Training method and device of atmospheric pollution regulation model, and prediction method and device

Abstract

The present disclosure provides a training method and device of an air pollution regulation model, an air pollution information prediction method and device, an electronic device, a computer readable storage medium and a computer program product. The training method comprises: obtaining first pollutant prediction information according to meteorological condition information and pollutant emission samples by using an air quality model; generating a training sample; obtaining second pollutant prediction information according to the training sample by using an air pollution regulation model; determining a model loss function according to the first pollutant prediction information and the second pollutant prediction information, and adjusting the air pollution regulation model. The present disclosure can predict pollutants such as PM2.5, regulate pollutant emissions, and the like. On the basis of ensuring accurate prediction of pollutants, the present disclosure can significantly reduce the demand for computing resources, improve computing efficiency, improve the accuracy of pollutant prediction, and improve the scientificity and accuracy of pollution control.

Description

Technical Field

[0001] This application relates to the field of atmospheric environmental science and technology, and in particular to a training method and device for an atmospheric pollution control model, a method and device for predicting atmospheric pollution information, an electronic device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] Atmospheric chemical transport models (CTMs) are used to simulate historical pollutant concentrations or provide operational air quality forecasts. Atmospheric pollution control models simulate, analyze, and predict the spatiotemporal distribution changes of air pollutants under different pollution control strategies by establishing mathematical models. Models utilizing atmospheric chemical transport principles for simulating air pollution control include various air quality models such as CMAQ (Community Multiscale Air Quality). Currently, existing atmospheric chemical transport models are typically based on partial differential equations (PDEs) that detail atmospheric physical and chemical processes. Due to the high coupling and complexity of PDE equations, the solution process is very complex, computationally inefficient, requires significant computational resources and time, and its applicability in real-world pollution control scenarios is limited. Summary of the Invention

[0003] This disclosure provides a training method and apparatus for an air pollution control model, an air pollution information prediction method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product.

[0004] According to a first aspect of this disclosure, a training method for an air pollution control model is provided, comprising: using an air quality model to obtain first pollutant prediction information based on meteorological condition information and pollutant emission samples; generating training samples, wherein the training samples include the meteorological condition information, the pollutant emission samples, and historical pollutant information, and the labels of the training samples are the first pollutant prediction information; using the air pollution control model to obtain second pollutant prediction information based on the training samples; determining a model loss function based on the first pollutant prediction information and the second pollutant prediction information; and adjusting the air pollution control model based on the model loss function.

[0005] Optionally, obtaining the second pollutant prediction information using the air pollution control model based on the training samples includes: generating meteorological variable input data based on the meteorological condition information; generating emission variable data based on the pollutant emission samples; generating historical background data based on the historical pollutant information; obtaining pollutant characteristic information using the air pollution control model based on the meteorological variable input data and the emission variable data, wherein the pollutant characteristic information includes pollutant concentration change information; summing the pollutant concentration change information and the historical background data using the air pollution control model to obtain pollutant concentration information; and obtaining the second pollutant prediction information, wherein the second pollutant prediction information includes the pollutant concentration information and the pollutant concentration change information.

[0006] Optionally, the air pollution control model includes a 3D convolutional layer, an activation function, a residual module, an encoder, a decoder, and a 2D convolutional layer; the step of obtaining pollutant characteristic information based on the meteorological variable input data and the emission variable data using the air pollution control model includes: using the 3D convolutional layer to extract 3D spatial features from the meteorological variable input data and the emission input data; using the activation function to activate the output data of the 3D convolutional layer, wherein the residual module establishes a skip connection between the input and output of the 3D convolutional layer; using the encoder to downsample and pool the output data of the activation function; and using the decoder to... The output data of the encoder is upsampled. The encoder includes 3D convolutional layers and pooling layers, and the decoder includes multiple transposed convolutional layers. Skip connections are established between the corresponding layers of the encoder and the decoder. The output data of the encoder is processed using the 2D convolutional layers to obtain the pollutant feature information. The number of 2D convolutional layers is multiple. The process of using the air pollution control model to sum the pollutant concentration change information and the historical background data to obtain the pollutant concentration information includes summing the pollutant concentration change information output from the 2D convolutional layers with the historical background data to obtain the pollutant concentration information.

[0007] Optionally, both the first pollutant prediction information and the second pollutant prediction information include pollutant concentration information and pollutant concentration change information. The adaptive weights include concentration adaptive weights and concentration change adaptive weights. Determining the model loss function based on the first pollutant prediction information, the second pollutant prediction information, and the adaptive weights of the air pollution control model includes: determining a concentration loss function based on the pollutant concentration information in the first pollutant prediction information and the pollutant concentration information in the second pollutant prediction information; determining a concentration change loss function based on the pollutant concentration change information in the first pollutant prediction information and the pollutant concentration change information in the second pollutant prediction information; determining a concentration accuracy loss function based on the concentration loss function, the concentration adaptive weights, and a preset constant; determining a concentration change accuracy loss function based on the concentration change loss function, the concentration change accuracy loss function, the concentration adaptive weights, and the concentration change adaptive weights; and determining the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weights, and the concentration change adaptive weights.

[0008] Optionally, determining the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weight, and the concentration change adaptive weight includes: determining penalty information based on the concentration adaptive weight and the concentration change adaptive weight; and determining the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration change adaptive weight, the concentration change adaptive weight, and the penalty information.

[0009] Optionally, the pollutant emission samples include a baseline scenario emission sample and the control scenario emission sample. The training method includes: sampling anthropogenic pollutant emission information in the atmospheric emission information to obtain a baseline scenario emission sample; obtaining atmospheric emission control information based on the anthropogenic pollutant emission information in the atmospheric emission information; and sampling the atmospheric emission control information to obtain a control scenario emission sample.

[0010] Optionally, obtaining atmospheric emission control information based on anthropogenic pollutant emission information in the atmospheric emission information includes: processing the anthropogenic pollutant emission information in the atmospheric emission information according to emission control factors, as well as spatial and temporal dimensions, to obtain the atmospheric emission control information.

[0011] Optionally, sampling the atmospheric emission control information to obtain control scenario emission samples includes: sampling the atmospheric emission control information according to the spatial dimension, the temporal dimension, and the sampling dimension to obtain the control scenario emission samples; sampling the anthropogenic pollutant emission information in the atmospheric emission information to obtain baseline scenario emission samples includes: sampling the anthropogenic pollutant emission information in the atmospheric emission information according to the spatial dimension, the temporal dimension, and the sampling dimension to obtain the baseline scenario emission samples; wherein, the sampling dimension includes a pollution source dimension and a pollutant type dimension, the pollution source dimension includes at least one dimension among power, industry, and other sectors; the pollutant type dimension includes at least one dimension among sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia, and particulate matter.

[0012] Optionally, the meteorological condition information can be obtained using a meteorological model based on meteorological reanalysis data, observation assimilation data, and topographic data.

[0013] Optionally, the historical pollutant information, the first pollutant prediction information, and the second pollutant prediction information include: the concentration of PM2.5 and oxides, the amount of concentration change, and the components of PM2.5 and the types of oxides.

[0014] According to a second aspect of this disclosure, an air pollution information prediction method is provided, comprising: generating model input data based on meteorological condition information, atmospheric emission information, and historical pollutant information; and obtaining pollutant prediction information based on the model input data using an air pollution control model; wherein the air pollution control model is trained by the training method described above.

[0015] According to a third aspect of this disclosure, a training device for an air pollution control model is provided, comprising: a first data acquisition module, configured to obtain first pollutant prediction information based on meteorological conditions and pollutant emission samples using an air quality model; a sample generation module, configured to generate training samples, wherein the training samples include the meteorological conditions, pollutant emission samples, and historical pollutant information, and the training samples are labeled with the first pollutant prediction information; a second data acquisition module, configured to obtain second pollutant prediction information based on the training samples using the air pollution control model; a loss determination module, configured to determine a model loss function based on the first pollutant prediction information and the second pollutant prediction information; and a model adjustment module, configured to adjust the air pollution control model based on the model loss function.

[0016] According to a fourth aspect of this disclosure, an air pollution information prediction device is provided, comprising: a data generation module for generating model input data based on meteorological condition information, atmospheric emission information, and historical pollutant information; and an information prediction module for generating pollutant prediction information based on the model input data using an air pollution control model; wherein the air pollution control model is trained using the training method described above.

[0017] According to a fifth aspect of this disclosure, an electronic device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to perform the method described above based on instructions stored in the memory.

[0018] According to a sixth aspect of this disclosure, a computer-readable storage medium is provided that stores computer instructions, which, when executed by a processor, implement the steps of the method described above.

[0019] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the method described above.

[0020] This disclosure discloses a training method, apparatus, air pollution information prediction method, apparatus, electronic equipment, computer-readable storage medium, and computer program product for an air pollution control model. It utilizes an air quality model to obtain prediction information for a first pollutant. Training samples are generated based on meteorological conditions, pollutant emission samples, and historical pollutant information. The training samples are labeled with the first pollutant prediction information and used to obtain prediction information for a second pollutant using the air pollution control model and the training samples. Based on the first and second pollutant prediction information, a model loss function is determined, and the air pollution control model is adjusted. The trained air pollution control model can predict pollutants such as PM2.5 and control pollutant emissions, significantly reducing the demand for computing resources, improving computational efficiency, enhancing the accuracy of pollutant prediction, and improving the scientific rigor and precision of pollution control. Attached Figure Description

[0021] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The accompanying drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. The above and other objects and advantages of this disclosure will be further described below with reference to specific embodiments and the accompanying drawings. In the drawings, the same or corresponding technical features or components will be represented by the same or corresponding reference numerals.

[0022] Figure 1 This is a flowchart illustrating some embodiments of the training method for the air pollution control model according to the present disclosure;

[0023] Figure 2 This is a schematic diagram of the process for obtaining pollutant emission samples in some embodiments of the training method for the air pollution control model according to this disclosure;

[0024] Figure 3 This is a schematic diagram of the multidimensional cube structure used for sampling;

[0025] Figure 4 This is a schematic diagram of the process for obtaining prediction information of a second pollutant in some embodiments of the training method for the air pollution control model according to the present disclosure.

[0026] Figure 5 This is a schematic diagram of the process for obtaining prediction information of a second pollutant in some embodiments of the training method for the air pollution control model according to the present disclosure.

[0027] Figure 6 This is a schematic diagram of the structure of an air pollution control model;

[0028] Figure 7 This is a flowchart illustrating the determination of the model loss function in some embodiments of the training method for the air pollution control model according to this disclosure;

[0029] Figure 8 A scatter plot for the test set validation of the air pollution control model;

[0030] Figure 9 PM2.5 concentration response validation diagram used to validate the air pollution control model;

[0031] Figure 10 This is a flowchart illustrating some embodiments of the air pollution information prediction method according to the present disclosure;

[0032] Figure 11 A schematic diagram of some embodiments of a training device for an air pollution control model according to the present disclosure;

[0033] Figure 12 Schematic diagrams of modules for training apparatus of an air pollution control model according to this disclosure;

[0034] Figure 13 The diagram shows some embodiments of an air pollution information prediction device according to this disclosure.

[0035] Figure 14 This is a schematic diagram of modules of some embodiments of an electronic device according to the present disclosure. Detailed Implementation

[0036] Exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of the embodiments are described in the specification. However, it should be understood that many implementation-specific settings must be made in carrying out the embodiments to achieve the developer's specific goals, such as complying with constraints related to the device and business, and these constraints may vary depending on the implementation. Furthermore, it should be understood that while development work can be very complex and time-consuming, such development work is merely a routine task for those skilled in the art who benefit from the present disclosure.

[0037] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0038] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of this disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0039] It should also be understood that in the embodiments disclosed herein, "a plurality of" may refer to two or more, and "at least one" may refer to one, two or more.

[0040] It should also be understood that any component, data or structure mentioned in the embodiments of this disclosure can generally be understood as one or more unless expressly defined or given to the contrary in the context.

[0041] Furthermore, the term "and / or" in this disclosure is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this disclosure generally indicates that the preceding and following related objects have an "or" relationship.

[0042] It should also be understood that the description of the various embodiments in this disclosure emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0043] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0044] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0045] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0046] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0047] Furthermore, to avoid obscuring this disclosure with unnecessary detail, only processing steps and / or apparatus structures closely related to at least the solutions according to this disclosure are shown in the accompanying drawings, while other details less relevant to this disclosure are omitted. It should also be noted that similar reference numerals and letters in the drawings indicate similar items, and therefore once an item is defined in one drawing, it need not be discussed again in subsequent drawings.

[0048] Figure 1 This is a flowchart illustrating some embodiments of the training method for the air pollution control model according to this disclosure. Figure 1 As shown, the training method for the air pollution control model includes steps S101-S105.

[0049] Step S101: Using an air quality model, based on meteorological conditions and pollutant emission samples, obtain prediction information for the first pollutant.

[0050] Air quality models can be of various types, such as the CMAQ model. Meteorological condition information can be obtained through various methods, for example, using a WRF (Weather Research and Forecasting Model). Pollutant emission samples can be obtained from pollutant emission inventories using various methods. By inputting meteorological condition information and pollutant emission samples into the air quality model, the first pollutant prediction information output by the air quality model is obtained.

[0051] Step S102: Generate training samples, which include meteorological condition information, pollutant emission samples and historical pollutant information. The labels of the training samples are the first pollutant prediction information.

[0052] Historical pollutant information comprises data and characteristics related to pollutants over a specific historical period. This data includes pollutant concentrations, for example, historical pollutant information includes pollutant concentration information for a specific historical period. This historical pollutant information can be used for model learning, understanding pollutant behavior patterns, change patterns, and interactions. Multiple training samples can be generated based on meteorological conditions, pollutant emission samples, and historical pollutant information to form a training sample set. The first pollutant prediction information obtained using an air quality model, corresponding to meteorological conditions and pollutant emission samples, can be used as the label information for the training samples.

[0053] Step S103: Using an air pollution control model, based on training samples, obtain prediction information for the second pollutant.

[0054] Air pollution control models can take many forms, such as convolutional neural network models and adversarial network models. Training samples are input into the air pollution control model to obtain the predicted information of the second pollutant output by the air pollution control model.

[0055] The pollutants in historical pollutant information, primary pollutant prediction information, and secondary pollutant prediction information include PM2.5, oxides, etc. This information can include the concentrations and changes in concentrations of PM2.5 and oxides, as well as information on PM2.5 components and oxide types. PM2.5 refers to particulate matter in the atmosphere with a diameter of 2.5 micrometers or less. Oxides can be various types of oxides. PM2.5 components can include sulfates, nitrates, ammonium salts, organic particulate matter, secondary organic particulate matter, black carbon, and other particulate matter. Oxides include ozone, nitrogen oxides, and atmospheric free radicals.

[0056] Step S104: Determine the model loss function based on the prediction information of the first pollutant and the prediction information of the second pollutant.

[0057] Step S105: Adjust the air pollution control model according to the model loss function.

[0058] Various training methods can be used to train the air pollution control model. For example, based on the predicted information of the first pollutant obtained from the air quality model and the predicted information of the second pollutant obtained from the air pollution control model, the model loss function can be determined. The air pollution control model can then be iteratively trained based on the model loss function. The parameters of the air pollution control model can be adjusted according to the function value of the model loss function, gradually reducing the function value of the model loss function until it is less than a preset threshold, thus obtaining a well-trained air pollution control model.

[0059] The training method for the air pollution control model disclosed herein utilizes an air quality model to obtain prediction information for the first pollutant. Training samples are generated based on meteorological conditions, pollutant emission samples, and historical pollutant information. The training samples are labeled with the first pollutant prediction information and used to obtain prediction information for the second pollutant using the air pollution control model and the training samples. A model loss function is determined based on the first and second pollutant prediction information for adjusting the air pollution control model. The trained air pollution control model can predict pollutants such as PM2.5 and control pollutant emissions, significantly reducing the demand for computational resources and improving computational efficiency. It provides a basic model tool and scientific decision support for pollution control, improving the accuracy of pollutant predictions. It can be applied to environmental management, policy formulation, and pollution prevention and control, enhancing the scientific rigor and precision of pollution control.

[0060] Meteorological models can be used to obtain meteorological condition information based on meteorological reanalysis data, observation assimilation data, and topographic data; meteorological models can be WRF models, etc.

[0061] In some embodiments, simulation nesting can be configured to cover regions such as China, and range parameters such as the projection method, grid resolution, and number of vertical layers of the simulation region can be determined. The configuration parameters for simulation nesting are shown in Table 1 below:

[0062] Projection method Lambert conformal projection Horizontal resolution 36km×36km Number of grids 127×172 Vertical number of layers 23 (Meteorology); 14 (Chemistry) Nesting 1

[0063] Table 1 - Setting Parameters for Simulated Nesting

[0064] Meteorological conditions can be simulated within a nested simulation framework based on parameters determined as shown in Table 1, utilizing global meteorological reanalysis data such as NCEP-FNL (as initial and boundary conditions) and topographic data such as USGS data. The WRF model sets various parameterization schemes for different physical processes, such as cloud microphysics modules, boundary layer modules, and longwave and shortwave radiation modules. The accuracy of simulating specific meteorological conditions varies significantly between different combinations of physical schemes. Based on meteorological principles, the optimal combination of parameterization schemes for the WRF model is selected. The parameterization scheme combinations for this simulation area are shown in Table 2 below.

[0065] Parameterization scheme Solution Name shortwave radiation Goddard shortwave scheme Longwave radiation RRTM scheme surface layer Pleim-Xiu surface layer Boundary layer ACM2(Pleim)PBL Cumulus convection Kain-Fritsch (new Eta) scheme Microphysics WSM 6-class graupel scheme

[0066] Table 2 - Combination of Parameterization Schemes for WRF Models

[0067] WRF models often suffer from systematic errors when simulating meteorological fields, such as difficulty in accounting for the attenuation effect of surface topography on wind speed. To improve the simulation accuracy of WRF models, a series of assimilation modules were added, using observational assimilation data such as NCEP-OBS data to constrain the simulation process of the WRF model. The assimilation information settings for the WRF model are shown in Table 3:

[0068] Assimilation module Assimilation variables Sea surface temperature regeneration Sea surface temperature Analysis of assimilation Temperature, specific humidity, and wind field (wind field only in the boundary layer). Observational assimilation Temperature, humidity, and airflow of the entire floor Soil assimilation Soil temperature, soil moisture, surface wind field

[0069] Table 3 - Assimilation Information Configuration Table for WRF Model

[0070] After configuring the WRF model, meteorological condition information is obtained by using the WRF model and meteorological analysis data such as NCEP-FNL global meteorological reanalysis data, observation assimilation data such as NCEP-OBS data, and topographic data such as USGS data. The meteorological condition information includes temperature, air pressure, humidity, wind field, cloud characteristics, precipitation, and radiation for the target area.

[0071] Figure 2 This is a schematic diagram illustrating the process of obtaining pollutant emission samples in some embodiments of the training method for the air pollution control model according to this disclosure. The pollutant emission samples include baseline scenario emission samples and control scenario emission samples, such as... Figure 2 As shown:

[0072] Step S201: Sample the anthropogenic pollutant emission information in the atmospheric emission information to obtain a baseline scenario emission sample.

[0073] Step S202: Obtain atmospheric emission control information based on anthropogenic pollutant emission information in atmospheric emission information.

[0074] Step S203: Sample atmospheric emission control information to obtain emission samples under control scenarios;

[0075] Step S204: After sampling and regulation, the natural source pollutant emission inventory is merged into the atmospheric emission information.

[0076] In some embodiments, atmospheric emission information can be obtained from multiple sources, such as anthropogenic emission inventories, which summarize emissions of various air pollutants generated by human activities. Natural source emission inventories are compilations and statistics of air pollutant emissions generated by natural processes. After sampling and regulation, natural source pollutant emission inventories are merged into the atmospheric emission information to generate new atmospheric emission information.

[0077] The MEIC-TOOL tool, written in Java, and the MEIC2CTM tool, written in Python, can be used to distribute the total emissions inventory of China calculated by the MEIC (Multi-resolution Emission Inventory for China) and MEIC-HR (Multi-resolution Emission Inventory for China-High Resolution) models to hourly grid data according to specific spatiotemporal parameters, and couple it with the MIX anthropogenic emissions inventory for East Asia. Key parameters of the total emissions inventory of China and the MIX anthropogenic emissions inventory are shown in Table 4 below.

[0078]

[0079]

[0080] Table 4 - Key Parameters of Emission Inventory

[0081] The meteorological field simulation results output from the WRF model can be processed using the MCIP (Meteorology-Chemistry Interface Processor) module. Natural source emission factors can be calculated using MODIS-LAI leaf area index and PFT vegetation type data. The meteorological conditions output from the MCIP module can then be input into the MEGAN (Model of Emissions of Gases and Aerosols) model to calculate the natural source emissions corresponding to the hourly meteorological conditions, thus obtaining a natural source emission inventory. Anthropogenic and natural source emission inventories can be merged to obtain regional total emission inventories and other atmospheric emission information.

[0082] The initial conditions (ICON model) and boundary conditions (BCON module) of the CMAQ model use the model's default data. The atmospheric chemical transport module (CCTM) of the CMAQ model uses optimized chemical mechanisms and online emissions. The output of CCTM includes various important pollutants, as well as free radicals and atmospheric chemical intermediates. The post-processing module outputs PM10 concentration, PM2.5 and its component concentrations, daily maximum eight-hour ozone concentration, daily maximum one-hour ozone concentration, NO2 concentration, SO2 concentration, etc. The configuration table of the CMAQ model is shown in Table 5 below.

[0083] Configuration Name Detailed settings Gas phase chemical mechanism CB05 Liquid phase chemical mechanism RADM Aerosol Module AERO6 Aerosol Thermodynamics ISORROPIA cloud module ACM Online emissions Turn on dust and sea salt aerosol emissions, turn off lightning emissions.

[0084] Table 5 - Configuration Table of CMAQ Model

[0085] Based on the simulation range, meteorological conditions, and atmospheric emission information set above, the pollutant concentrations for the baseline scenario can be obtained using the CMAQ model. A meteorological-emissions-air quality model system can be built on a high-performance computing platform to support parallel simulation of emission control scenarios.

[0086] In some embodiments, atmospheric emission control information can be obtained by controlling and processing anthropogenic pollutant emission information in atmospheric emission information based on emission control factors, as well as spatial and temporal dimensions.

[0087] For example, multiple control scenarios can be set up for model training. For each control scenario, a set of emission control factors (IFs) and their applicable spatiotemporal scales (spatial and temporal dimensions) can be set. The control factors are defined in the range of 0-1 to characterize the degree of control over the emissions of various pollutants such as sulfur dioxide, nitrogen oxides, and non-methane volatile organic compounds. The control factors are set between 0 (complete emission shutdown) and 1 (no control, maintaining consistency with the baseline state).

[0088] IFs are used to adjust anthropogenic pollutant emission information in the atmospheric emission information under the baseline state of the model, thereby generating different atmospheric emission control information. Spatial dimensions include at least one of national, provincial, and geographic grid dimensions. The national dimension means that the same control factor is used for the entire national grid; the provincial dimension means that the same control factor is used for the entire provincial grid; and the geographic grid dimension means that all grids within China use heterogeneous control factors. Temporal dimensions include at least one of year, season, month, and day. In the spatial dimension, control can be performed using a three-level scale: national, provincial, and geographic grid.

[0089] In some embodiments, atmospheric emission control information is sampled according to spatial, temporal, and sampling dimensions to obtain control scenario emission samples; atmospheric emission information is also sampled according to spatial, temporal, and sampling dimensions to obtain baseline scenario emission samples. The sampling dimensions include pollution source dimensions and pollutant type dimensions. The pollution source dimension includes at least one dimension from the power, industry, and other sectors. The pollutant type dimension includes at least one dimension from sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia, and particulate matter. The control scenario emission samples and baseline scenario emission samples may include pollutant emission information with temporal and spatial characteristics. The pollutants may include one or more of sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia, and particulate matter.

[0090] For example, various algorithms, such as the SOBOL algorithm, can be used to calculate the emission control tensor based on spatial, temporal, and sampling dimensions. This tensor is then applied to atmospheric emission control information (using spatial dimensions) for random sampling, resulting in a large-scale emission sample of the control scenario. A multidimensional cubic structure for sampling is shown below. Figure 3 As shown.

[0091] The pollution source dimension can be categorized into three dimensions: power generation, industry, and other sectors. The pollutant type dimension includes five dimensions: sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia, and particulate matter. Based on the pollution source and pollutant type dimensions, a 15-dimensional independent sampling space can be constructed to generate emission samples for control scenarios.

[0092] It can obtain samples of special control scenarios. By setting the number and types of pollutant categories, it can control emissions of nitrogen oxides and non-methane volatile organic compounds with stronger nonlinearity, and obtain emission samples of control scenarios such as emission control of single emission sources.

[0093] Specific emission control scenario samples can be developed by combining special cases from policy evaluations; additional measures can be added targeting NO. x The control scenario of gradually reducing NMVOCs can be used to improve the learning ability of the nonlinear response of PM2.5 and its component concentrations. It can also increase the control scenario of reducing emissions only for a certain pollution source to improve the adaptability to actual policy application.

[0094] Based on spatial, temporal, and sampling dimensions, SOBOL high-dimensional sampling quasi-random sampling technology can be used to perform 15-dimensional sampling space automated sampling in atmospheric emission information, efficiently obtain emission disturbance tensors, and apply them to baseline scenario emission samples.

[0095] The obtained baseline scenario emission samples and control scenario emission samples, along with meteorological condition information obtained through the WRF model, are input into the CMAQ model to simulate the baseline scenario and the massive control scenario in parallel, obtaining the first pollutant prediction information output by the CMAQ model. Environmental monitoring data can be used to verify the accuracy of the pollutant simulation corresponding to the baseline scenario emission samples. For the WRF model-CMAQ model, a preheating period of 17 days or more is added before each year / monthly simulation to eliminate the influence of initial conditions on the simulation results (first pollutant prediction information).

[0096] In some embodiments, the MEIC sectoral list (industrial, power, agricultural, residential, transportation) can be merged into industrial sources, power sources, and other sources based on different pollution source heights to determine the pollution source dimension. The pollutant dimension of the sampling space can be determined by combining the controlled species of emission control policies. For example, based on different emission policies, the sampled pollutants can be identified as sulfur dioxide, nitrogen oxides, ammonia, non-methane volatile organic compounds, particulate matter, etc. Due to PM... 2.5 Related pollutants typically use the same end-of-pipe treatment equipment, therefore the effect on primary PM2.5 is similar. 2.5 Black carbon (BC), organic carbon (OC), PM 10 Maintain the same dimensional perturbation.

[0097] Consider PM 2.5 The responses of emissions control measures to emission regulation vary across different seasons. Four months—January, April, July, and October—with varying emissions and meteorological conditions are used to represent different seasons (time dimension). Three spatial scales (national, provincial, and grid scales) are employed to characterize the spatial heterogeneity of emission control measures.

[0098] Figure 4 This is a schematic diagram illustrating the process of obtaining prediction information for a second pollutant in some embodiments of the training method for the air pollution control model according to this disclosure, such as... Figure 4 As shown:

[0099] Step S501: Generate meteorological variable input data based on meteorological condition information.

[0100] Step S502: Generate emission variable data based on pollutant emission samples.

[0101] Step S503: Generate historical background data based on pollutant information from historical states.

[0102] Step S504: Using an atmospheric pollution control model, pollutant characteristic information is obtained based on meteorological variable input data and emission variable data. The pollutant characteristic information includes pollutant concentration change information.

[0103] Step S505: Summing the pollutant concentration change information and historical background data to obtain pollutant concentration information.

[0104] Step S506: Obtain second pollutant prediction information, wherein the second pollutant prediction information includes pollutant concentration information and pollutant concentration change information; the second pollutant prediction information may also include the composition or type of pollutant.

[0105] In some embodiments, for an air pollution control model, the intervention state of the model is defined, including an emission baseline scenario and an emission control scenario; the emission baseline scenario is the actual emission scenario, and the emission control scenario is the scenario where control is applied based on the baseline scenario. The emission input without control is the baseline scenario, and the obtained multidimensional emission perturbation tensor, used to control the baseline scenario, is the control scenario.

[0106] The baseline and controlled scenario emission samples contain pollutant information for major anthropogenic pollutants, including PM2.5, black carbon (BC), organic carbon (OC), particulate matter with an aerodynamic diameter between 2.5 μm and 10 μm (PMcoarse), NMVOC, NH3, and NO. x Pollutants such as SO2 are included in the baseline and controlled scenario emission samples, covering power generation, industry, and other sources. Pollutant information from both samples can be vertically aggregated to obtain daily emissions in a unified two-dimensional spatial format. Variables from the WRF model output are extracted, primarily describing temperature, air pressure, humidity, wind field, cloud characteristics, precipitation, radiation, heat flux, and underlying surface characteristics. The vertical layers of the 3D variables from the CMAQ model can be flattened to feature dimensions, resulting in unified two-dimensional meteorological variables (meteorological condition information).

[0107] The input data for setting up an air pollution control model includes meteorological conditions, pollutant emission samples (baseline scenario emission samples and control scenario emission samples), and historical pollutant information (historical pollutant concentrations and oxidants). The historical pollutant information can be from years such as 2017, simulated by the WRF-CMAQ model based on the meteorological conditions of the intervention scenario and fixed anthropogenic emissions for 2017. The historical pollutant information is only relevant to meteorological data within a limited range of intervention scenarios; therefore, it is pre-prepared and integrated into the model. When using the model, users only need to ensure that the input defining the model's baseline state matches the meteorological data of the selected intervention scenario.

[0108] The min-max normalization method is used to normalize all input variables (test samples and test samples, etc.) of the air pollution control model to between 0 and 1; a data generator is constructed to generate model input features in batches.

[0109] The generated training and test samples include meteorological condition information, pollutant emission samples, and historical pollutant information; the labels for the training and test samples are the corresponding first pollutant prediction information. The ratio of test samples to test samples is 4:1, which can be achieved using a stratified random splitting method to ensure that emission samples from different types of control scenarios are evenly distributed across both the test and test sample sets.

[0110] Figure 5 This is a schematic diagram illustrating the process of obtaining predicted information for a second pollutant in some embodiments of the training method for the air pollution control model according to this disclosure. The air pollution control model includes 3D convolutional layers, activation functions, residual modules, encoders, decoders, and 2D convolutional layers, etc. Figure 5 As shown:

[0111] Step S601: Use 3D convolutional layers to extract 3D spatial features from meteorological variable input data and emission input data.

[0112] Step S602: Activation function is used to activate the output data of the 3D convolutional layer, wherein a jump connection is established between the input and output of the 3D convolutional layer through a residual module.

[0113] Step S603: Using the encoder, the output data of the activation function is downsampled and pooled.

[0114] Step S604: The encoder's output data is upsampled using the decoder. The encoder includes 3D convolutional layers and pooling layers, and the decoder includes multiple transposed convolutional layers. Skip connections are established between the corresponding layers of the encoder and the decoder.

[0115] Step S605: The encoder output data is processed using 2D convolutional layers to obtain pollutant feature information. Multiple 2D convolutional layers are used, and the pollutant feature information includes pollutant concentration variation information.

[0116] Step S606: The pollutant concentration change information output by the 2D convolutional layer is summed with the historical background data to obtain the pollutant concentration information.

[0117] Step S607: Obtain the second pollutant prediction information. The second pollutant prediction information includes pollutant concentration information and pollutant concentration change information from the output of the 2D convolutional layer. The second pollutant prediction information may also include the composition or type of the pollutant.

[0118] Air pollution control models can be of various types, such as three-dimensional residual UNet network models. Figure 6As shown, the atmospheric pollution control model includes 3D convolutional layers, activation functions, residual modules, encoders, decoders, and 2D convolutional layers. The atmospheric pollution control model is input with multivariate spatiotemporal characteristics such as meteorology, emissions, pollutants, and oxidants, where the concentrations of pollutants and oxidants are based on historical background information.

[0119] The initial 3D convolutional layer is used to extract 3D spatial features from the input data, and the activation function is used to increase the non-linear expressive power. Residual connections are built to add skip connections between the input and output of the 3D convolutional layer. This is used to reduce the gradient vanishing problem, improve the training stability of the model, and allow the network to learn incremental changes between the input and output, making the model easier to optimize.

[0120] The algorithm utilizes an encoder for downsampling, employing a series of 3D convolutional layers to reduce the spatial size of the feature map and extract high-level features. Pooling is then used to further reduce data dimensionality. A decoder then performs upsampling using transposed convolution to restore the feature map to its original size. Skip connections are then used to fuse high-resolution and low-resolution feature information.

[0121] A series of 2D convolutional layers are used to convert features into 2D variables and output pollutant concentration changes. Summation and concatenation are used for feature aggregation, and the adjusted pollutant concentration changes are superimposed on historical pollutant concentrations to obtain the adjusted pollutant concentration information. By adjusting the hyperparameters of each module of the air pollution control model, the optimal model structure configuration can be obtained.

[0122] Figure 7 This is a flowchart illustrating the determination of the model loss function in some embodiments of the training method for the air pollution control model according to this disclosure. Both the first and second pollutant prediction information include pollutant concentration information and pollutant concentration change information. The adaptive weights include concentration adaptive weights and concentration change adaptive weights, such as... Figure 7 As shown:

[0123] Step S801: Determine the concentration loss function based on the pollutant concentration information in the first pollutant prediction information and the pollutant concentration information in the second pollutant prediction information.

[0124] Step S802: Determine the concentration change loss function based on the pollutant concentration change information in the first pollutant prediction information and the pollutant concentration change information in the second pollutant prediction information.

[0125] Step S803: Determine the concentration accuracy loss function based on the concentration loss function, the concentration adaptive weight, and the preset constant.

[0126] Step S804: Determine the concentration change accuracy loss function based on the concentration change loss function, the concentration change adaptive weight, and the preset constant.

[0127] Step S805: Determine the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weight, and the concentration change adaptive weight.

[0128] Several methods can be used to determine the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weights, and the concentration change adaptive weights. For example, penalty information can be determined based on the concentration adaptive weights and the concentration change adaptive weights; the model loss function can be determined based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration change adaptive weights, the concentration change adaptive weights, and the penalty information.

[0129] In some embodiments, the first pollutant prediction information generated by the CMAQ model (WRF-CMAQ model) includes pollutant concentration information and pollutant concentration change information, and the second pollutant prediction information generated by the air pollution control model of this disclosure includes pollutant concentration information and pollutant concentration change information; the pollutants include PM2.5 and oxides, etc., the components of PM2.5 include seven types such as sulfate, nitrate, ammonium salt, organic particulate matter, secondary organic particulate matter, black carbon and other particulate matter, and oxides include three types, that is, the total number of components or types of pollutants is 10.

[0130] Based on the pollutant concentration information in the first pollutant prediction information and the pollutant concentration information in the second pollutant prediction information, the concentration loss function is determined as shown in Formula 1-1:

[0131]

[0132] Based on the pollutant concentration change information in the first pollutant prediction information and the pollutant concentration change information in the second pollutant prediction information, the concentration change loss function is determined as shown in Formula 1-2:

[0133]

[0134] Among them, V m,n The weights for each geographic grid are calculated using formulas 1-3:

[0135]

[0136] In this context, a geographic grid divides the Earth's surface into equal-sized, regularly shaped grid units according to certain rules and scales. These grid units can be squares, rectangles, hexagons, or other shapes, and each grid has a unique identifier. For example, the grid unit (horizontal resolution) set according to the simulation nesting parameters is 36km × 36km.

[0137] By setting up a geographic grid, a geographic grid matrix can be formed. The geographic grid matrix represents the grid simulation domain. The row and column indices of the grids in the grid simulation domain are the row and column indices (row number and column number) of the geographic grids in the geographic grid matrix. The grids in the grid simulation domain correspond to the grid cells in the geographic grid matrix.

[0138] In Equations 1-1 to 1-3, C represents concentration, ΔC represents the change in concentration; m and n represent the row and column indices of the grid in the grid simulation domain (geographic grid matrix), respectively; M and N represent the total number of rows and columns in the grid simulation domain (geographic grid matrix); AI represents the data source as an air pollution control model, and WRF-CMAQ represents the data source as a CMAQ model. The pollutant concentration information in the second pollutant prediction information, which characterizes the pollutants in the grid within the grid simulation domain, Pollutant concentration information in the first pollutant prediction information that characterizes the pollutants in the grid in the grid simulation domain; The pollutant concentration change information in the second pollutant prediction information, which characterizes the pollutants in the grid of the grid simulation domain, is used to represent the pollutants in the grid. The first pollutant prediction information in the grid simulation domain characterizes the pollutants in the grid. i is the component number of PM2.5 or the type number of oxides. The value of i is 1, 2...10. The components of PM2.5 include seven types: sulfate, nitrate, ammonium salt, organic particulate matter, secondary organic particulate matter, black carbon and other particulate matter. The types of oxides include three types: ozone, etc.

[0139] Based on the concentration loss function, the concentration adaptive weight, and the preset constant, a concentration accuracy loss function is determined to evaluate the accuracy of the concentration prediction task. The concentration accuracy loss function is shown in Formula 1-4 below:

[0140]

[0141] Where i is the component number or oxide type number of PM2.5, σ 1i ε represents the concentration adaptive weight of the i-th PM2.5 component or oxide; the concentration adaptive weight is used in the air pollution control model to automatically adjust the weight allocation according to different concentration-related factors; ε is a preset constant.

[0142] Based on the concentration change loss function, the adaptive weight of concentration change, and the preset constant, a concentration change accuracy loss function is determined to evaluate the accuracy of the concentration change prediction task. The concentration change accuracy loss function is shown in Formula 1-5 below:

[0143]

[0144] σ 2i The adaptive weighting is the concentration change of the i-th PM2.5 component or oxide type. The adaptive weighting is used in the air pollution control model to automatically adjust the weight allocation according to the relevant factors of different concentration changes.

[0145] The total loss function Ltotal consists of three components: Lconc, which evaluates the accuracy of concentration prediction; LΔconc, which evaluates the accuracy of concentration change prediction; and a logarithmic penalty term. The total loss function Ltotal is used as the model loss function, as shown in the formula...

[0146] As shown in Equation 1-6:

[0147]

[0148] Wherein, the subscripts "conc" and "Δconc" represent the terms of concentration and concentration change, respectively; σ is an adaptive weight automatically learned by the air pollution control model, where σ conc It is a concentration-adaptive weighting. It is an adaptive weight for concentration changes; ε is a small constant, for example, 0.01. ε acts as a constraint to prevent the denominator from getting too close to zero.

[0149] Supervised learning is carried out using the total loss function (model loss function) as the objective to dynamically adjust various parameters of the air pollution control model and obtain the optimal air pollution control model.

[0150] In some embodiments, various methods can be used to evaluate the performance of air pollution control models. For example, test samples are obtained, including meteorological condition information, pollutant emission samples, and historical pollutant information, with the test samples labeled as the first pollutant prediction information; the trained air pollution control model is then validated based on the test samples. The validation results using the training sample set are shown below. Figure 8 As shown, the AI ​​model is the air pollution control model disclosed in this paper.

[0151] The model can be used to extrapolate the atmospheric pollution control model based on the test samples, and the model prediction results can be comprehensively compared with the simulated true values ​​of the WRF-CMAQ model under various control scenarios. The model can be tested in different regions, at different times, with different particulate matter components and at different concentration levels.

[0152] Several application scenarios can be constructed based on actual environmental decision-making needs to assess whether the air pollution control model can support control applications. Three emission control scenarios with different intensities during periods of heavy pollution and at the urban scale are designed. The WRF-CMAQ model and the trained air pollution control model are used to predict pollutant prediction information for each of the three scenarios. This assesses whether the air pollution control model can support practical applications in environmental fields such as pollution emergency management. The verification results are as follows: Figure 9 As shown, the AI ​​model is the air pollution control model disclosed in this paper.

[0153] The training method for the air pollution control model in the above embodiments enables the prediction of pollutants such as PM2.5 and the control of pollutant emissions through a well-trained air pollution control model. This significantly reduces the demand for computing resources, improves computing efficiency, provides basic model tools and scientific decision support for pollution control, and enhances the accuracy of pollutant prediction. It can be applied to fields such as environmental management, policy making, and pollution prevention and control, thereby improving the scientific nature and precision of pollution control.

[0154] Figure 10 The following are schematic flowcharts illustrating some embodiments of the air pollution information prediction method according to this disclosure, such as... Figure 10 As shown:

[0155] Step S1101: Generate model input data based on meteorological conditions, atmospheric emissions, and historical pollutant information.

[0156] Meteorological condition information can be generated using various methods. For example, meteorological models such as the WRF model can be used to obtain meteorological condition information based on meteorological analysis data such as meteorological reanalysis data, observational assimilation data, and topographic data.

[0157] Atmospheric emission information can be obtained using various methods, including emission samples from control scenarios. For example, atmospheric emission information can be obtained by merging anthropogenic and natural source pollutant emission inventories; anthropogenic pollutant emission information can be sampled from the atmospheric emission information to obtain baseline scenario emission samples; atmospheric emission control information can be obtained based on anthropogenic pollutant emission information; and control scenario emission samples can be obtained by sampling the atmospheric emission control information.

[0158] Step S1102: Using an air pollution control model, obtain pollutant prediction information based on the model input data.

[0159] The air pollution control model is trained using the training method described in any of the above embodiments. The pollutant prediction information output by the air pollution control model includes the concentrations and concentration changes of PM2.5 and oxides, as well as the components of PM2.5 and the types of oxides.

[0160] The atmospheric pollution information prediction method disclosed herein can improve the simulation speed of atmospheric pollution control models to the minute level, support the simulation optimization of PM2.5 air quality continuous improvement paths and the optimization of emergency control plans for heavy pollution at the level of tens of thousands of sets of data down to the geographic grid. It can be widely used in environmental management, policy making and pollution prevention and control, and can serve research institutions to formulate refined pollution control strategies and optimize air quality improvement paths. Government ecological and environmental protection agencies can use atmospheric pollution control models to conduct PM2.5 pollution control scenario simulation and policy evaluation, which will help improve the scientificity and accuracy of PM2.5 pollution control in my country and promote the development of intelligent pollution control.

[0161] In some embodiments, such as Figure 11 As shown, this disclosure provides a training device 1200 for an air pollution control model, including a first data acquisition module 1201, a sample generation module 1202, a second data acquisition module 1203, a loss determination module 1204, and a model adjustment module 1205.

[0162] The first data acquisition module 1201 uses an air quality model to obtain the first pollutant prediction information based on meteorological conditions and pollutant emission samples. The sample generation module 1202 generates training samples, which include meteorological conditions, pollutant emission samples, and historical pollutant information. The training samples are labeled with the first pollutant prediction information.

[0163] The second data acquisition module 1203 uses the air pollution control model to obtain the prediction information of the second pollutant based on the training samples. The loss determination module 1204 determines the model loss function based on the prediction information of the first and second pollutants. The model adjustment module 1205 adjusts the air pollution control model according to the model loss function until the function value of the model loss function is less than a preset threshold.

[0164] like Figure 12 As shown, this disclosure provides a training device 1200' for an air pollution control model, which includes all the modules of the training device 1200 for an air pollution control model, as well as a scenario information generation module 1206 and a meteorological condition generation module 1207.

[0165] The pollutant emission samples include baseline scenario emission samples and control scenario emission samples. The scenario information generation module 1206 merges the anthropogenic pollutant emission inventory and the natural source pollutant emission inventory to obtain atmospheric emission information. The scenario information generation module 1206 samples the anthropogenic pollutant emission information in the atmospheric emission information to obtain the baseline scenario emission samples. The scenario information generation module 1206 obtains atmospheric emission control information based on the anthropogenic pollutant emission information in the atmospheric emission information. The scenario information generation module 1206 samples the atmospheric emission control information to obtain control scenario emission samples.

[0166] For example, the scenario information generation module 1206 regulates and processes the anthropogenic pollutant emission information in the atmospheric emission information based on emission control factors, as well as spatial and temporal dimensions, to obtain atmospheric emission control information.

[0167] The scenario information generation module 1206 samples atmospheric emission control information based on spatial, temporal, and sampling dimensions to obtain control scenario emission samples. The module also samples anthropogenic pollutant emission information from atmospheric emission information based on spatial, temporal, and sampling dimensions to obtain baseline scenario emission samples. The sampling dimensions include pollution source dimensions and pollutant type dimensions. The pollution source dimension includes at least one dimension from the power, industry, and other sectors. The pollutant type dimension includes at least one dimension from sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia, and particulate matter.

[0168] The meteorological conditions generation module 1207 is used to obtain meteorological conditions information by utilizing meteorological models, meteorological reanalysis data, observation assimilation data, and topographic data.

[0169] In some embodiments, such as Figure 13 As shown, this disclosure provides an air pollution information prediction device 1400, including a data generation module 1401 and an information prediction module 1402.

[0170] The data generation module 1401 generates model input data based on meteorological conditions, atmospheric emissions, and historical pollutant information. The information prediction module 1402 uses the atmospheric pollution control model to generate pollutant prediction information based on the model input data.

[0171] like Figure 14 As shown, the electronic device may include a memory 1501, a processor 1502, a communication interface 1503, and a bus 1504. The memory 1501 is used to store instructions, and the processor 1502 is coupled to the memory 1501. The processor 1502 is configured to execute the training method of the air pollution control model or the air pollution information prediction method described above based on the instructions stored in the memory 1501.

[0172] The memory 1501 can be a high-speed RAM, non-volatile memory, or a memory array. The memory 1501 may also be divided into blocks, and these blocks can be combined into virtual volumes according to certain rules. The processor 1502 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the training method for the air pollution control model or the air pollution information prediction method of this disclosure.

[0173] In some embodiments, this disclosure provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the training method for the air pollution control model as described in any of the foregoing embodiments.

[0174] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples (not an exhaustive list) of readable storage media may include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0175] Embodiments of this disclosure may also be computer program products, including computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of this disclosure as described in the "Exemplary Methods" section above.

[0176] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0177] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0178] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0179] It should also be noted that in the apparatus, devices, and methods of this disclosure, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions to this disclosure.

[0180] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0181] The above description has been given for illustrative and descriptive purposes. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although several exemplary aspects and embodiments have been discussed above, those skilled in the art will understand that the above embodiments are illustrative only and do not limit the scope of this disclosure. Those skilled in the art will understand that the above embodiments can be combined, modified, or replaced without departing from the scope and spirit of this disclosure.

Claims

1. A training method for an air pollution control model, comprising: Using air quality models, based on meteorological conditions and pollutant emission samples, we obtain prediction information for the first pollutant. Generate training samples, wherein the training samples include the meteorological condition information, the pollutant emission samples and historical pollutant information, and the labels of the training samples are the first pollutant prediction information; Using an air pollution control model, prediction information for the second pollutant is obtained based on the training samples. Based on the first pollutant prediction information and the second pollutant prediction information, determine the model loss function; The air pollution control model is adjusted based on the model loss function. The air pollution control model includes a 3D convolutional layer, an activation function, a residual module, an encoder, a decoder, and a 2D convolutional layer. The 3D convolutional layer is used to extract 3D spatial features from meteorological variable input data and emission variable data. The activation function is used to activate the output data of the 3D convolutional layer, and the residual module establishes skip connections between the input and output of the 3D convolutional layer. The encoder downsamples and pools the output data of the activation function. The decoder upsamples the output data of the encoder. The encoder includes a 3D convolutional layer and a pooling layer, and the decoder includes multiple transposed convolutional layers. Multiple 2D convolutional layers are also included, and skip connections are established between the corresponding layers of the encoder and the decoder. The 2D convolutional layer processes the output data of the encoder to obtain pollutant characteristic information.

2. The training method as described in claim 1, wherein, The method of using an air pollution control model to obtain prediction information for the second pollutant based on the training samples includes: Based on the meteorological conditions information, the meteorological variable input data is generated; The emission variable data are generated based on the pollutant emission samples; Based on the pollutant information of the historical state, generate historical state background data; Using the aforementioned air pollution control model, the pollutant characteristic information is obtained based on the meteorological variable input data and the emission variable data, wherein the pollutant characteristic information includes pollutant concentration change information; Using the aforementioned air pollution control model, the pollutant concentration change information and the historical background data are summed to obtain the pollutant concentration information; Obtain the second pollutant prediction information, wherein the second pollutant prediction information includes the pollutant concentration information and the pollutant concentration change information.

3. The training method as described in claim 2, wherein, The step of using the air pollution control model to sum the pollutant concentration change information and the historical background data to obtain the pollutant concentration information includes: The pollutant concentration change information output from the 2D convolutional layer is summed with the historical background data to obtain the pollutant concentration information.

4. The training method as described in claim 1, wherein, Both the first and second pollutant prediction information include pollutant concentration information and pollutant concentration change information. The adaptive weights include concentration adaptive weights and concentration change adaptive weights. Determining the model loss function based on the first pollutant prediction information, the second pollutant prediction information, and the adaptive weights of the air pollution control model includes: Based on the pollutant concentration information in the first pollutant prediction information and the pollutant concentration information in the second pollutant prediction information, a concentration loss function is determined; Based on the pollutant concentration change information in the first pollutant prediction information and the pollutant concentration change information in the second pollutant prediction information, determine the concentration change loss function; The concentration accuracy loss function is determined based on the concentration loss function, the concentration adaptive weight, and the preset constant; The concentration change accuracy loss function is determined based on the concentration change loss function, the concentration change adaptive weight, and the preset constant. The model loss function is determined based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weight, and the concentration change adaptive weight.

5. The training method as described in claim 4, wherein, The step of determining the model loss function based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration adaptive weight, and the concentration change adaptive weight includes: The penalty information is determined based on the concentration adaptive weight and the concentration change adaptive weight; The model loss function is determined based on the concentration accuracy loss function, the concentration change accuracy loss function, the concentration change adaptive weight, the concentration change adaptive weight, and the penalty information.

6. The training method as described in claim 1, wherein, The pollutant emission samples include baseline scenario emission samples and control scenario emission samples, and the training method includes: Anthropogenic pollutant emission information from atmospheric emission data is sampled to obtain baseline scenario emission samples; Based on the anthropogenic pollutant emission information in the atmospheric emission information, atmospheric emission control information is obtained; The atmospheric emission control information is sampled to obtain emission samples under control scenarios.

7. The training method as described in claim 6, wherein, The step of obtaining atmospheric emission control information based on anthropogenic pollutant emission information in the atmospheric emission information includes: Based on emission control factors, as well as spatial and temporal dimensions, the anthropogenic pollutant emission information in the atmospheric emission information is controlled and processed to obtain the atmospheric emission control information.

8. The training method as described in claim 7, wherein, The step of sampling anthropogenic pollutant emission information from the atmospheric emission control information to obtain emission samples for the control scenario includes: Based on the spatial dimension, the temporal dimension, and the sampling dimension, the anthropogenic pollutant emission information in the atmospheric emission control information is sampled to obtain the emission sample of the control scenario. The sampling of anthropogenic pollutant emission information from atmospheric emission information to obtain baseline scenario emission samples includes: Based on the spatial dimension, the temporal dimension, and the sampling dimension, the anthropogenic pollutant emission information in the atmospheric emission information is sampled to obtain the baseline scenario emission sample. The sampling dimensions include pollution source dimension and pollutant type dimension. The pollution source dimension includes at least one dimension from the power, industry and other sectors. The pollutant type dimension includes at least one dimension from sulfur dioxide, nitrogen oxides, non-methane volatile organic compounds, ammonia and particulate matter.

9. The training method as described in claim 1, comprising: The meteorological conditions information is obtained using a meteorological model based on meteorological reanalysis data, observational assimilation data, and topographic data.

10. The training method according to any one of claims 1 to 9, wherein, The historical pollutant information, the first pollutant prediction information, and the second pollutant prediction information include: the concentration of PM2.5 and oxides, the amount of concentration change, and the components of PM2.5 and the types of oxides.

11. A method for predicting air pollution information, comprising: The model input data is generated based on meteorological conditions, atmospheric emissions, and historical pollutant information. Using an air pollution control model, pollutant prediction information is obtained based on the input data of the model. The air pollution control model is obtained by training using the training method described in any one of claims 1 to 10.

12. A training device for an air pollution control model, comprising: The first data acquisition module is used to obtain the first pollutant prediction information based on meteorological conditions and pollutant emission samples using an air quality model. A sample generation module is used to generate training samples, wherein the training samples include the meteorological condition information, pollutant emission samples and historical pollutant information, and the label of the training samples is the first pollutant prediction information; The second data acquisition module is used to obtain prediction information of the second pollutant based on the training samples using the air pollution control model. The loss determination module is used to determine the model loss function based on the first pollutant prediction information and the second pollutant prediction information; The model adjustment module is used to adjust the air pollution control model according to the model loss function. The air pollution control model includes a 3D convolutional layer, an activation function, a residual module, an encoder, a decoder, and a 2D convolutional layer. The 3D convolutional layer is used to extract 3D spatial features from meteorological variable input data and emission variable data. The activation function is used to activate the output data of the 3D convolutional layer, and the residual module establishes skip connections between the input and output of the 3D convolutional layer. The encoder downsamples and pools the output data of the activation function. The decoder upsamples the output data of the encoder. The encoder includes a 3D convolutional layer and a pooling layer, and the decoder includes multiple transposed convolutional layers. Multiple 2D convolutional layers are also included, and skip connections are established between the corresponding layers of the encoder and the decoder. The 2D convolutional layer processes the output data of the encoder to obtain pollutant characteristic information.

13. An air pollution information prediction device, comprising: The data generation module is used to generate model input data based on meteorological conditions, atmospheric emissions, and historical pollutant information. The information prediction module is used to generate pollutant prediction information based on the input data of the air pollution control model. The air pollution control model is obtained by training using the training method described in any one of claims 1 to 10.

14. An electronic device comprising: Memory; And a processor coupled to the memory, the processor being configured to perform the method as described in any one of claims 1-11 based on instructions stored in the memory.

15. A computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1 to 11.

16. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

Images