Multi-stage optimization space-time tracing method and device and storage medium

Abstract

The invention relates to a multi-stage optimization space-time tracing method and device and a storage medium, and is applied to the technical field of pollution monitoring. The method specifically comprises the following steps: constructing a closed-loop pollution source contribution optimization and traceability system by combining an AERMOD model and a deep learning model, combining simulation output of AERMOD with actual monitoring data, meteorological data and geographic space information, and performing correction through the deep learning model. The optimized total concentration contribution is further decomposed into the contribution proportion of a single pollutant, multi-feature input is used for integrating multi-dimensional features such as historical contribution weight, meteorological data and geographic information, the output result of the model is dynamically fed back to a front-layer structure, and the pollution source contribution weight is continuously optimized, so that the pollution source contribution ratio is optimized; and closed-loop linkage from data input to strategy output is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of pollution monitoring, and particularly relates to a multi-stage optimized spatio-temporal tracing method, device and storage medium. Background Art

[0002] Accurately evaluating the contribution of emission sources to the pollutant concentration at monitoring points and analyzing the pollutant concentration contribution at different times have become key tasks for precise governance and air quality management. Traditional methods for identifying pollutant sources and predicting concentrations mainly rely on diffusion models and simple statistical analyses.

[0003] In the prior art, there is a method combining a traditional diffusion model with monitoring station data. Its technical solution includes the following steps: collecting the emission inventory of major pollution sources in the region, including pollutant emission intensity, emission source location, emission time, etc., as the basic data input for pollutant diffusion simulation, using the diffusion model to numerically simulate the spatial diffusion trajectory of pollutants, calculating the concentration contribution value of pollutants diffusing from the emission source to each monitoring point based on the physical formula of pollutant diffusion, calibrating and adjusting the simulation results in combination with the actual observed concentration data of pollutants at the regional monitoring stations to reduce the deviation between the simulated concentration and the actual concentration, and decomposing the pollutant concentration at the monitoring point into the contribution ratios of different emission sources through model calculation to provide a basis for pollution control;

[0004] Although these methods can provide a theoretical basis for pollutant diffusion and emission source contribution, there are still some deficiencies, such as insufficient spatial feature modeling, limited time feature capture ability, and insufficient real-time and accuracy of contribution decomposition. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a multi-stage optimized spatio-temporal tracing method, device and storage medium to solve the problems of insufficient spatial feature modeling, limited time feature capture ability, and insufficient real-time and accuracy of contribution decomposition in the prior art.

[0006] According to the first aspect of the embodiments of the present invention, a multi-stage optimized spatio-temporal tracing method is provided. The method includes:

[0007] Obtaining meteorological data within a preset range of a target site for a specified historical period, obtaining topographic and geomorphic data within the preset range of the target site, and obtaining monitoring data of various pollutant monitoring points within the preset range of the target site for the specified historical period;

[0008] Preprocessing the meteorological data, topographic and geomorphic data, and monitoring data of various pollutant monitoring points to obtain first input data;

[0009] Input the first input data into the pre-built AERMOD model to obtain the total concentration contribution of pollution sources at the target site;

[0010] Obtain the real-time monitoring data of various pollutant monitoring points, and obtain the initial weight of the pollution source contribution according to the proportion of each pollution source in the real-time monitoring data of various pollutant monitoring points;

[0011] Based on the total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the initial weight of the pollution source contribution, obtain the corrected total concentration contribution of pollution sources at the target site through the pre-built first spatio-temporal deep learning model DCRNN;

[0012] Based on the corrected total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, and the initial weight of the pollution source contribution, obtain the corrected single pollutant concentration contribution value through the pre-built second spatio-temporal deep learning model DCRNN;

[0013] Based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site, obtain the contribution weight of the pollution source through the pre-built third spatio-temporal deep learning model DCRNN;

[0014] Update the initial weight of the pollution source contribution through the contribution weight of the pollution source, and obtain the corrected single pollutant concentration contribution value through the updated contribution weight of the pollution source.

[0015] Preferably,

[0016] The obtaining of the corrected total concentration contribution of pollution sources at the target site through the pre-built first spatio-temporal deep learning model DCRNN based on the total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the initial weight of the pollution source contribution includes:

[0017] Preprocess the total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the initial weight of the pollution source contribution to obtain the second input data;

[0018] Construct a first node feature matrix according to the second input data to obtain graph structure information; input the first node feature matrix and the graph structure information into the pre-built first spatio-temporal deep learning model DCRNN to obtain the corrected total concentration contribution of pollution sources at the target site.

[0019] Preferably,

[0020] Based on the total concentration contribution of pollution sources at the corrected target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, and the initial weights of the pollution source contributions, obtaining the corrected single pollutant concentration contribution value through the pre-built second spatio-temporal deep learning model DCRNN includes:

[0021] Obtaining the corrected pollutant contribution according to the total concentration contribution of pollution sources at the corrected target site and the initial weights of the pollution source contributions, and obtaining the error evaluation of the first spatio-temporal deep learning model DCRNN according to the total concentration contribution of pollution sources at the corrected target site and the total concentration contribution of pollution sources at the target site;

[0022] Obtaining the real-time meteorological data within the preset range of the target site, and preprocessing the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, the initial weights of the pollution source contributions, the total concentration contribution of pollution sources at the corrected target site, the corrected pollutant contribution, and the error evaluation to obtain the third input data;

[0023] Constructing a second node feature matrix according to the third input data, and inputting the second node feature matrix and the graph structure information into the pre-built second spatio-temporal deep learning model DCRNN to obtain the corrected single pollutant concentration contribution value.

[0024] Preferably,

[0025] Based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site, obtaining the contribution weights of pollution sources through the pre-built third spatio-temporal deep learning model DCRNN includes:

[0026] Preprocessing the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site to obtain the fourth input data;

[0027] Constructing a third node feature matrix according to the fourth input data, and inputting the third node feature matrix and the graph structure information into the pre-built third spatio-temporal deep learning model DCRNN to obtain the contribution weights of pollution sources.

[0028] Preferably,

[0029] The obtaining of the graph structure information includes:

[0030] Obtaining the graph structure information according to the spatial distribution of the total concentration contribution of pollution sources at the target site within the preset range of the target site.

[0031] Preferably,

[0032] The first spatio-temporal deep learning model DCRNN, the second spatio-temporal deep learning model DCRNN, and the third spatio-temporal deep learning model DCRNN all include: a graph convolution module, a recurrent neural network module, and a decoding module;

[0033] The graph convolution module is used to extract the spatial features of the input features, the recurrent neural network module is used to extract the temporal features of the input features, and the decoding module is used to fuse the extracted spatial features and temporal features.

[0034] According to the second aspect of the embodiments of the present invention, a multi-stage optimized spatio-temporal traceability device is provided. The device includes:

[0035] A first data acquisition module: used to acquire meteorological data of a specified historical period within a preset range of a target site, acquire topographic and geomorphic data within the preset range of the target site, and acquire monitoring data of various pollutant monitoring points within the preset range of the target site during the specified historical period;

[0036] A first input module: used to preprocess the meteorological data, topographic and geomorphic data, and monitoring data of various pollutant monitoring points to obtain first input data;

[0037] An AERMOD module: used to input the first input data into a pre-built AERMOD model to obtain the total concentration contribution of pollution sources at the target site;

[0038] A second data acquisition module: used to acquire real-time monitoring data of various pollutant monitoring points, and obtain the initial weight of the pollution source contribution according to the proportion of each pollution source in the real-time monitoring data of various pollutant monitoring points;

[0039] A first deep learning module: used to obtain the corrected total concentration contribution of pollution sources at the target site through a pre-built first spatio-temporal deep learning model DCRNN based on the total concentration contribution of pollution sources at the target site, real-time monitoring data of various pollutant monitoring points, topographic and geomorphic data, and the initial weight of the pollution source contribution;

[0040] A second deep learning module: used to obtain the corrected single pollutant concentration contribution value through a pre-built second spatio-temporal deep learning model DCRNN based on the corrected total concentration contribution of pollution sources at the target site, real-time monitoring data of various pollutant monitoring points, real-time meteorological data, and the initial weight of the pollution source contribution;

[0041] A third deep learning module: used to obtain the contribution weight of the pollution source through a pre-built third spatio-temporal deep learning model DCRNN based on the corrected single pollutant concentration contribution value, real-time monitoring data of various pollutant monitoring points, topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site.

[0042] Circular update module: used to update the initial weight contributed by the pollution source through the contribution weight of the pollution source, and obtain the corrected contribution value of a single pollutant concentration through the updated contribution weight of the pollution source.

[0043] According to the third aspect of the embodiments of the present invention, a storage medium is provided. The storage medium stores a computer program, and when the computer program is executed by a main controller, each step in the above method is implemented.

[0044] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0045] In this application, a closed-loop pollution source contribution optimization and traceability system is constructed by combining the AERMOD model and the deep learning model. The simulated output of AERMOD is combined with actual monitoring data, meteorological data, and geospatial information, and corrected by the deep learning model. The optimized total concentration contribution is further decomposed into the contribution ratio of a single pollutant. Through integrating multi-dimensional features such as historical contribution weights, meteorological data, and geographical information, the deep learning model can capture complex spatio-temporal dependence relationships, ensuring that the decomposition result is more in line with the actual situation. The output result of the model can be dynamically fed back to the previous layer structure. By continuously optimizing the pollution source contribution weight, a closed-loop linkage from data input to strategy output is realized. In this application, deep learning methods such as DCRNN are introduced into the pollution traceability closed-loop system, combining the advantages of traditional simulation models, realizing the deep integration of data-driven and physical models, and realizing the linkage between AERMOD and the deep learning model through a dynamic feedback mechanism, improving the accuracy.

[0046] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0048] Figure 1 is a schematic flowchart of a multi-stage optimization spatio-temporal traceability method shown according to an exemplary embodiment;

[0049] Figure 2 is a schematic overall implementation process framework diagram shown according to another exemplary embodiment;

[0050] Figure 3 is a schematic diagram of the optimized total concentration contribution shown according to another exemplary embodiment;

[0051] Figure 4Schematic diagram of the corrected concentration contribution of a single pollutant shown according to another exemplary embodiment;

[0052] Figure 5 Schematic diagram of the source contribution weight shown according to another exemplary embodiment;

[0053] Figure 6 Schematic diagram of a multi-stage optimized spatio-temporal tracing device shown according to another exemplary embodiment;

[0054] In the drawings: 1 - First data acquisition module, 2 - First input module, 3 - AERMOD module, 4 - Second data acquisition module, 5 - First deep learning module, 6 - Second deep learning module, 7 - Third deep learning module, 8 - Loop update module. Detailed implementation mode

[0055] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all the implementation manners consistent with the present invention. On the contrary, they are only examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0056] Embodiment 1

[0057] Figure 1 Schematic diagram of the process of a multi-stage optimized spatio-temporal tracing method shown according to an exemplary embodiment, as Figure 1 shown, the method includes:

[0058] S1. Obtain meteorological data within a preset range of the target site for a specified historical period, obtain topographic and geomorphic data within the preset range of the target site, and obtain monitoring data of various pollutant monitoring points within the preset range of the target site for the specified historical period;

[0059] S2. Preprocess the meteorological data, topographic and geomorphic data, and monitoring data of various pollutant monitoring points to obtain first input data;

[0060] S3. Input the first input data into a pre-built AERMOD model to obtain the total concentration contribution of the pollution sources at the target site;

[0061] S4. Obtain real-time monitoring data of various pollutant monitoring points, and obtain the initial weight of the source contribution according to the proportion of each pollution source in the real-time monitoring data of various pollutant monitoring points;

[0062] S5. Based on the total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the initial weights of the pollution source contributions, obtain the corrected total concentration contribution of pollution sources at the target site through the pre-established first spatio-temporal deep learning model DCRNN;

[0063] S6. Based on the corrected total concentration contribution of pollution sources at the target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, and the initial weights of the pollution source contributions, obtain the corrected concentration contribution value of a single pollutant through the pre-established second spatio-temporal deep learning model DCRNN;

[0064] S7. Based on the corrected concentration contribution value of a single pollutant, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site, obtain the contribution weights of pollution sources through the pre-established third spatio-temporal deep learning model DCRNN;

[0065] S8. Update the initial weights of the pollution source contributions with the contribution weights of the pollution sources, and obtain the corrected concentration contribution value of a single pollutant through the updated contribution weights of the pollution sources;

[0066] It can be understood that as shown in the appendix Figure 2 First, it is necessary to determine the emission sources at the target site, then obtain the meteorological data for a specified historical period within the preset range of the target site (usually a circular area centered on the target site with a radius of Y), obtain the geospatial information within the preset range of the target site (generally topographic and geomorphic data), and obtain the monitoring data of various pollutant monitoring points within the preset range of the target site for a specified historical period; preprocess the above data, which generally involves deleting outliers, filling in missing values, and standardizing data from different sources and in different formats, to obtain the first input data;

[0067] Input the first input data into the pre-established AERMOD model, and the model outputs the total contribution of pollution source concentrations at the target site; among them, the AERMOD model is based on the Gaussian steady-state plume diffusion model, generally used to predict the concentration distribution of pollutants in the atmosphere. By considering meteorological conditions, terrain factors, etc., this model can provide relatively accurate concentration prediction data; it is applicable to the emissions of various emission sources (including point sources, area sources, and volume sources), and is applicable to the simulation and prediction of various emission diffusion scenarios such as rural environments and urban environments, flat terrains and complex terrains, ground sources and elevated sources;

[0068] Then obtain the current meteorological data and the real-time monitoring data of various pollutant monitoring points, and obtain the initial weights of the pollution source contributions according to the proportions of each pollution source in the real-time monitoring data of various pollutant monitoring points; construct the second input data according to the output of the AERMOD model, the current meteorological data, the real-time monitoring data of various pollutant monitoring points, the initial weights of the pollution source contributions, etc.; construct the first node feature matrix X according to the second input data, that is, the multi-feature data set of multiple sites; at the same time, obtain the graph structure information according to the spatial distribution of the total concentration contribution of the pollution sources at the target site within the preset range of the target site; use the first node feature matrix X and the graph structure information as the input of the pre-built first spatio-temporal deep learning model DCRNN, and obtain the corrected total concentration contribution of the pollution sources at the target site through the first spatio-temporal deep learning model DCRNN, as shown in the appendix Figure 3 as shown;

[0069] Among them, the spatio-temporal deep learning model DCRNN, that is, the diffusion convolutional recurrent neural network, uses diffusion convolution to capture spatial features with the topological structure of the graph (the relationship between nodes and edges). Diffusion convolution is a convolutional operation based on a directed graph. It captures the information propagation from one node to other nodes by simulating random walks and uses an extended version based on GRU to construct a decoding module to accept the spatio-temporal features of the hidden state and map them to the target variables; simply put, the spatio-temporal deep learning model DCRNN includes a graph convolution module, a recurrent neural network module, and a decoding module. The graph convolution module is used to extract the spatial features of the input features, the recurrent neural network module is used to extract the temporal features of the input features, and the decoding module fuses the spatial features and the temporal features to obtain spatio-temporal features.

[0070] Obtain the corrected pollutant contributions according to the corrected total concentration contributions of the pollution sources at the target site and the initial weights of the pollution source contributions, and obtain the error evaluation; obtain the real-time meteorological data, and preprocess the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, the initial weights of the pollution source contributions, the corrected total concentration contributions of the pollution sources at the target site, the corrected pollutant contributions, and the error evaluation to ensure the consistency of the input time series and spatial features, and obtain the third input data; construct the second node feature matrix X according to the third input data, and input the second node feature matrix and the graph structure information into the pre-built second spatio-temporal deep learning model DCRNN to obtain the corrected single pollutant concentration contribution value, as shown in the appendix Figure 4 as shown;

[0071] Then, preprocess the corrected contribution values of individual pollutant concentrations, real-time monitoring data of various pollutant monitoring points, topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site to obtain the fourth input data; construct a third node feature matrix based on the fourth input data, and input the third node feature matrix and the graph structure information into the pre-built third spatio-temporal deep learning model DCRNN to obtain the contribution weights of pollution sources, as shown in the appendix Figure 5 shown;

[0072] Update the contribution weights to the above optimization process. The updated contribution weights of pollution sources can more accurately reflect the impact of pollution sources on the target site; by continuously optimizing the contribution weights of pollution sources, a closed-loop linkage from data input to strategy output is achieved.

[0073] Embodiment 2:

[0074] Figure 6 is a system schematic diagram of a multi-stage optimized spatio-temporal tracing device shown according to another exemplary embodiment. The device includes:

[0075] The first data acquisition module 1: used to acquire meteorological data within a preset range of the target site for a specified historical period, acquire topographic and geomorphic data within a preset range of the target site, and acquire monitoring data of various pollutant monitoring points within a preset range of the target site for a specified historical period;

[0076] The first input module 2: used to preprocess the meteorological data, topographic and geomorphic data, and monitoring data of various pollutant monitoring points to obtain the first input data;

[0077] The AERMOD module 3: used to input the first input data into the pre-built AERMOD model to obtain the total concentration contribution of pollution sources at the target site;

[0078] The second data acquisition module 4: used to acquire real-time monitoring data of various pollutant monitoring points, and obtain the initial weights of pollution source contributions according to the proportion of each pollution source in the real-time monitoring data of various pollutant monitoring points;

[0079] The first deep learning module 5: used to obtain the corrected total concentration contribution of pollution sources at the target site through the pre-built first spatio-temporal deep learning model DCRNN based on the total concentration contribution of pollution sources at the target site, real-time monitoring data of various pollutant monitoring points, topographic and geomorphic data, and the initial weights of pollution source contributions;

[0080] The second deep learning module 6: It is used to obtain the corrected single pollutant concentration contribution value through the pre-established second spatio-temporal deep learning model DCRNN based on the total concentration contribution of pollution sources at the corrected target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data, and the initial weights of pollution source contributions;

[0081] The third deep learning module 7: It is used to obtain the contribution weight of the pollution source through the pre-established third spatio-temporal deep learning model DCRNN based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic and geomorphic data, and the total concentration contribution of pollution sources at the target site;

[0082] The loop update module 8: It is used to update the initial weights of the pollution source contributions through the contribution weights of the pollution source, and obtain the corrected single pollutant concentration contribution value through the updated contribution weights of the pollution source.

[0083] Embodiment 3:

[0084] This embodiment provides a storage medium, which stores a computer program. When the computer program is executed by the main controller, it implements each step in the above method;

[0085] It can be understood that the above-mentioned storage medium can be a read-only memory, a magnetic disk or an optical disc, etc.

[0086] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and the content not detailed in some embodiments can be referred to the same or similar content in other embodiments.

[0087] It should be noted that in the description of the present invention, terms such as "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "a small number of sparsely distributed" means at least two.

[0088] Any process or method description shown in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of executable instructions including one or more sparsely distributed steps for implementing a specific logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the technical field to which the embodiments of the present invention belong.

[0089] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, a small number of sparsely distributed steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0090] Those of ordinary skill in the art can understand that all or part of the steps carried by the method of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0091] In addition, in each embodiment of the present invention, the functional units can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0092] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, etc.

[0093] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or a small number of sparsely distributed embodiments or examples.

[0094] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-stage optimized spatiotemporal tracing method, characterized in that: The method comprises: Obtain meteorological data for a specified historical period within a preset range of the target site, obtain topographic data within a preset range of the target site, and obtain monitoring data for various pollutant monitoring points within a specified historical period within the preset range of the target site; Preprocessing the meteorological data, topographic data, and monitoring data of various pollutant monitoring points to obtain first input data; Input the first input data into the pre-built AERMOD model to obtain the total concentration contribution of the pollution source of the target site; Obtaining real-time monitoring data of various pollutant monitoring points, and obtaining initial weights of pollution source contributions according to the proportions of various pollution sources in the real-time monitoring data of various pollutant monitoring points; Based on the total pollution source concentration contribution of the target site, the real-time monitoring data of various pollutant monitoring points, the topographic data and the initial weight of the pollution source contribution, the corrected total pollution source concentration contribution of the target site is obtained through the pre-built first spatiotemporal deep learning model DCRNN; Based on the total pollution source concentration contribution of the corrected target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data and the initial weight of the pollution source contribution, the corrected single pollutant concentration contribution value is obtained through the pre-built second spatiotemporal deep learning model DCRNN; Based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, topographic data and the total concentration contribution of pollution sources at the target site, the contribution weight of the pollution source is obtained through the pre-built third spatiotemporal deep learning model DCRNN; The initial weight of the pollution source contribution is updated by the contribution weight of the pollution source, and the corrected single pollutant concentration contribution value is obtained by the updated contribution weight of the pollution source.

2. The method according to claim 1, characterized in that The total concentration contribution of pollution sources of the target site based on the total pollution source concentration contribution of the target site, the real-time monitoring data of various pollutant monitoring points, the topographic data and the initial weight of the pollution source contribution, and the total pollution source concentration contribution of the target site after correction obtained by the pre-built first spatiotemporal deep learning model DCRNN include: Preprocessing the total concentration contribution of pollution sources at the target site, real-time monitoring data of various pollutant monitoring points, topographic data, and initial weights of pollution source contributions to obtain second input data; A first node feature matrix is ​​constructed according to the second input data to obtain graph structure information; the first node feature matrix and graph structure information are input into a pre-built first spatiotemporal deep learning model DCRNN to obtain the corrected total concentration contribution of pollution sources at the target site.

3. The method according to claim 2, characterized in that The total pollution source concentration contribution of the corrected target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data and the initial weight of the pollution source contribution are obtained through the pre-built second spatiotemporal deep learning model DCRNN to obtain the corrected single pollutant concentration contribution value, including: Obtaining a corrected pollutant contribution according to the corrected total pollution source concentration contribution of the target site and the initial weight of the pollution source contribution, and obtaining an error evaluation of the first spatiotemporal deep learning model DCRNN according to the corrected total pollution source concentration contribution of the target site and the total pollution source concentration contribution of the target site; Acquire real-time meteorological data within a preset range of the target site, pre-process the real-time monitoring data of the various pollutant monitoring points, the real-time meteorological data, the initial weight of the pollution source contribution, the total concentration contribution of the pollution source of the target site after correction, the pollutant contribution after correction, and the error evaluation to obtain third input data; A second node feature matrix is ​​constructed according to the third input data, and the second node feature matrix and graph structure information are input into a pre-built second spatiotemporal deep learning model DCRNN to obtain a corrected single pollutant concentration contribution value.

4. The method according to claim 3, characterized in that The contribution weight of the pollution source is obtained by using the pre-built third spatiotemporal deep learning model DCRNN based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic data and the total concentration contribution of the pollution source of the target site, including: Preprocessing the corrected single pollutant concentration contribution value, real-time monitoring data of various pollutant monitoring points, topographic data, and total pollution source concentration contribution of the target site to obtain fourth input data; A third node feature matrix is ​​constructed according to the fourth input data, and the third node feature matrix and graph structure information are input into a pre-built third spatiotemporal deep learning model DCRNN to obtain the contribution weight of the pollution source.

5. The method according to claim 2, characterized in that: The obtaining of graph structure information includes: The graph structure information is obtained according to the spatial distribution of the total concentration contribution of the pollution sources of the target site within a preset range of the target site.

6. The method according to claim 5, characterized in that The first spatiotemporal deep learning model DCRNN, the second spatiotemporal deep learning model DCRNN and the third spatiotemporal deep learning model DCRNN all include: a graph convolution module, a recurrent neural network module and a decoding module; The graph convolution module is used to extract the spatial features of the input features, the recurrent neural network module is used to extract the temporal features of the input features, and the decoding module is used to fuse the extracted spatial features and temporal features.

7. A multi-stage optimized space-time tracing device, characterized in that: The device comprises: The first data acquisition module is used to acquire meteorological data of a specified historical period within a preset range of the target site, acquire topographic data within a preset range of the target site, and acquire monitoring data of various pollutant monitoring points within a specified historical period within the preset range of the target site; The first input module is used to pre-process the meteorological data, topographic data and monitoring data of various pollutant monitoring points to obtain first input data; AERMOD module: used to input the first input data into the pre-built AERMOD model to obtain the total concentration contribution of the pollution source of the target site; The second data acquisition module is used to obtain real-time monitoring data of various pollutant monitoring points, and obtain the initial weight of the contribution of the pollution source according to the proportion of each pollution source in the real-time monitoring data of various pollutant monitoring points; The first deep learning module is used to obtain the corrected total concentration contribution of pollution sources at the target site through a pre-built first spatiotemporal deep learning model DCRNN based on the total concentration contribution of pollution sources at the target site, real-time monitoring data of various pollutant monitoring points, topographic data, and initial weights of pollution source contributions; The second deep learning module is used to obtain the corrected single pollutant concentration contribution value through the pre-built second spatiotemporal deep learning model DCRNN based on the total pollution source concentration contribution of the corrected target site, the real-time monitoring data of various pollutant monitoring points, the real-time meteorological data and the initial weight of the pollution source contribution; The third deep learning module is used to obtain the contribution weight of the pollution source through the pre-built third spatiotemporal deep learning model DCRNN based on the corrected single pollutant concentration contribution value, the real-time monitoring data of various pollutant monitoring points, the topographic data and the total concentration contribution of the pollution source of the target site; Circular update module: used to update the initial weight of the contribution of the pollution source through the contribution weight of the pollution source, and obtain the corrected single pollutant concentration contribution value through the updated contribution weight of the pollution source.

8. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the main controller, each step in the multi-stage optimized spatiotemporal tracing method as described in any one of claims 1-6 is implemented.

Images