A single variable pollutant emission source inversion method, device, equipment and medium

Abstract

The present application relates to a method, device, equipment and medium for inverting single-variable pollutant emission sources. The method combines the advantages of variational assimilation and deep learning. After performing three-dimensional variational assimilation on the pollutant concentration observation data to obtain a more accurate pollutant concentration reanalysis field, a trained emission inversion model with prediction function is used to perform inversion prediction of pollutant emission sources to obtain a more accurate pollutant emission inventory. The use of this method can meet the demand for high-time-resolution emission inventories for atmospheric pollution forecasting and control, and significantly improves computational efficiency. In addition, the pollutant emission inventory obtained based on this method can also be used to improve the pollutant concentration forecast level of atmospheric chemistry models, providing strong support for atmospheric pollution prevention, control, governance, forecasting and early warning.

Description

Technical Field

[0001] The present application relates to the field of atmospheric environment technology, and in particular to a method, device, equipment and medium for inverting a single variable pollutant emission source. Background Art

[0002] Accurate air pollution emission inventories are a prerequisite for air pollution control. Typical emissions inventories are compiled using a bottom-up approach, but this process relies on a large amount of emission statistics. The subsequent assumption of hourly factors to differentiate emissions over time further increases errors, making updates difficult and subject to considerable uncertainty.

[0003] Currently, the main top-down methods used to constrain bottom-up emission estimates are ensemble Kalman filtering and four-dimensional variational assimilation. However, these methods are complex to model and require extensive computational resources, limiting their practical application. In recent years, deep learning methods have been applied to air pollution forecasting, achieving promising results. However, these methods typically generate average emissions on a monthly or daily timescale, limiting their applicability for estimating emissions during short-term pollution events. Therefore, an inversion method that can provide higher temporal resolution for emission inventories is urgently needed. Summary of the Invention

[0004] Based on this, it is necessary to provide a single-variable pollutant emission source inversion method, device, equipment and medium to address the above technical problems, the purpose of which is to achieve high-time-resolution emission source inversion on the basis of improving calculation efficiency and accuracy.

[0005] A single variable pollutant emission source inversion method, the method comprising:

[0006] Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area;

[0007] The enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0008] Using atmospheric chemistry models and disturbed emission source data, hourly forward simulations of the meteorological field and pollutant concentration field in the study area were performed to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0009] The meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0010] The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field;

[0011] The meteorological element dataset, pollutant concentration reanalysis field and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0012] In one embodiment, after obtaining single-variable pollutant concentration observation data with short life cycle characteristics, meteorological reanalysis grid data, and prior emission source data in the study area, the method further includes:

[0013] The pollutant concentration observation data are quality controlled and averaged according to the study area and grid resolution of the WRF-Chem model.

[0014] In one embodiment, the enhanced disturbance strategy, the weakened disturbance strategy, and the cross-level adjustment disturbance strategy are respectively used to perturb the prior emission source data to obtain the disturbed emission source data, including:

[0015] According to the actual increase, decrease and normal emission ratio of pollution sources, the grid ratio of increased emission, decreased emission and normal emission per hour in the study area is set;

[0016] Create a set of random numbers that conform to the normal distribution and have the same dimension as the prior emission source data, perform Gaussian filtering on the random numbers, and perform initial perturbations by multiplying the emissions in each type of grid by random numbers that obey different means and standard deviations and conform to the normal distribution. This yields perturbed emissions in enhanced emission grids, weakened emission grids, and normal emission grids.

[0017] An enhanced perturbation strategy is adopted, that is, based on the initial perturbation, the perturbed emissions in all grids are multiplied by a random number greater than 1 generated from a uniform distribution to obtain an enhanced emission inventory for simulating emission source enhancement;

[0018] A disturbance reduction strategy is adopted, that is, based on the initial disturbance, the disturbance emissions in all grids are multiplied by a random number less than 1 generated from a uniform distribution to obtain a reduced emission inventory for simulating emission source reduction;

[0019] A cross-level adjustment disturbance strategy is adopted, that is, before the initial disturbance, the emission at each moment in the prior emission source data is adjusted by adding a random number in a set interval. At the same time, the emission is divided into 8 intervals according to the preset threshold, and the emission in each interval is subjected to cross-level adjustment disturbance. Then, the emission in each grid is multiplied by a random number that obeys different means and standard deviations and conforms to the normal distribution, so as to obtain an emission inventory in which the emission is controlled within the preset range and is used to simulate the migration of existing emission sources and the formation of new emission sources.

[0020] In one embodiment, the initial perturbation is performed by multiplying the emissions in each type of grid by random numbers that have different means and standard deviations and conform to normal distribution, including:

[0021] The emissions within the enhanced emission grid are initially perturbed by multiplying them by a random number with a mean greater than 1, the emissions within the weakened emission grid are initially perturbed by multiplying them by a random number with a mean less than 1, and the emissions within the normal emission grid are initially perturbed by multiplying them by a random number with a mean equal to 1.

[0022] In one embodiment, performing cross-level adjustment disturbance on the emissions in each interval includes:

[0023] Two types of grids are set in the study area. For one type of grid, a random number is directly added to the emissions within the grid to cause disturbance, so that the emission value after disturbance changes in magnitude. For the other type of grid, a random number is multiplied by the emissions within the grid to cause disturbance.

[0024] When disturbing emissions in different intervals, the situations at different times are taken into consideration. Emissions are reduced when the prior emissions are overestimated, and new emission sources are added when the prior emissions are underestimated.

[0025] In one embodiment, an atmospheric chemistry model and disturbed emission source data are used to perform hourly forward simulations of the meteorological field and pollutant concentration field in the study area to obtain hourly resolution meteorological element datasets and pollutant concentration datasets, including:

[0026] Based on the prior emission source data, hourly forward simulations were performed during the study period. The simulation time window was set to 1 hour, and the hourly simulation results were used as the background field.

[0027] Taking the background field at the initial moment of the study as the initial field, the atmospheric chemistry model and the emission inventory obtained by using three perturbation strategies are used to perform hourly forward simulations of the meteorological field and pollutant concentration field in the study area to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0028] In one embodiment, a meteorological element dataset, a pollutant concentration dataset, and prior emission source data are input into a pre-built emission inversion model based on deep learning for training, thereby obtaining a trained emission inversion model, including:

[0029] The hourly-resolution meteorological element dataset, pollutant concentration dataset and prior emission source data obtained by forward simulation are used as input features and input into the deep learning-based emission inversion model for training. The training target is defined as the difference in hourly emissions between the disturbed emission source data and the prior emission source data, that is, the emission increment. The model inversion accuracy is evaluated by comparing the difference between the model-predicted emissions and the disturbed emission source data until a trained emission inversion model that meets the preset inversion accuracy is obtained.

[0030] A single variable pollutant emission source inversion device, the device comprising:

[0031] The data acquisition module is used to obtain single-variable pollutant concentration observation data with short life cycle characteristics in the study area, meteorological reanalysis grid data and prior emission source data;

[0032] A disturbance module is used to disturb the prior emission source data by adopting an enhanced disturbance strategy, a weakened disturbance strategy, and a cross-level adjustment disturbance strategy to obtain disturbed emission source data;

[0033] The forward simulation module is used to perform hourly forward simulation of the meteorological field and pollutant concentration field in the study area using atmospheric chemistry models and disturbed emission source data, and obtain hourly resolution meteorological element datasets and pollutant concentration datasets;

[0034] The model training module is used to input meteorological element datasets, pollutant concentration datasets, and prior emission source data into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0035] The variational assimilation module is used to assimilate the hourly pollutant concentration observation data in the study area using a three-dimensional variational assimilation method to generate a pollutant concentration reanalysis field;

[0036] The emission inversion module is used to input the meteorological element dataset, pollutant concentration reanalysis field and prior emission source data into the trained emission inversion model for prediction, obtain the emission increment field, and obtain the pollutant emission inventory with hourly resolution in the study area by superimposing the emission increment field onto the prior emission field.

[0037] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0038] Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area;

[0039] The enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0040] Using atmospheric chemistry models and disturbed emission source data, hourly forward simulations of the meteorological field and pollutant concentration field in the study area were performed to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0041] The meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0042] The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field;

[0043] The meteorological element dataset, pollutant concentration reanalysis field and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0044] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0045] Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area;

[0046] The enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0047] Using atmospheric chemistry models and disturbed emission source data, hourly forward simulations of the meteorological field and pollutant concentration field in the study area were performed to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0048] The meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0049] The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field;

[0050] The meteorological element dataset, pollutant concentration reanalysis field and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0051] The above-mentioned single-variable pollutant emission source inversion method, device, equipment and medium combine the advantages of variational assimilation and deep learning. After performing three-dimensional variational assimilation on the pollutant concentration observation data to obtain a more accurate pollutant concentration reanalysis field, the trained emission inversion model with prediction function is used to perform pollutant emission source inversion prediction to obtain a more accurate pollutant emission inventory. Compared with the existing technology, the emission inversion model trained by this application can predict pollutant emissions with hourly resolution, which greatly reduces the consumption of computing resources and shortens the computing time, so that this application has the ability to solve the emission estimation problem in short-term pollution events. It can not only meet the demand for high-time-resolution emission inventories for atmospheric pollution forecasting and control, but also significantly improve computing efficiency. In addition, the pollutant emission inventory obtained based on this method can also be used to improve the pollutant concentration forecast level of atmospheric chemistry models, providing strong support for atmospheric pollution prevention and control and forecasting and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 1 is a flow chart of a single variable pollutant emission source inversion method in one embodiment;

[0053] Figure 2 A schematic diagram of the correlation between SO2 emissions predicted by an emission inversion model and “real” emissions in one embodiment;

[0054] Figure 3 Schematic diagram of the interpolation of SO2 concentrations to ground stations and the observed concentrations at the stations at 17:00 UTC (Universal Standard Time) on January 21, 2023, respectively, based on the MEIC_2016 emission inventory and the inversion of this application in one embodiment;

[0055] Figure 4 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0057] In one embodiment, Figure 1As shown, a single variable pollutant emission source inversion method is provided, including the following steps:

[0058] Step 1: Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area.

[0059] Specifically, after obtaining the single-variable pollutant concentration observation data with short life cycle characteristics in the study area, meteorological reanalysis grid data and prior emission source data, it is also necessary to perform quality control on the pollutant concentration observation data and average the pollutant concentration observation data falling within the same grid according to the study area and grid resolution of the WRF-Chem model.

[0060] Quality control effectively ensures the accuracy and reliability of observational data, more accurately reflecting actual pollutant concentrations within the study area and laying a solid foundation for subsequent data analysis and forecasting. The WRF-Chem model, developed by the NOAA (National Oceanic and Atmospheric Administration) Forecast Systems Laboratory (FSL), is a regional air quality model that online couples a meteorological model (WRF) with a chemical model (Chem). Averaging the observed data using the WRF-Chem model helps simplify the data structure and reduce complexity and dispersion.

[0061] In step 2, the enhanced disturbance strategy, the weakened disturbance strategy and the cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0062] Among them, the enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used for the enhancement, weakening, addition and migration of emission sources respectively, which helps to provide a more comprehensive and realistic emission source simulation. Specifically, three different disturbances are performed on the prior emission source data, including the following steps:

[0063] First, based on the actual ratio of emission increase, emission reduction and normal emission of pollution sources, the grid ratio of enhanced emission, weakened emission and normal emission per hour in the study area is set.

[0064] Then, a set of random numbers that conform to a normal distribution and have the same dimensions as the prior emission source data is created. The random numbers are Gaussian filtered, and the emissions within each type of grid are initially perturbed by multiplying them by random numbers that follow a normal distribution with different means and standard deviations. This yields the perturbed emissions within the enhanced emission grid, the weakened emission grid, and the normal emission grid. The emissions within the enhanced emission grid are initially perturbed by multiplying them by a random number with a mean greater than 1, the emissions within the weakened emission grid are initially perturbed by multiplying them by a random number with a mean less than 1, and the emissions within the normal emission grid are initially perturbed by multiplying them by a random number with a mean equal to 1.

[0065] An enhanced perturbation strategy is adopted, that is, based on the initial perturbation, the perturbed emissions in all grids are multiplied by a random number greater than 1 generated from a uniform distribution to obtain an enhanced emission inventory for simulating emission source enhancement.

[0066] A disturbance reduction strategy is adopted, that is, based on the initial disturbance, the disturbed emissions in all grids are multiplied by a random number less than 1 generated from a uniform distribution to obtain a reduced emission inventory for simulating emission source reduction.

[0067] A cross-level adjustment disturbance strategy is adopted, that is, before the initial disturbance, the emission at each moment in the prior emission source data is adjusted by adding a random number in a set interval. At the same time, the emission is divided into 8 intervals according to the preset threshold, and the emission in each interval is subjected to cross-level adjustment disturbance. Then, the emission in each grid is multiplied by a random number that obeys different means and standard deviations and conforms to the normal distribution, so as to obtain an emission inventory in which the emission is controlled within the preset range and is used to simulate the migration of existing emission sources and the formation of new emission sources.

[0068] Among them, cross-level adjustment disturbances include:

[0069] Two types of grids are set in the study area. For one type of grid, a random number is directly added to the emissions within the grid to cause disturbance, so that the emission value after disturbance changes in magnitude. For the other type of grid, a random number is multiplied by the emissions within the grid to cause disturbance.

[0070] When disturbing emissions in different intervals, the situations at different times are taken into consideration. For moments when the prior emissions are overestimated, emissions are reduced. For moments when the prior emissions are underestimated, new emission sources are added. This can reduce high-value emissions and increase a certain proportion of low-value emissions, simulating the migration of existing emission sources and the formation of new emission sources.

[0071] Step 3: Using the atmospheric chemistry model and the disturbed emission source data, the meteorological field and pollutant concentration field in the study area are forward simulated hourly to obtain the meteorological element dataset and pollutant concentration dataset with hourly resolution.

[0072] Specifically, based on prior emission source data, hourly forward simulations were first performed over the study period, with a one-hour simulation window. The hourly simulation results served as the background field. Then, using the background field at the initial study moment as the initial field, hourly forward simulations of the meteorological and pollutant concentration fields within the study area were performed using an atmospheric chemistry model and emission inventories derived using three perturbation strategies. This yielded hourly-resolution meteorological and pollutant concentration datasets.

[0073] In step 4, the meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model.

[0074] Specifically, the hourly-resolution meteorological element dataset, pollutant concentration dataset, and prior emission source data obtained through forward simulation are used as input features and fed into a deep learning-based emission inversion model for training. The training target is defined as the difference in hourly emissions between the perturbed emission source data and the prior emission source data, i.e., the emission increment. The model inversion accuracy is evaluated by comparing the difference between the model-predicted emissions and the perturbed emission source data, until a trained emission inversion model that meets the preset inversion accuracy is obtained. In this embodiment, the U-Net model is specifically selected as the emission inversion model.

[0075] In step 5, the three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate a pollutant concentration reanalysis field.

[0076] In step 6, the meteorological element dataset, pollutant concentration reanalysis field, and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0077] Three-dimensional variational assimilation (3D-VAR) is a data assimilation method used to combine observational data from various sources (such as meteorological stations and satellite remote sensing data) with background fields from numerical models (such as atmospheric circulation models) to obtain more accurate estimates of the state of geophysical systems such as the atmosphere and ocean. By performing 3D variational assimilation on pollutant concentration observations, more accurate pollutant concentration reanalysis fields can be obtained. These reanalysis fields can then be input into a trained emission inversion model to perform inversion predictions of pollutant emission sources, resulting in more accurate model predictions.

[0078] In a specific example, taking the optimization of sulfur dioxide (SO2) emissions from fireworks during the Spring Festival period from January 20 to January 23, 2023, as an example, a single-variable pollutant emission source inversion method provided in this application was used to obtain a SO2 emission analysis field. The study area was China, and the specific steps included:

[0079] 1. SO2 concentration data from ground-based monitoring stations were collected and quality-controlled from January 11 to February 9, 2022, and January 20 to 23, 2023. FNL reanalysis data for the study period were also obtained. FNL reanalysis data are global reanalysis data provided by the National Centers for Environmental Prediction (NECP) / National Center for Atmospheric Research (NCAR). The 2016 Multi-Resolution Emission Inventory of China (MEIC_2016), compiled by Tsinghua University, was used as the prior emission source data. SO2 concentration data within the same grid were averaged based on the study area and grid resolution of the WRF-Chem model.

[0080] 2. Regarding SO2 emissions in MEIC_2016, considering that only using the MEIC inventory for January 2016 for forward simulation will result in insufficient sample diversity, thus limiting the ability of the deep learning model to learn various features under different conditions. To address this limitation and enrich the training dataset, three perturbation strategies are used to adjust SO2 emissions in MEIC_2016. Each method generates 30 daily emission inventories from January 11 to February 9, 2022. The specific steps are as follows:

[0081] 2.1. Initial disturbance:

[0082] First, according to the actual ratio of emission increase, emission reduction and normal emission, the grid ratio of hourly emission enhancement, emission reduction and normal emission in the study area is set.

[0083] Then, a set of random numbers that conform to the normal distribution and have the same dimension as the prior emission source data are created. To ensure the correlation of disturbances in adjacent grids, the random numbers are Gaussian filtered, and the SO2 emissions in each type of grid are initially disturbed by multiplying them by random numbers that obey different means and standard deviations and conform to the normal distribution. The disturbed SO2 emissions in the enhanced emission grid, weakened emission grid, and normal emission grid are obtained.

[0084] The set grid ratio and random number information are shown in Table 1.

[0085] Table 1 Grid ratio and random number information

[0086]

[0087] Specifically, the random number Select three sets of numbers with the same dimension as the emission that conform to the Gaussian distribution and , after the initial perturbation, the grid points Disturbance SO2 emissions at Expressed as

[0088] ;

[0089] in, and represents the horizontal and vertical coordinates of the grid points, Represents SO2 emissions from the MEIC_2016 inventory, for different classification grid points, Taken from different :

[0090] ;

[0091] in, N represents a normal distribution, is the mean, standard deviation It is taken from The random number in the interval is chosen to introduce more variability while ensuring the actual disturbance level. Gaussian filtering is performed to ensure the spatial consistency of the disturbance between adjacent grid points.

[0092] 2.2 Enhanced disturbance strategy: Based on the initial disturbance, the disturbance SO2 emissions in all grids are Multiply it by a value drawn from a uniform distribution in the interval Random number , and obtain the enhanced emission inventory used to simulate emission source enhancement .

[0093] 2.3. Disturbance reduction strategy: Based on the initial disturbance, the disturbance SO2 emissions in all grids are reduced. Multiply it by a value drawn from a uniform distribution in the interval Random number , and obtain the emission reduction inventory used to simulate emission source reduction .

[0094] 2.4. Cross-level adjustment disturbance strategy: Before the initial disturbance, the level of SO2 emissions is adjusted by adding a random number of a set interval to the emissions at each moment in the prior emission source data. At the same time, the SO2 emissions are divided into 8 intervals according to the preset threshold, which are expressed as

[0095] ;

[0096] in, Indicates SO2 emissions;

[0097] After the SO2 emissions in the eight intervals were adjusted across the order of magnitude, the SO2 emissions in each grid were multiplied by random numbers with different means and standard deviations and in accordance with the normal distribution, and the emissions were controlled within the range of 0-600 mol / km. 2 / h and is used to simulate the migration of existing emission sources and the formation of new emission sources. Specifically, the calculation formula for the cross-level adjustment disturbance strategy is as follows: ;

[0098] in, is taken from the interval The random number is used to adjust the emission between different intervals so that Compared with the emission after the initial disturbance, the magnitude of the increase or decrease is achieved. In order to ensure that the emission has actual physical meaning (emission cannot be negative), Constrained to numbers greater than or equal to 0. This adjustment method is intended to simulate the migration of existing emission sources and the emergence of new emission sources.

[0099] 3. Based on the MEIC_2016 emissions inventory, hourly forward simulations were performed throughout the study period, with a one-hour simulation window. The hourly simulation results served as the background field. Subsequently, using the background field at the study's initial moment as the initial field, hourly forward simulations of meteorological and pollutant concentration fields were performed from January 11 to February 9, 2022, using three versions of the emissions inventory derived from the three perturbation strategies. Through this process, three different forward simulation results, excluding the background field, were obtained every hour during the study period. The differences between the three forward simulation results were entirely due to the different emissions inventories used.

[0100] 4. Use the simulated hourly meteorological element data set (including ground pressure, boundary layer height, 2-meter water vapor mixing ratio, 10-meter wind direction and speed, and 2-meter temperature), SO2 concentration, and prior emission source data as input features and input them into the training emission inversion model. The training target is defined as the difference in hourly SO2 emissions between the perturbed emission source data and the prior emission source data, that is, the emission increment, which aims to allow the emission inversion model to learn and establish a mapping relationship between the SO2 emission increment and the relevant independent variables. The perturbation emissions from January 20 to January 23, 2023 are generated according to the same perturbation method as the training data set. It is assumed that the perturbation emissions of these four days are "real" emissions. The WRF_Chem model is combined for forward simulation, and the simulated meteorological element data and SO2 concentration are input as features into the trained model to estimate the hourly SO2 emission increment. The inversion accuracy of the emission inversion model is evaluated by comparing the difference between the SO2 emissions obtained by model inversion and the "real" emissions. Figure 2 The correlation between the SO2 emissions predicted by the emission inversion model and the “real” emissions in the test set is shown. It can be seen that the SO2 emissions predicted by the emission inversion model are very consistent with the “real” emissions, with a slope of 0.99, indicating that the model reproduces the disturbed SO2 emissions very well.

[0101] 5. Apply the three-dimensional variational assimilation method to assimilate hourly surface SO2 observation data from January 20 to January 23, 2023, into the model background field, optimize the simulation results and generate a SO2 concentration reanalysis field.

[0102] 6. The generated reanalysis SO2 concentration reanalysis field, hourly meteorological element data, and prior emission source data are input into the trained emission inversion model to generate the corresponding SO2 emission increment field. The SO2 emission increment field is then superimposed on the prior emission field to obtain the final predicted SO2 emissions.

[0103] Finally, if Figure 3 As shown in the figure, the WRF-Chem model was used to simulate and analyze the SO2 concentration based on the MEIC_2016 emission inventory and the SO2 concentration inverted in this application, and the simulation results of SO2 concentration were verified using ground station observation data. Figure 3 The figure shows the scatter plot of the simulated SO2 concentration under the two emission inventories at 17:00 UTC on January 21, 2023, after interpolation to the ground station and the observed concentration at the station. Figure 3 It can be seen that the emission sources obtained by the method proposed in this application have a significant improvement effect on forecasts and can well reproduce the peak SO2 concentration during the concentrated period of fireworks display during the Spring Festival. It has the ability to solve the emission estimation problem during short-term pollution events (large-scale fireworks displays or forest fires). It not only meets the demand for high-temporal resolution emission inventories for atmospheric pollution forecasting and control, but also significantly reduces computing resource consumption and calculation time. In addition, the emission inventory obtained based on this method can also be used to improve the pollutant concentration forecast level of atmospheric chemistry models, providing strong support for atmospheric pollution prevention, control, and forecasting and early warning.

[0104] In one embodiment, a single-variable pollutant emission source inversion device is provided, comprising:

[0105] The data acquisition module is used to obtain single-variable pollutant concentration observation data with short life cycle characteristics in the study area, meteorological reanalysis grid data and prior emission source data;

[0106] A disturbance module is used to disturb the prior emission source data by adopting an enhanced disturbance strategy, a weakened disturbance strategy, and a cross-level adjustment disturbance strategy to obtain disturbed emission source data;

[0107] The forward simulation module is used to perform hourly forward simulation of the meteorological field and pollutant concentration field in the study area using atmospheric chemistry models and disturbed emission source data, and obtain hourly resolution meteorological element datasets and pollutant concentration datasets;

[0108] The model training module is used to input meteorological element datasets, pollutant concentration datasets, and prior emission source data into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0109] The variational assimilation module is used to assimilate the hourly pollutant concentration observation data in the study area using a three-dimensional variational assimilation method to generate a pollutant concentration reanalysis field;

[0110] The emission inversion module is used to input the meteorological element dataset, pollutant concentration reanalysis field and prior emission source data into the trained emission inversion model for prediction, obtain the emission increment field, and obtain the pollutant emission inventory with hourly resolution in the study area by superimposing the emission increment field onto the prior emission field.

[0111] For the specific definition of the single-variable pollutant emission source inversion device, please refer to the definition of the single-variable pollutant emission source inversion method above, and will not be repeated here. The various modules in the above-mentioned single-variable pollutant emission source inversion device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0112] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a single variable pollutant emission source inversion method is implemented. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0113] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0114] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0115] Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area;

[0116] The enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0117] Using atmospheric chemistry models and disturbed emission source data, hourly forward simulations of the meteorological field and pollutant concentration field in the study area were performed to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0118] The meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0119] The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field;

[0120] The meteorological element dataset, pollutant concentration reanalysis field and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0121] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0122] Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area;

[0123] The enhanced disturbance strategy, weakened disturbance strategy and cross-level adjustment disturbance strategy are used to perturb the prior emission source data to obtain the disturbed emission source data.

[0124] Using atmospheric chemistry models and disturbed emission source data, hourly forward simulations of the meteorological field and pollutant concentration field in the study area were performed to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

[0125] The meteorological element dataset, pollutant concentration dataset, and prior emission source data are input into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model;

[0126] The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field;

[0127] The meteorological element dataset, pollutant concentration reanalysis field and prior emission source data are input into the trained emission inversion model for prediction to obtain the emission increment field. By superimposing the emission increment field on the prior emission field, the pollutant emission inventory with hourly resolution in the study area is obtained.

[0128] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0129] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A single variable pollutant emission source inversion method, characterized in that: The method comprises: Obtain single-variable pollutant concentration observation data, meteorological reanalysis grid data, and prior emission source data with short life cycle characteristics in the study area; The prior emission source data are disturbed by using an enhanced disturbance strategy, a weakened disturbance strategy, and a cross-level adjustment disturbance strategy to obtain disturbed emission source data; Using the atmospheric chemistry model and the disturbed emission source data, the meteorological field and pollutant concentration field in the study area are forward simulated hourly to obtain hourly resolution meteorological element datasets and pollutant concentration datasets; Inputting the meteorological element dataset, pollutant concentration dataset, and prior emission source data into a pre-built deep learning-based emission inversion model for training to obtain a trained emission inversion model; The three-dimensional variational assimilation method is used to assimilate the hourly pollutant concentration observation data in the study area to generate the pollutant concentration reanalysis field; The meteorological element dataset, pollutant concentration reanalysis field, and prior emission source data are input into the trained emission inversion model for prediction to obtain an emission increment field. The emission increment field is superimposed on the prior emission field to obtain an hourly resolution pollutant emission inventory for the study area. The enhanced disturbance strategy, the weakened disturbance strategy, and the cross-level adjustment disturbance strategy are respectively used to disturb the prior emission source data to obtain the disturbed emission source data, including: According to the actual increase, decrease and normal emission ratio of pollution sources, the grid ratio of increased emission, decreased emission and normal emission per hour in the study area is set; Create a set of random numbers that conform to the normal distribution and have the same dimension as the prior emission source data, perform Gaussian filtering on the random numbers, and perform initial perturbations by multiplying the emissions in each type of grid by random numbers that conform to the normal distribution and have different means and standard deviations, to obtain perturbed emissions in enhanced emission grids, weakened emission grids, and normal emission grids; An enhanced perturbation strategy is adopted, that is, based on the initial perturbation, the perturbed emissions in all grids are multiplied by a random number greater than 1 generated from a uniform distribution to obtain an enhanced emission inventory for simulating emission source enhancement; A disturbance reduction strategy is adopted, that is, based on the initial disturbance, the disturbance emissions in all grids are multiplied by a random number less than 1 generated from a uniform distribution to obtain a reduced emission inventory for simulating emission source reduction; A cross-level adjustment disturbance strategy is adopted, that is, before the initial disturbance, the emission at each moment in the prior emission source data is adjusted by adding a random number in a set interval. At the same time, the emission is divided into 8 intervals according to the preset threshold, and the emission in each interval is subjected to cross-level adjustment disturbance. Then, the emission in each grid is multiplied by a random number that obeys different means and standard deviations and conforms to the normal distribution, so as to obtain an emission inventory in which the emission is controlled within the preset range and is used to simulate the migration of existing emission sources and the formation of new emission sources.

2. The method according to claim 1, characterized in that After obtaining single-variable pollutant concentration observation data with short life cycle characteristics in the study area, meteorological reanalysis grid data and prior emission source data, it also includes: The pollutant concentration observation data are quality controlled, and the pollutant concentration observation data falling within the same grid are averaged according to the study area and grid resolution of the WRF-Chem model.

3. The method according to claim 1, characterized in that Initial perturbations are performed by multiplying the emissions in each type of grid by random numbers that have different means and standard deviations and conform to normal distributions, including: The emissions within the enhanced emission grid are initially disturbed by multiplying a random number with a mean greater than 1, the emissions within the weakened emission grid are initially disturbed by multiplying a random number with a mean less than 1, and the emissions within the normal emission grid are initially disturbed by multiplying a random number with a mean equal to 1.

4. The method according to claim 1, wherein The emissions in each interval are adjusted across levels, including: Two types of grids are set in the study area. For one type of grid, a random number is directly added to the emissions within the grid to cause disturbance, so that the emission value after disturbance changes in magnitude. For the other type of grid, a random number is multiplied by the emissions within the grid to cause disturbance. When disturbing emissions in different intervals, the situations at different times are taken into consideration. Emissions are reduced when the prior emissions are overestimated, and new emission sources are added when the prior emissions are underestimated.

5. The method according to claim 1, wherein Using the atmospheric chemistry model and the disturbed emission source data, the meteorological field and pollutant concentration field in the study area are simulated hourly forward to obtain hourly resolution meteorological element datasets and pollutant concentration datasets, including: Based on the prior emission source data, hourly forward simulations were performed during the study period. The simulation time window was set to 1 hour, and the hourly simulation results were used as the background field. Taking the background field at the initial moment of the study as the initial field, the atmospheric chemistry model and the emission inventory obtained by using three perturbation strategies are used to perform hourly forward simulations of the meteorological field and pollutant concentration field in the study area to obtain hourly resolution meteorological element datasets and pollutant concentration datasets.

6. The method according to claim 1, characterized in that The meteorological element dataset, pollutant concentration dataset and prior emission source data are input into a pre-built emission inversion model based on deep learning for training to obtain a trained emission inversion model, including: The hourly-resolution meteorological element dataset, pollutant concentration dataset and prior emission source data obtained by forward simulation are used as input features and input into the deep learning-based emission inversion model for training. The training target is defined as the difference in hourly emissions between the disturbed emission source data and the prior emission source data, that is, the emission increment. The model inversion accuracy is evaluated by comparing the difference between the model-predicted emissions and the disturbed emission source data until a trained emission inversion model that meets the preset inversion accuracy is obtained.

7. A single variable pollutant emission source inversion device, characterized in that: The device comprises: The data acquisition module is used to obtain single-variable pollutant concentration observation data with short life cycle characteristics in the study area, meteorological reanalysis grid data and prior emission source data; A disturbance module is used to disturb the prior emission source data by adopting an enhanced disturbance strategy, a weakened disturbance strategy, and a cross-level adjustment disturbance strategy to obtain disturbed emission source data; A forward simulation module is used to perform hourly forward simulation of the meteorological field and pollutant concentration field in the study area using the atmospheric chemistry model and the perturbed emission source data, thereby obtaining a meteorological element dataset and a pollutant concentration dataset with hourly resolution; A model training module is used to input the meteorological element dataset, pollutant concentration dataset and prior emission source data into a pre-built emission inversion model based on deep learning for training, thereby obtaining a trained emission inversion model; The variational assimilation module is used to assimilate the hourly pollutant concentration observation data in the study area using a three-dimensional variational assimilation method to generate a pollutant concentration reanalysis field; An emission inversion module is used to input the meteorological element dataset, pollutant concentration reanalysis field, and prior emission source data into the trained emission inversion model for prediction, obtain an emission increment field, and obtain an hourly resolution pollutant emission inventory for the study area by superimposing the emission increment field onto the prior emission field; The disturbance module is specifically used for: According to the actual increase, decrease and normal emission ratio of pollution sources, the grid ratio of increased emission, decreased emission and normal emission per hour in the study area is set; Create a set of random numbers that conform to the normal distribution and have the same dimension as the prior emission source data, perform Gaussian filtering on the random numbers, and perform initial perturbations by multiplying the emissions in each type of grid by random numbers that conform to the normal distribution and have different means and standard deviations, to obtain perturbed emissions in enhanced emission grids, weakened emission grids, and normal emission grids; An enhanced perturbation strategy is adopted, that is, based on the initial perturbation, the perturbed emissions in all grids are multiplied by a random number greater than 1 generated from a uniform distribution to obtain an enhanced emission inventory for simulating emission source enhancement; A disturbance reduction strategy is adopted, that is, based on the initial disturbance, the disturbance emissions in all grids are multiplied by a random number less than 1 generated from a uniform distribution to obtain a reduced emission inventory for simulating emission source reduction; A cross-level adjustment disturbance strategy is adopted, that is, before the initial disturbance, the emission at each moment in the prior emission source data is adjusted by adding a random number in a set interval. At the same time, the emission is divided into 8 intervals according to the preset threshold, and the emission in each interval is subjected to cross-level adjustment disturbance. Then, the emission in each grid is multiplied by a random number that obeys different means and standard deviations and conforms to the normal distribution, so as to obtain an emission inventory in which the emission is controlled within the preset range and is used to simulate the migration of existing emission sources and the formation of new emission sources.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images