A global pollutant diffusion field simulation method and system based on sparse data

Abstract

The present invention provides a global pollutant diffusion field simulation method and system based on sparse data, which relates to the technical field of pollutant diffusion simulation, including: S1, collecting meteorological data, pollution source data and site monitoring concentration data at each natural time; S2, generating an initial particle set at natural time T based on meteorological data and pollution source data at natural time T and before natural time T; S3, combining the particle set at natural time T and particle time t with the meteorological data and pollution source data at natural time T, estimating the state of the particle set at natural time T and particle time t+1, and performing error evaluation. When the error meets the acceptable condition, the RF estimation field at natural time T is estimated based on the state of the particle set at natural time T and particle time t+1, and going to step S4; S5, predicting the predicted pollutant concentration field at natural time T+1 through a generator; S6, evaluating the predicted pollutant concentration field at natural time T+1 through a discriminator.

Description

Technical Field

[0001] The present invention relates to the technical field of pollutant diffusion simulation, and in particular to a method and system for simulating a global pollutant diffusion field based on sparse data. Background Art

[0002] Currently, the methods available for simulating atmospheric pollutant diffusion fields can be divided into the following three categories:

[0003] 1. Classical methods: These include Gaussian models, Lagrangian methods, CMAQ, and CALPUFF. These methods are usually based on theoretical models of physics and chemistry, assume that emission sources and meteorological conditions are relatively known, and are solved using specific mathematical formulas or equations.

[0004] 2. Machine learning / deep learning algorithms: These include attention networks, gated recurrent units, and attention mechanisms. These methods rely on data-driven models and can automatically adjust model parameters for prediction and simulation by learning the complex relationships between inputs and outputs from historical data.

[0005] 3. Statistical algorithms: These include particle filters, genetic algorithms, and simulated annealing. These are optimization algorithms typically used to solve parameter tuning problems in complex systems. They can find suitable solutions through search and optimization. In diffusion simulations, they can be used to optimize model parameters or sample and infer different hypotheses.

[0006] However, each of these three methods has certain limitations, especially for the simulation scenario of the global pollutant diffusion field. Because the monitoring stations in a large area are unevenly distributed, there is no emission information or the emission information is incomplete in some areas. In this case, pollutant diffusion simulation and estimation will face certain challenges. The lack or incomplete emission information will affect the accuracy and reliability of the model. Therefore, a single method often has limitations in accuracy, adaptability, and real-time performance. Summary of the Invention

[0007] The purpose of the present invention is to provide a global pollutant diffusion field simulation method and system based on sparse data, which mainly combines particle filtering and generative adversarial network algorithms. It can also solve the problem of global pollutant diffusion field simulation well in sparse data and complex environments.

[0008] In order to solve the above technical problems, the present invention adopts the following solutions:

[0009] A global pollutant diffusion field simulation method based on sparse data, the global pollutant diffusion field simulation method comprising the following steps:

[0010] S1. Collect meteorological data, pollution source data, and site monitoring concentration data at each natural moment, and package the meteorological data, pollution source data, and site monitoring concentration data at the same natural moment into one moment data;

[0011] S2. Generate an initial particle set at natural time T based on meteorological data and pollution source data before natural time T, where the initial particle set refers to a particle set whose particle time is t=0;

[0012] S3. The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through the diffusion model, and an error evaluation is performed. When the error meets the acceptable condition, the RF estimated field at natural time T is obtained based on the state estimation of the particle set at natural time T and particle time t+1, and the process goes to step S4.

[0013] S4. Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the predicted pollutant concentration field at natural time T+1 is obtained through the generator prediction;

[0014] S5. Obtain the real pollutant concentration field at natural time T+1, combine it with the predicted pollutant concentration field at natural time T+1, and evaluate the predicted pollutant concentration field at natural time T+1 through the discriminator.

[0015] Furthermore, the meteorological data includes pressure, wind speed, wind direction, temperature, precipitation, relative humidity, solar radiation, cloud cover, and grid ID; the pollution source data includes the location of the pollution source, emission intensity, emission height, and type of pollutants emitted; the site monitoring concentration data is the pollutant concentration data monitored by the monitoring site at each moment, including PM2.5, PM10, NO2, SO2, and CO.

[0016] Furthermore, in S2, the process of generating the initial particle set at natural time T according to the meteorological data and pollution source data at natural time T and before natural time T is as follows:

[0017] Generate a particle set at natural time T, initialize the particle set at natural time T according to meteorological data and pollution source data at natural time T and before natural time T, randomly initialize each particle with noise or initialize with prior information to obtain the initial state of the particle, and then obtain the initial particle set at natural time T.

[0018] Furthermore, in S3, the error evaluation includes the following steps:

[0019] SA1, perform error evaluation on the particle set at natural time T and particle time t+1;

[0020] SA2. After obtaining the error assessment result, update the particle set at natural time T and particle time t+1 based on the site monitoring concentration data at natural time T to obtain an updated particle set at natural time T and particle time t+1, and go to step SA3.

[0021] SA3. Judge the error assessment results:

[0022] When the error meets the acceptable conditions, the RF estimation field of the natural time T is obtained based on the updated particle set at the natural time T and the particle time t+1, and the process goes to step S4; when the error does not meet the acceptable conditions, t is updated to t', t'=t+1, and the process goes to step SA1.

[0023] Furthermore, in SA2, the process of obtaining the updated particle set at natural time T and particle time t+1 is as follows:

[0024] Update the weight of each particle in the particle set at natural time T and particle time t+1 according to the site monitoring concentration data at natural time T;

[0025] Particle resampling is performed according to the updated weights of each particle to update the weights of the particle set at natural time T and particle time t+1, thereby obtaining an updated particle set at natural time T and particle time t+1.

[0026] Furthermore, the process of updating the weight of each particle in the particle set at natural time T and particle time t+1 according to the site monitoring concentration data at natural time T is as follows:

[0027] Obtain the site monitoring concentration data at natural time T and the state of the particle set at natural time T and particle time t, compare the site monitoring concentration data at natural time T with the state of the particle set at natural time T and particle time t, and update the weight of each particle according to the comparison result.

[0028] Furthermore, the particle resampling is used to remove particles with lower weights in the particle set, and the weight of the particle set at the natural time T and the particle time t+1 is updated.

[0029] Furthermore, the process of estimating the RF estimated field at natural time T based on the updated particle set at natural time T and particle time t+1 is as follows:

[0030] The weighted average calculation is performed on the particles in the particle set at the updated natural time T and the particle time t+1, and the obtained weighted average value is used as the final concentration field, that is, the RF estimation field at the natural time T.

[0031] A global pollutant diffusion field simulation system based on sparse data, applying the global pollutant diffusion field simulation method based on sparse data, comprises:

[0032] Data acquisition module: collects meteorological data, pollution source data and site monitoring concentration data at each natural moment, and packages the meteorological data, pollution source data and site monitoring concentration data at the same natural moment into one moment data;

[0033] Particle set initialization module: generates an initial particle set at natural time T according to meteorological data and pollution source data before natural time T. The initial particle set refers to the particle set with particle time t=0;

[0034] Concentration distribution field estimation module: The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through the diffusion model. The error is then evaluated. When the error meets the acceptable conditions, the RF estimated field at natural time T is obtained based on the state of the particle set at natural time T and particle time t+1.

[0035] Generator prediction module: Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the generator predicts the pollutant concentration field at natural time T+1;

[0036] Discriminator evaluation module: obtains the actual pollutant concentration field at natural time T+1, combines it with the predicted pollutant concentration field at natural time T+1, and evaluates the predicted pollutant concentration field at natural time T+1 through the discriminator.

[0037] Beneficial effects of the present invention:

[0038] The present invention provides a global pollutant diffusion field simulation method and system based on sparse data. In the case of sparse data and uneven distribution of monitoring sites, PF-GAN can effectively compensate for data gaps between monitoring sites and fill blank areas through its flexible generative model, nonlinear modeling capabilities, adversarial training process, and combination with particle filtering. It can generate a reasonable pollutant concentration distribution field under irregular data input and provide more accurate diffusion estimation results. In addition, by alternatingly updating the particle filter and GAN, the estimation of the pollutant concentration field is gradually optimized, so that the model can fully utilize existing data to accurately simulate the pollutant diffusion field in sparse data scenarios. After each particle filtering step, the pollutant concentration field is predicted by the generator and then adversarial training is performed by the discriminator. The GAN training process can improve the accuracy of the generated pollutant concentration field, while the particle filter maintains the state estimation of the pollutant concentration field. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of a method for simulating a global pollutant diffusion field based on sparse data in Example 1 of the present invention.

[0040] Figure 2 Schematic diagram of the particle filtering process in Example 1 of the present invention.

[0041] Figure 3 Schematic diagram of the process of adversarial network training in Example 1 of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is in no way intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] Unless otherwise specifically stated, the relative arrangement of components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention.

[0044] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0045] Additionally, descriptions of well-known structures, functions, and configurations may be omitted for clarity and conciseness. Those skilled in the art will recognize that various changes and modifications can be made to the examples described herein without departing from the spirit and scope of the present disclosure.

[0046] Technologies, methods and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, such technologies, methods and equipment should be considered part of the authorization specification.

[0047] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0048] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments:

[0049] Example 1

[0050] At present, monitoring stations over a large area are usually unevenly distributed, resulting in some blank areas with no emission information or incomplete emission information. In this case, pollutant diffusion simulation and estimation will face certain challenges. The lack or incomplete emission information will affect the accuracy and reliability of the model. Therefore, a single method often has limitations in accuracy, adaptability, and real-time performance.

[0051] To address the aforementioned issues, this embodiment proposes a global pollutant diffusion field simulation method based on sparse data. Based on a nested statistical algorithm and machine / deep learning, the particle filter-generative adversarial network (PF-GAN) algorithm is employed. This method effectively addresses the data-sparse diffusion simulation problem, offering greater adaptability and real-time performance, while maintaining both accuracy and flexibility to accommodate rapidly changing environments. The PF-GAN combines particle filtering and generative adversarial networks, effectively handling the challenges posed by sparse data without being constrained by gridding. Therefore, it offers significant advantages in this specific scenario, as follows:

[0052] (1) No gridding requirement, flexible response to uneven data distribution: PF-GAN does not rely on a regular grid structure and can directly process data with sparse data and uneven distribution of monitoring sites. Traditional grid-based methods, such as PF-CNN, require that data must be regularly gridded, which will cause many difficulties in sparse data environments. PF-GAN, by generating a model, can generate a reasonable pollutant concentration field under irregular data input and fill the data gaps between sites.

[0053] (2) Using adversarial training to fill data gaps: The core advantage of the generative adversarial network (GAN) lies in its generative ability. The generator network can generate concentration fields based on existing sparse data, meteorological and pollution source information. The adversarial training process between the generator and the discriminator can make the generated data closer and closer to the actual pollutant concentration distribution. Even if the original data points are sparse, the GAN can learn the global characteristics of the data and generate a smooth and reasonable concentration field to fill the gaps in the sparse data area.

[0054] (3) Efficient modeling of nonlinear relationships: The pollutant diffusion process is usually highly nonlinear and is affected by a variety of complex factors (such as meteorological conditions, geographical environment, emission sources, etc.). Especially in the generator part, this nonlinear relationship can be captured through a complex neural network structure. It is more flexible than traditional physical model-based methods (such as Gaussian diffusion model) and can automatically learn the complex patterns in the pollutant diffusion process.

[0055] (4) Enhanced adaptability to observed data through particle filtering: Particle filtering (PF) can dynamically estimate the state of the pollutant concentration field and update it using observed data. By updating the weights of particles, PF can enhance the adaptability to the uncertainty and dynamic changes of observed data, especially in environments with sparse data and high noise. When combined with GAN, particle filtering can help the generator generate pollutant concentration fields more accurately, ensuring that the generated results conform to the distribution characteristics of the observed data.

[0056] (5) Improve simulation accuracy and generate diverse simulation results: PF-GAN can generate multiple different concentration field simulation results, which provides a diverse perspective and uncertainty assessment for the diffusion prediction of atmospheric pollutants. This is very important in practical applications because the diffusion of atmospheric pollutants is highly uncertain. Generating multiple simulation results helps provide more comprehensive decision support.

[0057] (6) Enhanced model robustness and generalization capabilities: PF-GAN can effectively avoid overfitting through generative adversarial training, and enhance the model's generalization capabilities in the face of sparse data or different types of pollutant diffusion fields. In particular, when the site distribution is uneven and the data points are scarce, GAN can learn the latent spatial structure extracted from the uneven data, which can be better applied to diffusion simulations in other regions and time periods.

[0058] In summary, in the case of sparse data and uneven distribution of monitoring sites, PF-GAN can effectively compensate for data gaps between monitoring sites and fill blank areas through its flexible generative model, nonlinear modeling capabilities, adversarial training process, and combination with particle filtering. It can generate reasonable pollutant concentration fields under irregular data input and provide more accurate diffusion prediction results.

[0059] Specifically, a global pollutant diffusion field simulation method based on sparse data, such as Figure 1 As shown, the method includes the following steps:

[0060] S1. Collect meteorological data, pollution source data, and site monitoring concentration data at each natural moment, and package the meteorological data, pollution source data, and site monitoring concentration data at the same natural moment into one moment data;

[0061] S2. Generate an initial particle set at natural time T based on meteorological data and pollution source data before natural time T, where the initial particle set refers to a particle set whose particle time is t=0;

[0062] S3. The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through the diffusion model, and an error evaluation is performed. When the error meets the acceptable condition, the RF estimated field at natural time T is obtained based on the state estimation of the particle set at natural time T and particle time t+1, and the process goes to step S4.

[0063] S4. Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the predicted pollutant concentration field at natural time T+1 is obtained through the generator prediction;

[0064] S5. Obtain the real pollutant concentration field at natural time T+1, combine it with the predicted pollutant concentration field at natural time T+1, and evaluate the predicted pollutant concentration field at natural time T+1 through the discriminator.

[0065] The meteorological data are: , where the dimension of meteorological data is: meteorological data at time T, including pressure, wind speed, wind direction, temperature, precipitation, relative humidity, solar radiation, cloud cover, grid ID and other D-dimensional features.

[0066] The pollution source data are: , where the dimension of pollution source data is: pollution source information at time T, including location, emission intensity, etc., with a total of S-dimensional features.

[0067] The monitoring concentration data of the site are: , where the dimension of the site monitoring concentration data is: pollutant concentration data of M sites at time T, including PM2.5, PM10, NO2, SO2, CO and other parameters.

[0068] The time series data is: T time data { , , }, the meteorological data, pollution source data and site monitoring concentration data at the same time are packaged into one moment data.

[0069] After data collection, it was found that the distribution of monitoring stations was often uneven under the global pollutant diffusion field simulation, and data in some areas was scarce or completely missing. Traditional grid-based methods, such as PF-CNN, require that data must be regularly gridded, which will cause many difficulties in such a sparse data environment. Therefore, in this invention, a particle filter (PF) is first used to dynamically estimate the state of the pollutant concentration field and update it using the observed data.

[0070] The process of dynamically estimating the state of the pollutant concentration field using particle filtering (PF), such as Figure 2 As shown in Figure 2, the first step is to initialize the particle set. First, generate a particle set at natural time T. Initialize the particle set at natural time T according to the meteorological data and pollution source data before natural time T. Randomly initialize each particle with noise or initialize it with prior information to obtain the initial state of the particle. Then, obtain the initial particle set at natural time T, where each particle represents a possible pollutant concentration field. After initializing the particle set, multiple possible states of the pollutant concentration field can be obtained. The specific process is as follows:

[0071] Input data: meteorological data at the initial time t=0 and pollution source data , that is, the particle set is initialized with the meteorological data and pollution source data at the initial time t=0;

[0072] Get the initialized particle set: ; Each particle , represents a possible pollutant concentration field, is the initial state of the ith particle, and N is the total number of particles.

[0073] The second step is particle prediction, which mainly predicts the pollutant concentration field of particles at the next moment based on meteorological data and pollution source data. Here, it is necessary to first obtain the particles at the current moment T from the particle set. and its corresponding meteorological data and pollution source data , and then the state of the particle set at particle time t=1 is predicted by the diffusion model. The specific process is as follows:

[0074] Input data: particles at the current moment , meteorological data and pollution source data , that is, obtaining the meteorological data and pollution source data of particles from the particle set;

[0075] Prediction step: For each particle , update the particle concentration field according to the diffusion model f(·), that is, sequentially input the meteorological data and pollution source data of the particles in the initial particle set at natural time T and particle time t=0 into the diffusion model, and predict the state of the particles at the next time through the diffusion model to obtain the state of the particle set at particle time t=1, that is, the concentration field of the particle set at particle time t=1. Specifically, the prediction is: ,in, Represents the updated particle concentration field.

[0076] The third step is particle weighting. By comparing the concentration field predicted by the particle with the actual observation data, the particle weight is updated. The update of the particle weight reflects the degree of fit of the particle to the observation data. The particle with a higher weight indicates that the prediction of the concentration field is more accurate. Specifically, the weight of each particle in the particle set at natural time T and particle time t+1 is updated according to the site monitoring concentration data at natural time T. The process is as follows:

[0077] Input data: site monitoring concentration data at the current time T , according to the concentration field predicted at the current time T ;

[0078] Weighting step: The weight of the particle is calculated by the likelihood function: ; Among them, the weight of each particle , which indicates the degree of match between each particle state and the concentration data monitored at the site.

[0079] The fourth step is the resampling step, which mainly resamples according to the weight of the particles to generate a new particle set. This not only removes particles with lower weights, but also ensures that particles with higher weights are more likely to be selected, which has a greater impact on the prediction of the next moment. Specifically, the particles are resampled according to the updated weights of each particle to update the weights of the particle set at natural time T and particle time t+1. The process is as follows:

[0080] Input data: current particle set , particle weight ;

[0081] Resampling step: Resample according to the particle weights to generate a new particle set: ; Among them, the dimension of the new particle set is the same as that of the particle set, both are R M .

[0082] The fifth step is the result output, which mainly calculates the concentration field prediction result at the current time based on the weighted average of the particles. Specifically, the RF estimated field at the natural time T is obtained based on the updated particle set at the natural time T and the particle time t+1. The concentration distribution field at the current time t can be estimated through particle filtering to provide input for the subsequent GAN generation process. The process is as follows:

[0083] Output data: The weight of each particle in the updated particle set is calculated by weighted average, and the weighted average of the particles is estimated as the final pollutant concentration field, that is, the concentration distribution field of the particle set at the current time t. The formula is: ; Among them, the concentration distribution field at the current time T is, Represents the predicted results of the pollutant concentration field at the current time T.

[0084] After obtaining the predicted results of the pollutant concentration field at the current time T through particle filtering, generative adversarial network (GAN) training is selected to generate the future pollutant concentration field. The adversarial network is trained adversarially through the generator and discriminator, which can make the generated pollutant concentration field closer to the true value.

[0085] like Figure 3 As shown, first, a generator is constructed. The generator can generate the pollutant concentration field at the next moment based on the pollutant concentration field, meteorological data, and pollution source data at the current moment. Specifically, the generator generates the concentration distribution field of the particle set at the next moment T+1 based on the updated concentration distribution field of the particle set at the current moment T, meteorological data, and pollution source data. The generator generates a near-real future pollutant concentration field through adversarial training. The process is as follows:

[0086] Input data: prediction results of pollutant concentration field output by particle filter , meteorological data , pollution source data ;

[0087] Generator model: Generator G generates the predicted pollutant concentration field at the next time T+1 based on the input data: ;in, It represents the generated pollutant concentration field, that is, the predicted pollutant concentration field at the next time T+1.

[0088] Then, a discriminator is constructed to evaluate whether the generated pollutant concentration field is realistic and matches the actual observed data. Specifically, the discriminator obtains the predicted pollutant concentration field of the particle set at the next time T+1 and determines whether the predicted pollutant concentration field at the next time T+1 matches the actual pollutant concentration field at the next time T+1. The discriminator makes the pollutant concentration field generated by the generator closer to the actual value through adversarial training with the generator. The process is as follows:

[0089] Input data: The predicted pollutant concentration field generated at the next moment T+1 And the actual pollutant concentration field at the next moment T+1 ;

[0090] Discriminator model: The discriminator D outputs a value indicating whether the input concentration field is real ,in, Represents the discriminator's evaluation of the authenticity of the generated results.

[0091] Next, the generator and discriminator are optimized based on the matching results. Specifically, by optimizing the loss functions of the generator and discriminator, the pollutant concentration field generated by the generator is made closer to the true value. The process is as follows:

[0092] Loss function (generator): ;

[0093] Among them, L G ∈R, represents the loss of the generator;

[0094] Loss function (discriminator):

[0095] ;

[0096] Among them, L D ∈R, represents the loss of the discriminator.

[0097] Finally, the accuracy of the model can be evaluated by comparing the actual monitoring data and simulation results based on the error between the calculated pollutant concentration field and the actual observation data, such as the mean square error (MSE). For example: , the pollutant concentration field that can be simulated and the actual pollutant concentration field Error assessment.

[0098] Example 2

[0099] A global pollutant diffusion field simulation system based on sparse data, applying the global pollutant diffusion field simulation method based on sparse data, comprises:

[0100] Data acquisition module: collects meteorological data, pollution source data and site monitoring concentration data at each natural moment, and packages the meteorological data, pollution source data and site monitoring concentration data at the same natural moment into one moment data;

[0101] Particle set initialization module: generates an initial particle set at natural time T according to meteorological data and pollution source data before natural time T. The initial particle set refers to the particle set with particle time t=0;

[0102] Concentration distribution field estimation module: The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through the diffusion model. The error is then evaluated. When the error meets the acceptable conditions, the RF estimated field at natural time T is obtained based on the state of the particle set at natural time T and particle time t+1.

[0103] Generator prediction module: Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the generator predicts the pollutant concentration field at natural time T+1;

[0104] Discriminator evaluation module: obtains the actual pollutant concentration field at natural time T+1, combines it with the predicted pollutant concentration field at natural time T+1, and evaluates the predicted pollutant concentration field at natural time T+1 through the discriminator.

[0105] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Based on the technical essence of the present invention and within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement of the above embodiment shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A global pollutant diffusion field simulation method based on sparse data, characterized by: The global pollutant diffusion field simulation method comprises the following steps: S1. Collect meteorological data, pollution source data, and site monitoring concentration data at each natural moment, and package the meteorological data, pollution source data, and site monitoring concentration data at the same natural moment into one moment data; S2. Generate an initial particle set at natural time T based on meteorological data and pollution source data before natural time T, where the initial particle set refers to a particle set whose particle time is t=0; S3. The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through a diffusion model, and an error assessment is performed. The error assessment includes the following steps: SA1, perform error evaluation on the particle set at natural time T and particle time t+1; SA2. After obtaining the error assessment result, update the particle set at natural time T and particle time t+1 based on the site monitoring concentration data at natural time T to obtain an updated particle set at natural time T and particle time t+1, and go to step SA3. SA3. Judge the error assessment results: When the error satisfies the acceptable condition, the concentration distribution field at the natural time T is estimated based on the updated particle set at the natural time T and the particle time t+1, and the process goes to step S4; wherein, the process of estimating the concentration distribution field at the natural time T based on the updated particle set at the natural time T and the particle time t+1 is as follows: performing a weighted average calculation on the particles in the updated particle set at the natural time T and the particle time t+1, and using the obtained weighted average as the final pollutant concentration field, that is, the concentration distribution field at the natural time T; When the error does not meet the acceptable condition, update t to t', t'=t+1, and go to step SA1; S4. Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the predicted pollutant concentration field at natural time T+1 is obtained through the generator prediction; S5. Obtain the actual pollutant concentration field at natural time T+1, combine it with the predicted pollutant concentration field at natural time T+1, and evaluate the predicted pollutant concentration field at natural time T+1 through the discriminator.

2. The method for simulating global pollutant diffusion fields based on sparse data according to claim 1, characterized in that: The meteorological data includes pressure, wind speed, wind direction, temperature, precipitation, relative humidity, solar radiation, cloud cover, and grid ID; the pollution source data includes the location of the pollution source, emission intensity, emission height, and type of pollutants emitted; the site monitoring concentration data is the pollutant concentration data monitored by the monitoring site at each moment, including PM2.5, PM10, NO2, SO2, and CO.

3. The method for simulating global pollutant diffusion fields based on sparse data according to claim 1, characterized in that: In S2, the process of generating the initial particle set at natural time T based on the meteorological data and pollution source data at natural time T and before natural time T is as follows: Generate a particle set at natural time T, initialize the particle set at natural time T according to meteorological data and pollution source data at natural time T and before natural time T, randomly initialize each particle with noise or initialize with prior information to obtain the initial state of the particle, and then obtain the initial particle set at natural time T.

4. The method for simulating global pollutant diffusion fields based on sparse data according to claim 1, characterized in that: In SA2, the process of obtaining the updated particle set at natural time T and particle time t+1 is as follows: Update the weight of each particle in the particle set at natural time T and particle time t+1 according to the site monitoring concentration data at natural time T; Particle resampling is performed according to the updated weights of each particle to update the weights of the particle set at natural time T and particle time t+1, thereby obtaining an updated particle set at natural time T and particle time t+1.

5. The method for simulating global pollutant diffusion fields based on sparse data according to claim 4 is characterized in that: The process of updating the weight of each particle in the particle set at natural time T and particle time t+1 according to the site monitoring concentration data at natural time T is as follows: Obtain the site monitoring concentration data at natural time T and the state of the particle set at natural time T and particle time t, compare the site monitoring concentration data at natural time T with the state of the particle set at natural time T and particle time t, and update the weight of each particle according to the comparison result.

6. The method for simulating global pollutant diffusion fields based on sparse data according to claim 4, characterized in that: The particle resampling is used to remove particles with lower weights in the particle set, and the weights of the particle set at the natural time T and the particle time t+1 are updated.

7. A global pollutant diffusion field simulation system based on sparse data, characterized by: Applying a global pollutant diffusion field simulation method based on sparse data as described in any one of claims 1 to 6, comprising: Data acquisition module: collects meteorological data, pollution source data and site monitoring concentration data at each natural moment, and packages the meteorological data, pollution source data and site monitoring concentration data at the same natural moment into one moment data; Particle set initialization module: generates an initial particle set at natural time T according to meteorological data and pollution source data before natural time T. The initial particle set refers to the particle set with particle time t=0; Concentration distribution field estimation module: The particle set at natural time T and particle time t is combined with the meteorological data and pollution source data at natural time T, and the state of the particle set at natural time T and particle time t+1 is estimated through the diffusion model. The error is then evaluated. When the error meets the acceptable conditions, the RF estimated field at natural time T is obtained based on the state of the particle set at natural time T and particle time t+1. Generator prediction module: Based on the concentration distribution field at natural time T and before natural time T, combined with the meteorological data and pollution source data at natural time T and before natural time T, the generator predicts the pollutant concentration field at natural time T+1; Discriminator evaluation module: obtains the actual pollutant concentration field at natural time T+1, combines it with the predicted pollutant concentration field at natural time T+1, and evaluates the predicted pollutant concentration field at natural time T+1 through the discriminator.

Images