each node v has the initial features The goal of the graph is to learn more semantic feature representations by continuously updating node representations.

[0092] Information propagation: In each round of propagation, a node receives information from its neighboring nodes and updates its own representation. The updated representation contains not only its own features, but also combines the features of neighboring nodes. This process is achieved by aggregating the information of neighboring nodes. Common aggregation operations include summation, averaging, and weighted sum. At the kth layer, the representation update formula of node v is:

[0093]

[0094] Among them, N(v) represents the neighbor set of node v, and AGGREGATE is an aggregation function used to integrate the information of neighbor nodes.

[0095] Nonlinear transformation and feature update: After aggregating the information of neighboring nodes, the node features will be updated through a nonlinear activation function (such as ReLU), thereby introducing nonlinear transformation and enhancing the model's expressiveness. The update formula is:

[0096]

[0097] Where W is the trainable weight matrix and σ is the activation function.

[0098] Multi-layer propagation and global information capture: GNNs are usually composed of multiple layers, and each layer of node representation gradually expands its receptive field and combines information from more distant neighbor nodes. Therefore, through multi-layer GNNs, the final representation of the node not only contains its own features, but also captures information from multi-hop neighbors, realizing the learning of the global graph structure.

[0099] Figure recovery process.

[0100] The image recovery layer performs the opposite operation to the coarsening layer. In this step, the image is upsampled using spatial interpolation. is the node feature you want to recover, and the node set is X f ,and is the node feature of the coarsened graph, and the node set is X c . r j =h j ,j∈X c , where h j It appears in X c The feature of node j, r j It appears in X f The characteristics of node j in X. f But not in X c In the example above, we can locate X cThen the feature of node i can be calculated by the following formula:

[0101] in

[0102] in, is the spatial distance between nodes i and j. Each f Each node i in the i , so we can restore C to F. After each upsampling, a fully connected layer is used to extract features, that is, smoothing. The formula is as follows:

[0103] h i =MLP(r i ),h i ∈F.

[0104] Decoding process.

[0105] The decoder is used to decode the output of the graph recovery layer to obtain the actual parameter information of each node and the concentration value and diffusion of pollutants in the entire grid area.

[0106] It can be understood that the training data set of the model; a large number of effective training data sets simulated by CFD using meteorological and terrain parameters in a real environment serve as the training data source of the model provided by this method.

[0107] In summary, the flow field data diffusion calculation method provided by this application uses the GNN model instead of CFD to calculate the diffusion of pollutant data in the unstructured grid area, which can multiply the calculation speed and accurately obtain the pollutant diffusion results of the unstructured grid. In the case of a sufficiently large training data set, the calculation is also effective for a new regional grid, and there is no need to use CFD for re-modeling and calculation, which shortens the entire data processing and calculation cycle.

[0108] See also Figure 6 , the embodiment of the present application can also provide a flow field data diffusion calculation device, such as Figure 6 As shown, the device for executing the above-mentioned flow field data diffusion calculation method may include:

[0109] The grid data acquisition unit 601 is used to acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration;

[0110] A graph structure conversion unit 602, used to convert the grid data into a graph structure;

[0111] A coarsening unit 603 is used to encode nodes and edges in the graph structure by using physical information, position information and node types of nodes in the grid, perform coarsening processing on the graph structure in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and classify all nodes of the graph structure into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set;

[0112] A graph structure feature extraction unit 604, used to summarize and extract the features of the coarsened graph structure by message passing using a neural network block; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update the features to obtain an updated graph structure;

[0113] An upsampling unit 605 is used to upsample the updated graph structure using a spatial interpolation method in a graph recovery layer and extract features using a fully connected layer;

[0114] The decoding unit 606 is used to use a decoder to decode the result output by the graph recovery layer, obtain the actual parameter information of each node, and obtain the concentration value and diffusion status of the pollutant in the entire grid area.

[0115] The present application may also provide a flow field data diffusion computing device, the device comprising a processor and a memory:

[0116] The memory is used to store program code and transmit the program code to the processor;

[0117] The processor is used to execute the steps of the above-mentioned flow field data diffusion calculation method according to the instructions in the program code.

[0118] like Figure 7 As shown, a flow field data diffusion computing device provided in an embodiment of the present application may include: a processor 10, a memory 11, a communication interface 12 and a communication bus 13. The processor 10, the memory 11 and the communication interface 12 communicate with each other through the communication bus 13.

[0119] In the embodiment of the present application, the processor 10 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic devices, etc.

[0120] The processor 10 may call a program stored in the memory 11. Specifically, the processor 10 may execute operations in an embodiment of the flow field data diffusion calculation method.

[0121] The memory 11 is used to store one or more programs, which may include program codes, and the program codes include computer operation instructions. In the embodiment of the present application, the memory 11 at least stores programs for implementing the following functions:

[0122] Acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration;

[0123] Converting the grid data into a graph structure;

[0124] The nodes and edges in the graph structure are encoded using physical information, position information and node types of the nodes in the grid, the graph structure is coarsened in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and all nodes of the graph structure are classified into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set;

[0125] The features of the coarsened graph structure are summarized and extracted by using a neural network block through message passing; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update the features to obtain an updated graph structure;

[0126] In the graph recovery layer, a spatial interpolation method is used to upsample the updated graph structure and a fully connected layer is used to extract features;

[0127] The decoder is used to decode the result output by the graph recovery layer to obtain the actual parameter information of each node, and the concentration value and diffusion condition of the pollutant in the entire grid area.

[0128] In one possible implementation, the memory 11 may include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required for at least one function (such as a file creation function, a data reading and writing function), etc.; the data storage area can store data created during use, such as initialization data, etc.

[0129] In addition, the memory 11 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0130] The communication interface 12 may be an interface of a communication module, and is used to connect to other devices or systems.

[0131] Of course, it should be noted that Figure 7 The structure shown does not constitute a limitation on the flow field data diffusion calculation device in the embodiment of the present application. In actual applications, the flow field data diffusion calculation device may include Figure 7 More or fewer components than shown, or combinations of certain components.

[0132] The embodiment of the present application may also provide a computer-readable storage medium, wherein the computer-readable storage medium is used to store program codes, and the program codes are used to execute the steps of the above-mentioned flow field data diffusion calculation method.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0134] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0135] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A flow field data diffusion calculation method, characterized in that: include: Acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration; Converting the grid data into a graph structure; The nodes and edges in the graph structure are encoded using physical information, position information and node types of the nodes in the grid, the graph structure is coarsened in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and all nodes of the graph structure are classified into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set; The features of the coarsened graph structure are summarized and extracted by using a graph neural network block through message passing; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update the features to obtain an updated graph structure; In the graph recovery layer, a spatial interpolation method is used to upsample the updated graph structure and a fully connected layer is used to extract features; The decoder is used to decode the result output by the graph recovery layer to obtain the actual parameter information of each node, and the concentration value and diffusion condition of the pollutant in the entire grid area.

2. The flow field data diffusion calculation method according to claim 1, characterized in that: After the grid data is converted into a graph structure, the encoder uses a multi-layer perceptron to embed the converted graph features into a high-dimensional latent vector.

3. The flow field data diffusion calculation method according to claim 2, characterized in that: The graph features are embedded into a high-dimensional latent vector as follows: V i =[pos i ,physical parameters],E ij =[u(pos i ,pos j )] v i =MLP v (V i ),e ij =MLP e (E ij ) Where: pos i is the coordinate of node i, and physical parameters are some scalars, u(pos i ,pos j ) represents the distance between node i and node j.

4. The flow field data diffusion calculation method according to claim 1, characterized in that: The connectivity relationship of the coarsened graph structure is obtained using the continuation matrix and the adjacency matrix of the graph.

5. The flow field data diffusion calculation method according to claim 4, characterized in that: The connectivity relationship is expressed by the following formula: A c =P T AP Where: P represents the extension matrix, A represents the adjacency matrix of the refined graph, and A c Represents the adjacency matrix of the coarsened graph.

6. The flow field data diffusion calculation method according to claim 1, characterized in that: The message passing of the graph neural network block is expressed as follows: Where: e i ' j and v i ′ represents the output of the previous graph neural network block, N i is the set of adjacent nodes of node i.

7. The flow field data diffusion calculation method according to claim 1, characterized in that: The characteristics of node i are calculated using the following formula: in Where: Represents the spatial distance between nodes i and j.

8. A flow field data diffusion calculation device, characterized in that: The device is used to execute the flow field data diffusion calculation method according to any one of claims 1 to 7, comprising: A grid data acquisition unit, used to acquire regional unstructured grid data of pollutants at different release concentrations, wherein the data packet of each grid in the grid data at least includes latitude and longitude, terrain, temperature, humidity, wind speed, wind direction, cloud height, and pollutant concentration; A graph structure conversion unit, used to convert the grid data into a graph structure; A coarsening unit is used to encode nodes and edges in the graph structure by using physical information, position information and node types of nodes in the grid, to coarsen the graph structure in combination with a coarsening layer including an algebraic multigrid algorithm to obtain a coarsened graph structure, and to classify all nodes of the graph structure into a coarse node set and a detail point set, wherein the coarse node set is a subset of the detail point set; A graph structure feature extraction unit, used to summarize and extract features of the coarsened graph structure by using a neural network block through message passing; in the graph neural network block, a node aggregates information from all adjacent nodes, and uses the aggregated information to update features to obtain an updated graph structure; An upsampling unit, configured to upsample the updated graph structure using a spatial interpolation method in a graph recovery layer and extract features using a fully connected layer; The decoding unit is used to decode the result output by the graph recovery layer using a decoder to obtain the actual parameter information of each node and the concentration value and diffusion condition of the pollutant in the entire grid area.

9. A flow field data diffusion calculation device, characterized in that: The device comprises a processor and a memory: The memory is used to store program codes and transmit the program codes to the processor; The processor is used to execute the flow field data diffusion calculation method according to any one of claims 1 to 7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the flow field data diffusion calculation method according to any one of claims 1 to 7.

Images