Mesoscale atmospheric pollutant transmission and migration rapid evaluation method based on convolutional neural network

Abstract

The invention relates to a mesoscale atmospheric pollutant transmission and migration rapid evaluation method based on a convolutional neural network, and solves the technical problems of high requirements on computing resources, long computing time and the like in the prior art. The method comprises the following steps: 1, acquiring mesoscale meteorological original data and static topographic data; 2, generating key meteorological element spatial distribution characteristic data; 3, acquiring source item parameters of the simulation area and performing normalization processing; 4, acquiring three-dimensional space characteristic data of the pollutants by means of a mesoscale air quality mode; 5, generating a sample data set; 6, normalizing the sample data set to obtain a time sequence data set; 7, extracting corresponding features as input parameters according to the time sequence data set; 8, building a convolutional neural network model and setting a loss function value for model training; 9, judging a training result; 10, extracting the trained convolutional neural network model, and designing a corresponding interface to generate a rapid evaluation model; and 11, evaluating by using the model.

Description

Technical Field

[0001] The present invention relates to a method for simulating the diffusion of air pollution, and more particularly to a method for rapidly evaluating the mesoscale diffusion and migration of air pollution based on a convolutional neural network-long short-term memory. Background Art

[0002] The study of the diffusion and migration of pollutants in the atmospheric environment is of great significance to research fields such as emergency prediction, chemical process analysis, air pollution prevention, and even infectious disease prevention. The transport of pollutants in the atmospheric environment is largely affected by atmospheric turbulence, terrain elevation, and the type of underlying surface. Therefore, it is a typical mesoscale problem strongly coupled with three-dimensional complex meteorological and terrain parameters. How to perform mesoscale simulation and prediction of the diffusion and migration of atmospheric pollutants is of crucial importance for the response decision-making of emergencies and the pre-analysis of physical processes in the above research fields.

[0003] At present, research scholars have developed typical mesoscale assessment models to simulate the transport characteristics of pollutants under complex source terms, meteorological, and terrain conditions. Most of the existing assessment models adopt traditional numerical simulation methods based on mathematical and physical equations. Usually, the Euler method is used to discretize the entire atmospheric space into grids, and the corresponding numerical solution methods are used to perform discretized numerical solutions for the entire simulation area. Although this assessment model has a strong ability to finely describe physical parameters in three dimensions, the Euler model consumes a large amount of computing resources. The required computing time is usually several hours, and there are certain limitations in the convergence of numerical calculations. Therefore, it does not meet the rapid assessment requirements for the atmospheric transport process of pollutants under emergencies.

[0004] For example, Ulas Im, Kostandinos Markakis, Alper Unal et al. published "Study of a winter PM episode in Istanbul using the high resolution WRF / CMAQ modeling system." (Atmospheric Environment, 44, 3085 - 3094) in 2010. This literature mainly simulated the concentration distribution characteristics of winter particulate matter in Istanbul based on the WRF-CMAQ mesoscale model.

[0005] Another example is that V. Simsek, L. Pozzoli, A. Unal, T. Kindap, M. Karaca. et al. published "Simulation of 137"Cs transport and deposition after the chernobyl nuclear power plant accident and radiological doses over the Anatolian peninsula." (Sci. Total Environ., 499 (2014), pp. 74 - 88.), Zhenhui Ma, Tengyue Ma, Baosheng Wang et al. published "Meso-scale numerical analysis for transport and deposition behaviors of radioactive aerosols under severe nuclear accident" (Progress in Nuclear Energy, 150, 104314.) in 2022, and Zhenhui Ma, Zhiming Li, Xiuhuan Tang et al. published "Meso-scale numerical analysis for transport and deposition behaviors of radioactive aerosols under severe nuclear accident" (Frontiers in Environmental Science, 12:1455273.) in 2024. These three literatures mainly focus on the assessment of the diffusion and migration of harmful substances in energy facilities under accident conditions.

[0006] The meso-scale methods for the above assessments can all achieve the assessment of the transport and migration of atmospheric pollutants to a certain extent. However, they all involve three-dimensional grid discretization and the solution of the N - S equations based on the Euler method, which require high computing resources and long computing time, and cannot quickly give the calculation results in a short time. Therefore, they cannot be used for the rapid assessment of the distribution and migration of harmful substances in scenarios such as accident emergency. Summary of the Invention

[0007] The purpose of the present invention is to solve the technical problems of the existing assessment methods, such as high requirements for computing resources and long computing time, and to provide a rapid assessment method for meso-scale atmospheric pollutant transport and migration based on a convolutional neural network.

[0008] To achieve the above purpose, the technical solution provided by the present invention is as follows:

[0009] A rapid evaluation method for mesoscale atmospheric pollutant transport and migration based on a convolutional neural network, which is characterized in that it includes the following steps:

[0010] [1], respectively obtain the mesoscale meteorological original data and static terrain data of the simulation area; the mesoscale meteorological original data includes the original meteorological reanalysis data and the assimilation data of the simulation area;

[0011] [2], based on the mesoscale meteorological original data and static terrain data obtained in step [1], rely on the mesoscale advanced meteorology model to conduct continuous numerical simulations to generate the spatial distribution characteristic data of the key meteorological elements in the simulation area;

[0012] [3], obtain the source term parameters of the simulation area, perform normalization processing on them, and then set the emission condition of the source term to hourly emission and the emission amount to unit intensity;

[0013] [4], combine the spatial distribution characteristic data of the key meteorological elements generated in step [2] and the source term parameters after normalization processing in step [3], rely on the mesoscale air quality model to conduct mesoscale atmospheric diffusion numerical simulations, so as to obtain the three-dimensional spatial characteristic data of pollutants in the simulation area under the condition of hourly emission of the normalized source term;

[0014] [5], respectively perform data annotation and data alignment on the spatial distribution characteristic data of the key meteorological elements generated in step [2] and the three-dimensional spatial characteristic data of pollutants obtained in step [4] according to different time steps, so as to generate a sample data set;

[0015] [6], perform normalization processing on the sample data set generated in step [5], and then organize it in chronological order to obtain a time series data set;

[0016] [7], respectively extract the time series features and spatial features in the time series data set, and then use them as input parameters, and divide the input parameters into a training set and a test set in chronological order;

[0017] [8], build a convolutional neural network model and set the loss function value, then input the training set described in step [7] into the convolutional neural network model for model training, and then output the training result;

[0018] [9], judge whether the loss function value in the training result described in step [8] is consistent with the set loss function value. If not, combine the test set to perform iterative optimization on the convolutional neural network model in step [8] and then continue with model training; if it is consistent, execute step

[10] ;

[0019]

[10] Extract the trained convolutional neural network model, design corresponding meteorological data interfaces, terrain data interfaces, and output data interfaces, and generate a rapid assessment model for mesoscale atmospheric pollutant transport and migration;

[0020]

[11] Input the actual parameters of the simulation area into the rapid assessment model for mesoscale atmospheric pollutant transport and migration in step

[10] , and then complete the rapid assessment of mesoscale atmospheric pollutant transport and migration based on the convolutional neural network.

[0021] Furthermore, in step 【2】, the continuous numerical simulation refers to configuring the horizontal grid of the simulation area in a three-layer grid nesting manner, and using the innermost grid as the final simulation area to carry out mesoscale meteorological simulations for 1 to 3 years;

[0022] The key meteorological elements include temperature, humidity, wind speed, wind direction, atmospheric pressure, precipitation, and boundary layer height.

[0023] Furthermore, in step 【2】, the horizontal grid resolution of the innermost grid is 1 to 3 km.

[0024] Furthermore, in step 【5】, the sample data set includes multiple independent sample data groups, and each sample data group contains meteorological emission information and pollutant concentration information.

[0025] Furthermore, in step 【6】, the organization in chronological order means organizing the normalized sample data set in chronological order using the sliding window method.

[0026] Furthermore, in step 【6】, the time series data set includes multiple time series data groups with different starting times and equal simulation durations.

[0027] Furthermore, in step 【8】, the training refers to first performing multi-layer stacked long short-term memory processing on the training set, and then performing three-dimensional convolution operations.

[0028] Furthermore, in step 【8】, the loss function value is the sum of the squares of the absolute values of the concentration differences between the training output results and the traditional simulation results for all grid points in the computational domain.

[0029] Furthermore, in step 【9】, the iterative optimization means using the loss function value as the iterative index for model training, and adjusting the hyperparameters according to the iterative results.

[0030] Furthermore, step

[11] also includes the step of evaluating the evaluation model and the evaluation results using statistical indicators after completing the evaluation.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The rapid assessment method for mesoscale atmospheric pollutant transport and migration based on convolutional neural network provided by the present invention is based on the mesoscale meteorological raw data and static terrain data of the simulation area. Relying on the mesoscale advanced meteorology model and mesoscale air quality model, the three-dimensional spatial characteristic data of pollutants in the simulation area is obtained. Then, this data is used as input parameters to be trained by the convolutional neural network to obtain a rapid assessment model for mesoscale atmospheric pollutant transport and migration, thereby achieving rapid assessment. The present invention can give full play to the advantages of the long short-term memory architecture of the convolutional neural network, fully capture the long-term dependence relationship in time during the pollutant diffusion process, not only improve the prediction efficiency, but also improve the prediction accuracy of pollutants.

[0033] 2. The rapid assessment method for mesoscale atmospheric pollutant transport and migration based on convolutional neural network provided by the present invention can, without the need to rely on high-performance server computing, use a single machine to complete the rapid evaluation of the three-dimensional distribution trend and evolution law of pollutants in the next few days for emergencies within a short time (a few minutes), so as to meet the actual needs of rapid simulation and evaluation for future multi-accident scenarios.

[0034] 3. The rapid assessment method for mesoscale atmospheric pollutant transport and migration based on convolutional neural network provided by the present invention has a coupling interface with the industry-wide common GIS geographic information system and three-dimensional meteorological reanalysis data, providing convenient conditions for the operational operation of the method and model.

[0035] 4. The rapid assessment method for mesoscale atmospheric pollutant transport and migration based on convolutional neural network provided by the present invention can be deployed on a server or cloud platform / quasi-cloud platform in the future, so it supports rapid rolling forecasting of pollutants in real-time or near real-time for multiple scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a schematic flow chart of an embodiment of the present invention;

[0037] Figure 2 is a schematic structural diagram of the convolutional neural network model in an embodiment of the present invention;

[0038] Figure 3 is a schematic diagram of the data flow structure of the convolutional neural network model in an embodiment of the present invention;

[0039] Figure 4 is a comparison chart of the preliminary prediction results between an embodiment of the present invention and the traditional numerical simulation method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] To make the objectives, advantages, and features of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention, and the purpose is not to limit the protection scope of the present invention.

[0041] With the rapid development of artificial intelligence technology, emerging artificial intelligence learning technologies such as convolutional neural networks and long short-term memory have gradually merged and interacted with traditional theoretical research fields. Such technologies are mainly used to find the correlations between different data, rather than exploring the causal relationships through traditional numerical simulation methods. This makes it possible for the solution of mesoscale problems to deviate from traditional mathematical and physical algorithms, and to a certain extent, promotes the development of rapid evaluation technologies. The present invention combines a convolutional neural network-long short-term memory architecture with a traditional mesoscale meteorological-atmospheric transport model, and generates a rapid evaluation model for mesoscale air pollution diffusion and migration through processes such as sample data generation, processing, model optimization, and iteration, to meet the needs of rapid evaluation of pollutant atmospheric transport and consequences in fields including but not limited to the atmospheric environment, chemical industry, and health disease prevention.

[0042] As Figure 1 shown, this embodiment provides a rapid evaluation method for mesoscale atmospheric pollutant transport and migration based on a convolutional neural network, including the following steps:

[0043] 【1】. Obtain the mesoscale meteorological raw data and static terrain data of the simulation area respectively. The mesoscale meteorological raw data includes raw meteorological reanalysis data and assimilation data of the simulation area.

[0044] In this embodiment, the source of the reanalysis data can be meteorological reanalysis data released by the National Centers for Environmental Prediction (NCEP) of the United States, the European Centre for Medium-Range Weather Forecasts (ECMWF), or the China Meteorological Administration (CMA). The assimilation data can use surface / sounding data or actual observation data released by official agencies.

[0045] 【2】. Based on the mesoscale meteorological raw data and static terrain data obtained in step 【1】, rely on the mesoscale advanced meteorology model to conduct continuous numerical simulations to generate data on the spatial distribution characteristics of key meteorological elements in the simulation area. The key meteorological elements include temperature, humidity, wind speed, wind direction, atmospheric pressure, precipitation, and boundary layer height, and the time resolution of the simulation results of these key meteorological elements is hourly.

[0046] In this embodiment, continuous numerical simulation means that the horizontal grid of the simulation area is configured in a nested manner with three layers of grids, and a 1-3-year mesoscale meteorological simulation is carried out with the innermost grid as the final simulation area. The horizontal grid resolution of the innermost grid is generally 1-3 km.

[0047] The continuous simulation results of the three-dimensional meteorological field in the entire simulation area for 1 to 3 years provide meteorological driving for the subsequent mesoscale atmospheric diffusion simulation. The simulation objects of mesoscale air pollution diffusion include gases and aerosols. For gases, the three-dimensional spatial concentration data of the main pollutants are mainly simulated. For aerosol pollutants, in addition to the spatial concentration distribution data, the settlement calculation data on the ground surface also need to be obtained.

[0048] For aerosol pollutants, according to the aerodynamic distribution characteristics of aerosols with different particle sizes, aerosols with different particle sizes can be classified and discussed: taking 10 μm as the boundary, they are divided into two groups of large particle size aerosols and small particle size aerosols. Generally, small particle size aerosols follow a lognormal distribution, so they can be considered as a continuous spectrum; for large particle size aerosols, particle size grouping can be carried out, and each particle size group has a typical average particle size, which is described by unified particle size and related aerodynamic physical property parameters.

[0049] 【3】 Obtain the source term parameters of the simulation area, normalize them, and then set the emission conditions of the source term to hourly emission with the emission amount being the unit intensity.

[0050] In this embodiment, the source term parameters used in the mesoscale air quality model need to be normalized in terms of source strength and time (it is considered that the pollutants are only emitted for one hour according to the unit amount). Then, according to the continuous numerical simulation results for 1 to 3 years in step 【2】, the normalized source term data are added hourly in the mesoscale air quality model according to the corresponding time series. The pollutant concentration and settlement simulation data corresponding to each normalized source term are statistically output in the model for 7 days after release.

[0051] 【4】 Combine the key meteorological element spatial distribution characteristic data generated in step 【2】 and the source term parameters after normalization in step 【3】, and rely on the mesoscale air quality model to conduct mesoscale atmospheric diffusion numerical simulation on them, so as to obtain the three-dimensional spatial characteristic data of pollutants in the simulation area under the condition of hourly emission of the normalized source term.

[0052] Among them, the mesoscale atmospheric diffusion numerical simulation can be carried out by using the simulation method disclosed in the publication number CN114547890A, or other methods that can achieve such results can be used for numerical simulation.

[0053] 【5】 Label and align the key meteorological element spatial distribution characteristic data generated in step 【2】 and the three-dimensional spatial characteristic data of pollutants obtained in step 【4】 according to different time steps respectively, so as to generate a sample data set that meets the requirements of the convolutional neural network-long short-term memory architecture.

[0054] The sample data set includes multiple independent sample data groups, and each sample data group contains complete input information (i.e., meteorological data, source term data meteorological emissions that have been normalized and arranged hourly in time series) and output information (i.e., pollutant concentration and deposition information).

[0055] 【6】. Normalize the sample data set generated in step 【5】, and then organize it in chronological order to obtain a time series data set, so as to improve the stability of the subsequent constructed convolutional neural network-long short-term memory architecture.

[0056] In this embodiment, organizing in chronological order means organizing the normalized sample data set in chronological order using the sliding window method, that is, keeping the time length of each group of sample data groups unchanged, and the starting time changes hourly to generate multiple time series data groups with different starting times but equal simulation durations, so as to obtain a time series data set suitable for the long short-term memory architecture.

[0057] 【7】. Extract the time series features and spatial features in the time series data set respectively, use them as the input parameters of the subsequent convolutional neural network-long short-term memory architecture, and divide the input parameters into a training set and a test set in chronological order, so as to ensure that the established rapid evaluation method has strong generalization ability for different cases and scenarios.

[0058] The input parameters include meteorological input feature parameters and emission feature input parameters. Among them, the meteorological input feature parameters mainly include wind speed, wind direction, precipitation and boundary layer height, and the emission feature input parameters are mainly the longitude and latitude positions, heights of emission sources and the surrounding terrain information.

[0059] 【8】. As Figure 2 shown, build a convolutional neural network model and set the loss function value, then input the training set described in step 【7】 into the convolutional neural network model for model training, and then output the training result.

[0060] The convolutional neural network model in this embodiment is a convolutional neural network-long short-term memory unit structure, which combines the characteristics of convolutional neural network and long short-term memory network, and mainly includes the following components: the input X at the current time step t , the hidden state H at the previous time step t-1 and the cell state C t-1 . This unit extracts spatial features through convolutional operations and includes four main gating mechanisms: the forget gate F t determines how much information of the previous cell state is retained, the input gate I t controls the writing of the current input information, the candidate cell state generates a new candidate state through the tanh function, and the output gate O tOutput that determines the current cell state. Cell state C t is updated by combining the previous state and the current candidate state, while the hidden state H t is generated based on the updated cell state and is implemented according to an activation function, which is used to introduce non-linearity, i.e.: its output range is 0 to 1; its output range is -1 to 1.

[0061] This design enables the convolutional neural network-long short-term memory unit structure to effectively capture temporal data and spatial information.

[0062] The training in this embodiment refers to first performing multi-layer stacked long short-term memory processing on the training set, and then feeding the data back to the output layer of the above architecture through three-dimensional convolution operations. The loss function value is set to the sum of the squares of the absolute values of the concentration differences between the training output results and the traditional simulation results at all grid points in the computational domain. The output training results are the output parameters of the sample data, including the three-dimensional concentration distribution and surface settlement distribution of gaseous and aerosol pollutants at future times.

[0063] Figure 3 FIG. 9 is a schematic diagram of a data flow structure based on a convolutional neural network-long short-term memory. The input in this data flow structure diagram includes the wind speed in the u direction, the wind speed in the v direction, the boundary layer height, and emission point information. Among them, concn represents the predicted concentration value at the nth moment, uvn represents the wind vector parameter at the nth moment, and pbln represents the boundary layer height at the nth moment. The first layer of the convolutional neural network-long short-term memory unit of the network receives the above input data and processes it through the convolutional neural network-long short-term memory unit to capture complex temporal features and spatial information. Subsequently, the subsequent layers further refine the features to better focus on the key temporal information. Finally, the output of the convolutional neural network-long short-term memory unit will be passed to a three-dimensional convolutional layer, which is responsible for further processing the extracted features and converting them into the final output, which is a single-channel result representing the predicted concentration.

[0064] 【9】 Judge whether the loss function value in the training result described in step 【8】 matches the set loss function value. If it matches, execute step

[10] ; if it does not match, iteratively optimize the convolutional neural network model in step 【8】 in combination with the test set and then continue with model training to make the comprehensive evaluation index including the loss function meet the actual requirements of the evaluation. The iterative optimization described here refers to using the loss function value as the iterative index for model training and tuning the hyperparameters according to the iterative results. The specific hyperparameters that need to be tuned include but are not limited to the learning rate, batch size, number of long short-term memory layers, and number of units, etc.

[0065]

[10] Extract the trained convolutional neural network model, design corresponding meteorological data interfaces, terrain data interfaces, and output data interfaces, and then generate a rapid assessment model for mesoscale atmospheric pollutant transport and migration.

[0066]

[11] Input the actual parameters of the simulation area (including source term parameters, geographic information data, meteorological reanalysis data, etc. of the actual simulation case) into the rapid assessment model for mesoscale atmospheric pollutant transport and migration in step

[10] , and then complete the rapid assessment of mesoscale atmospheric pollutant transport and migration based on the convolutional neural network.

[0067] Figure 4 It is a comparison chart of the preliminary prediction results between the prediction method of this embodiment and the traditional numerical simulation method. The left figure is the concentration diffusion simulation result of the traditional numerical simulation method, and the right figure is the concentration diffusion simulation result generated by the rapid assessment method based on the convolutional neural network-long short-term memory described in the present invention under the same conditions. It can be seen from this figure that the concentration diffusion simulation results generated by the rapid assessment method are basically consistent with the results of the traditional simulation method in terms of diffusion trend, diffusion range, and concentration characteristics, thus verifying the rationality of this method.

[0068] In addition, statistical indicators can be used to further evaluate the evaluation model and evaluation results after the evaluation is completed. The statistical indicators for evaluation include but are not limited to mean square error, root mean square error, mean absolute error, and correlation coefficient.

[0069] The rapid assessment method for mesoscale atmospheric pollutant transport and migration based on the convolutional neural network provided by the present invention can deeply consider the influence of atmospheric turbulence motion and vertical changes on pollutant transport, and deeply integrate sudden source terms, complex three-dimensional meteorological fields, and complex terrain / surface conditions, so as to realize the numerical simulation of mesoscale three-dimensional atmospheric diffusion under the influence of multiple factors. At the same time, without the need to rely on high-performance server computing, it can use a single machine to quickly evaluate the three-dimensional distribution trend and evolution law of pollutants in the next few days for emergencies within a short time (a few minutes), so as to meet the actual needs of future rapid simulation evaluation of multiple accident scenarios, and in the future, the model can be deployed on a server or a cloud platform / quasi-cloud platform, so as to support real-time or near-real-time rapid rolling forecasting of pollutants in multiple scenarios.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural networks, characterized in that: The following steps are involved: 【1】Obtain the mesoscale meteorological raw data and static terrain data of the simulation area respectively; The mesoscale meteorological raw data include original meteorological reanalysis data and assimilated data of the simulation area; 【2】Based on mesoscale meteorological raw data and static terrain data, continuous numerical simulation is carried out with the help of mesoscale advanced meteorological models to generate spatial distribution characteristic data of key meteorological elements in the simulation area; 【3】, obtain the source term parameters of the simulation area, normalize them, and then set the emission conditions of the source term to hourly emission and unit intensity; 【4】Combining the spatial distribution characteristic data of key meteorological elements generated in step 【2】 and the normalized source item parameters, numerical simulation of mesoscale atmospheric diffusion is performed based on the mesoscale air quality model, so as to obtain the three-dimensional spatial characteristic data of pollutants in the simulation area under the normalized source item hourly emission conditions; [5] The key meteorological element spatial distribution characteristic data and pollutant three-dimensional spatial characteristic data generated in step [2] are labeled and aligned according to different time steps to generate a sample data set; 【6】Normalize the sample data set and then organize it in chronological order to obtain a time series data set; 【7】Extract the temporal features and spatial features from the time series data set respectively, use them as input parameters, and divide the input parameters into a training set and a test set in chronological order; 【8】Build a convolutional neural network model and set the loss function value, then input the training set into the convolutional neural network model for model training, and then output the training results; [9], determine whether the loss function value in the training result is consistent with the set loss function value. If not, iteratively optimize the convolutional neural network model based on the test set before continuing model training; if consistent, execute step [10]; 【10】Extract the trained convolutional neural network model, and design the corresponding meteorological data interface, terrain data interface and output data interface to generate a rapid assessment model for the transport and migration of mesoscale atmospheric pollutants; 【11】The actual parameters of the simulation area are input into the rapid assessment model of mesoscale atmospheric pollutant transport and migration, and then the rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network is completed.

2. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1 is characterized in that: In step [2], the continuous numerical simulation refers to constructing the horizontal grid of the simulation area by nesting three layers of grids, and using the innermost grid as the final simulation area to carry out 1-3 years of mesoscale meteorological simulation; The key meteorological elements include temperature, humidity, wind speed, wind direction, atmospheric pressure, precipitation and boundary layer height.

3. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 2 is characterized in that: In step [2], the horizontal grid resolution of the innermost grid is 1 to 3 km.

4. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1 is characterized in that: In step [5], the sample data set includes multiple independent sample data groups, and each sample data group contains meteorological emission information and pollutant concentration information.

5. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1 is characterized in that: In step [6], the organization in chronological order refers to organizing the normalized sample data set in chronological order using a sliding window method.

6. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 5 is characterized in that: In step [6], the time series data set includes multiple time series data sets with different starting times and simulation durations.

7. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1 is characterized in that: In step [8], the training refers to first performing multi-layer stacked long short-term memory processing on the training set and then performing a three-dimensional convolution operation.

8. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 7 is characterized in that: In step [8], the loss function value is the sum of the squares of the absolute values ​​of the concentration differences between the training output results and the traditional simulation results in all grid points in the calculation domain.

9. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1, characterized in that: In step [9], the iterative optimization refers to using the loss function value as the iterative indicator of model training and tuning the hyperparameters according to the iterative results.

10. The method for rapid assessment of mesoscale atmospheric pollutant transport and migration based on convolutional neural network according to claim 1, characterized in that: Step [11] also includes the step of using statistical indicators to evaluate the evaluation model and evaluation results after the evaluation is completed.

Images