CN114841066A - A dust particle drift trajectory model and its modeling method - Google Patents

Abstract

The invention discloses a flying dust particle drift track model and a modeling method thereof, wherein the model comprises the following steps: 1) a data preprocessing stage: preprocessing data; 2) a model pre-training stage: dividing grids according to the geographical position of the station equipment; 3) a model pre-training stage: model pre-training is carried out through simulation data and a traditional physical method; 4) a transfer learning fine adjustment stage: and predicting the real flying dust particle drift track by using the trained model. The invention introduces the concept of physical field, replaces the calculation of speed field by the network of codec structure, and optimizes the calculation of flow term, diffusion term and volume force term. Experiments show that the raise dust track prediction model provided by the invention has greatly improved calculation speed and calculation stability compared with the traditional method, and has better universality and interpretability compared with a neural network method.

Description

A dust particle drift trajectory model and its modeling method

technical field

The invention relates to the technical field of dust trajectory prediction, in particular to a dust particle drift trajectory model and a modeling method thereof.

Background technique

As the speed of urban and social development continues to accelerate, more and more people are beginning to pay attention to environmental and residential issues. Air pollution is still a problem for many developing countries at this stage, and the latest GBD (Global Burden of Disease) analysis continues to identify air pollution as one of the most important risk factors for death and disability. Atmospheric particulate matter, a component of air pollution, is listed as the sixth-leading risk factor for premature death.

Most of the dust trajectory prediction models are based on traditional physical models or neural networks. Most of the traditional methods are based on experimental field research, numerical simulation research, meteorological field analysis research and data mining methods. These traditional physical models have cumbersome calculation steps. The volume is very large and tends to be unstable; neural network-based methods such as convolutional neural network and recurrent neural network prediction have their specific application areas, and they also lack interpretability.

The invention introduces the concept of the field, replaces the calculation of the velocity field in the traditional method with the network of the codec structure, and optimizes the calculation of the convection term, the diffusion term and the body force term at the same time. Experiments show that the dust trajectory prediction model proposed by the present invention has greatly improved the accuracy and calculation speed compared with the traditional method, and at the same time improves the stability of calculation, and has better generality and interpretability than the neural network method.

SUMMARY OF THE INVENTION

The invention mainly aims at the problems of heavy calculation tasks, lack of precision and difficulty in calculating the velocity field of the dust movement in the traditional method, improves the relevant traditional physical algorithm and combines the neural network method, and proposes a dust particle drift trajectory model and a modeling method.

The technical scheme of the present invention is as follows:

The dust data used in the present invention includes the following information: location information of the equipment site, equipment number, equipment longitude, latitude, equipment data status, temperature, humidity, wind speed, wind direction, PM2.5 concentration, and pressure.

A drift trajectory model of fugitive dust particles and a modeling method thereof, characterized in that it comprises the following steps:

1) Data preprocessing stage: preprocessing data;

2) Model pre-training stage: divide the grid according to the geographical location of the site equipment;

3) Model pre-training stage: model pre-training through simulated data and traditional physical methods;

4) Transfer learning fine-tuning stage: use the trained model to predict the real drift trajectory of dust particles.

Further, 1) Data preprocessing: In order to improve the accuracy of model prediction, the following data preprocessing needs to be carried out:

1.1) Data deduplication: remove the duplicate data uploaded during the device collection process by connecting table query;

1.2) Wind speed processing: Take the latitude and longitude of the earth as the coordinate axis, and calculate the longitudinal component u of the wind speed and the lateral component v of the wind speed according to the wind speed and wind direction parameters in the dust data;

1.2.1) Calculate the angle θ between it and the coordinate axis of the geodetic plane according to the wind direction;

1.2.2) Calculate the wind speed V Calculate the transverse component u and the longitudinal component v:

Transverse component u=Vcos(θ); longitudinal component v=Vsin(θ);

1.3) Longitude and latitude processing: In order to reduce the influence of station longitude and latitude errors and thereby improve the accuracy of the prediction results, the present invention adopts Gaussian projection algorithm to convert (L, B) under ordinary geodetic coordinates into (x, y) under Gaussian coordinates;

1.3.1) Calculate the central meridian longitude, band number and longitude difference (the distance from the point to the central meridian) according to the latitude and longitude. The 3° central meridian is selected for this model, and the calculation process is as follows:

1.3.1.1) Calculate the 3° band n ₃ : n ₃ =int(L/3+0.5);

1.3.1.2) Calculate the central meridian L ₀ of 3° with n ₃ : L ₀ =3n ₃ ;

1.3.1.3) Calculate the longitude difference l (the distance between the station and the central meridian): l=LL ₀ .

1.3.2) Calculate the meridian arc length X according to the selection of the ellipsoid, and this model selects the Krasovsky ellipsoid for calculation (which can be applied to the 54 Beijing coordinate system):

X=111134.861B-16036.480 sin 2B+16.828 sin 4B-0.022 sin 6B;

1.3.3) Calculate the value of each parameter in the positive calculation formula of Gaussian coordinates, wherein, B is the geodetic latitude, t is the tangent of the geodetic dimension, e is the eccentricity of the ellipse, N is the radius of curvature of the Maoxi circle, and a is the long radius of the ellipsoid parameter:

1.3.4) Substitute the parameters obtained by the above calculation into the Gaussian coordinate formula to find the coordinates (x, y) in the Gaussian coordinate system:

x=X+N sin B cos Bl ² +N sin B cos ³ B(5-t ² -9e ² cos ² B)l ⁴

y=N cos Bl+N cos ³ B(1-t ² +e ² cos ² B)l ³ +N cos ⁵ B(5-18t ² +t ⁴ )l ⁵

Further, 2) grid processing:

In the dust movement, the spatial interaction of adjacent stations is very close, so the present invention no longer studies the dust laws of a single station, but from the overall point of view, through the velocity field, concentration field, temperature field, pressure field , humidity field to study the distribution and variation of dust in space. To describe the physics, further meshing is required based on the geographic location of the site equipment.

2.1) Cluster analysis: In order to improve the accuracy of prediction, before dividing the grid, the present invention performs DBScan cluster analysis according to the physical location of the site equipment, and obtains the site equipment cluster cluster under Gaussian coordinates;

2.2) Grid division: After obtaining a cluster with better effect, divide a uniform square grid according to the geographical location of the site equipment in the cluster;

2.3) Data interpolation: Linear interpolation is performed according to the distance between equipment sites and the dust data value, and finally the dust data is stored in the grid in the form of velocity field, concentration field, temperature field, pressure field, and humidity field.

Further, 3) model pre-training:

The two mainstream methods currently used to predict the drift trajectory of dust particles are the forward calculation method based on traditional physical methods and the neural network method based on numerical simulation, but both methods have certain drawbacks: the calculation of traditional physical methods is too complicated, The steps are cumbersome; neural network methods lack interpretability and it is generally difficult to obtain real velocity fields as training labels. Based on this, the present invention provides a drift trajectory model of fugitive dust particles and a modeling method thereof.

3.1) Generate simulation data: simulate and generate dust simulation data at time T ₀ in the divided grid, including dust velocity field, concentration field, temperature field, pressure field and humidity field;

3.2) Forward calculation based on traditional physical methods: On the basis of the simulated data, the body force term is updated according to the traditional physical method (mainly based on the NS Navier-Stokes equation), and the calculation of the convection term and diffusion term is performed forward Calculate to obtain the augmented data set of the fugitive dust physics field of the subsequent time nodes T ₁ , T ₂ , T ₃ , ..., T _n , T _n+1 :

3.2.1) N-S equations: N-S Navier-Stokes equations (Navier-Stokes equations) are the equations of fluid motion that describe the conservation of fluid momentum. The specific formula is as follows:

是流体关于速度场的物质导数。欧拉视角主要关注的是空间中的网格点，因而物质导数

等于网格点上流体速度随时间的变化率

与流体在速度场作用下变化率u·▽u之和。其他参数方面：ρ是流体密度；u是速度矢量；p是压力；f是单位体积流体受的外力，μ是流体粘度。in,

is the material derivative of the fluid with respect to the velocity field. Euler's perspective is mainly concerned with grid points in space, so the matter derivative

is equal to the rate of change of fluid velocity over time at grid points

and the sum of the rate of change u·▽u of the fluid under the action of the velocity field. Other parameters: ρ is the fluid density; u is the velocity vector; p is the pressure; f is the external force per unit volume of the fluid, and μ is the fluid viscosity.

的计算后的得到状态W₂，最后经过扩散项(μ▽²u)的计算后得到状态W₃。The above formula is the update process of the fluid state, W ₀ represents the initial state of the particle, the state after the calculation of the body force term (f) is W ₁ , and after the convection term

The state W ₂ is obtained after the calculation of , and finally the state W ₃ is obtained after the calculation of the diffusion term (μ▽ ² u).

3.2.2) Body force term: The calculation formula of the body force term is f=ρg (g is the acceleration of gravity, ρ is the fluid density).

网格点上粒子对应的速度场横向分量为u，纵向分量为v,求解1个时间步长Δt后网格点物理量场值

的步骤为：3.2.3) Convective term: In fluid simulation, the convection term is automatically executed. The essence of convection is that under the action of different velocity fields, adjacent grid points transfer fluid micelles to each other and update the grid points. The physical quantity value of , given a grid point position (x, y), the physical quantity field value of this point is

The transverse component of the velocity field corresponding to the particle on the grid point is u, and the longitudinal component is v. After solving one time step Δt, the physical quantity field value of the grid point is

The steps are:

3.2.3.1) Track backward for 1 time step to get the grid position of the particle (x-uΔt, y-vΔt);

3.2.3.2) Construct the Catmull-Rom spline interpolation function according to the physical field grid to calculate the mean value of the change of the particle physical quantity field

的计算公式为：

3.2.3.3) Final

The calculation formula is:

已知，现需要求解第n时刻的物理量

其具体计算步骤为：3.2.4) Diffusion term: The calculation of the diffusion term is another very important part of fluid simulation. The calculation formula of the diffusion term of the fluid is a _v = μ▽ ² u, where a _v is the acceleration of the particle under the action of the fluid viscous force, and μ is the viscosity coefficient of the fluid. Assuming the n+1th moment, the physical quantity at the grid coordinates (x, y)

Known, now we need to solve the physical quantity at the nth time

The specific calculation steps are:

其中Δt为时间步长；3.2.4.1) Establish the backward Euler equation

where Δt is the time step;

3.2.4.2) Use the central difference method to calculate the Laplace operator:

为

网格四周的物理量，Δx为网格宽度，Δy为网格高度，本专利使用正方形网格，固有Δx＝Δy；in

for

Physical quantities around the grid, Δx is the width of the grid, Δy is the height of the grid, this patent uses a square grid, inherent Δx=Δy;

将常数项提出得到：3.2.4.3) Order

Taking the constant term out, we get:

3.2.4.4) Convert it to matrix form, and use the conjugate gradient method to solve the large-scale positive definite sparse matrix.

进而由增广数据集

进行预训练。(3.2.5) According to the update and calculation of the body force term, the convection term and the diffusion term, the physical field data of the dust at the subsequent time nodes T ₁ , T ₂ , ..., T _n , T _n+1 are obtained

then augmented by the dataset

Do pre-training.

作为数据集输入神经网络模型U进行预训练：(3.3) Pre-training: use the dust physical information of time nodes T ₁ , T ₂ , ..., T _n , T _n+1 calculated by traditional physical methods in Section (3.2)

Input the neural network model U as a dataset for pre-training:

(3.3.1) Model selection: the present invention introduces the concept of grid and field when studying the dust data. The data set D is essentially a set of spatial distribution data that expresses the variation of various physical variable fields with time, so it can be An analogy is the pixel distribution data of an "image". Therefore, compared with other networks, the U-net model, which has excellent performance in the field of image segmentation prediction, is more suitable for the dust physical field data. Therefore, the present invention selects the U-net model as the pre-training model.

(3.3.2) Training details: The input channel size is 32×32×6×2, of which 32×32 is the grid size, and 6×2 represents 6 physical field data (transverse component of velocity field, longitudinal component of velocity field, concentration field, temperature field, pressure field, humidity field) and two adjacent time fields. The output channel is 32×32×3×1, where 32×32 is the grid size, and 3×1 represents the predicted generated 3 physical fields (transverse component of velocity field, longitudinal component of velocity field and concentration field) and 1 time field .

Further, 4) transfer learning and fine-tuning:

Input the real data set into the pre-trained model U, and adjust the number of iterations (epoch), the training batch (batch-size) and the loss function (loss) to obtain the final predicted trajectory of dust particle drift.

(4.1) Model migration: retain all the parameters in the pre-trained U-Net model and use it as the training model for the real data set;

(4.2) Data training: the input channel size is 32×32×4×2, of which 32×32 is the grid size, and 4×2 represents 4 real physical field data (concentration field, temperature field, pressure field, humidity field) and two adjacent time fields. The output channel is 32×32×3×1, where 32×32 is the grid size, and 3×1 represents the predicted generated 3 physical fields (transverse component of velocity field, longitudinal component of velocity field and concentration field) and 1 time field ;

(4.3) Model fine-tuning: adjust the value of the parameter epoch to 240, adjust the value of the training batch parameter batchsize to 8, set L2 as the loss function, and train to obtain the final dust drift trajectory. Among them, the L2 loss function is the Least Square Error function. The purpose of using LSE is to minimize the square sum of the difference between the real value _yi and the training result f(x _i ) D _L2 , and its function formula is as follows :

The beneficial effects of the present invention are as follows: the model of the present invention is superior to the traditional method in terms of accuracy, calculation time, and stability compared with the traditional physical method, and has better generality than the numerical estimation method only using the neural network. Therefore, the model can be more widely applicable to most scenarios.

Description of drawings

Fig. 1 is the overall structure diagram of the present invention;

Fig. 2 is the Gaussian coordinate schematic diagram of the present invention;

Fig. 3 is a grid schematic diagram of the present invention; in Fig. 3, the white color level of each small square represents the size of the concentration field, and the higher the white color level is, the higher the dust concentration in this grid is represented;

Fig. 4 is the semi-Lagrangian method schematic diagram of the present invention;

Fig. 5 is the dust velocity field calculated by the traditional method of the present invention;

Fig. 6 is the dust concentration field calculated by the traditional method of the present invention; Fig. 6 is composed of 32×32 small squares, and the white color level of each small square represents the size of the concentration field. The higher the dust concentration in this grid;

FIG. 7 is a schematic diagram of the U-net network structure used in the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings.

1-7, a dust particle drift trajectory model and its modeling method, the specific steps are as follows:

1) Data preprocessing:

(1.1) De-duplication: query with the table to screen out the repeated dust data uploaded by the equipment;

(1.2) Wind speed processing:

(1.2.1) Calculate the angle θ between it and the coordinate axis of the geodetic plane according to the wind direction;

(1.2.2) Calculate the wind speed V Calculate the transverse component u and the longitudinal component v:

The transverse component u=Vcos(θ); the longitudinal component v=Vsin(θ).

(1.3) Gaussian projection is carried out on the site coordinates: the goal of this step is to convert (L, B) under the latitude and longitude coordinates into (x, y) under the Gaussian coordinates, and the specific steps can refer to accompanying drawing 2;

(1.3.1) Calculate the central meridian longitude, band number and longitude difference (the distance from the point to the central meridian) according to the latitude and longitude. The 3° central meridian is selected for this model, and the calculation process is as follows:

(1) Calculate the 3° band n ₃ : n ₃ =int(L/3+0.5);

(2) Calculate the central meridian L ₀ with n ₃ of 3°: L ₀ =3n ₃ ;

(3) Calculate the longitude difference l (the distance between the station and the central meridian): l=LL ₀ ;

(1.3.2) Calculate the meridian arc length X according to the selection of the ellipsoid. This model selects the Krasovsky ellipsoid for calculation (which can be applied to the 54 Beijing coordinate system):

X=111134.861B-16036.480 sin 2B+16.828 sin 4B-0.022 sin 6B;

(1.3.3) Calculate the value of each parameter in the positive calculation formula of Gaussian coordinates, where B is the geodetic latitude, t is the tangent value of the geodetic dimension, e is the eccentricity of the ellipse, N is the radius of curvature of the Maoxi circle, and a is the length of the ellipsoid Radius parameter:

(1.3.4) Substitute the parameters obtained by the above calculation into the Gaussian coordinate positive formula to find the coordinates (x, y) in the Gaussian coordinate system:

x=X+N sin B cos Bl ² +N sin B cos ³ B(5-t ² -9e ² cos ² B)l ⁴

y=N cos Bl+N cos ³ B(1-t ² +e ² cos ² B)l ³ +N cos ⁵ B(5-18t ² +t ⁴ )l ⁵

2) Grid processing:

(2.1) Cluster analysis: The cluster C is obtained through the DBscan clustering algorithm, and the steps are as follows:

(1) Mark all objects as unvisited; randomly select an unvisited object P;

(2) If there are at least MinPts objects in the ε field of P, create a new cluster C, and add P to the cluster C;

(3) Let N be the set of objects in the ε field of P. If there are at least MinPts objects in the ε field of P, add these objects to N; if P is not yet a member of any cluster, add P to cluster C.

(2.2) Linear interpolation to obtain a uniform grid (see Figure 3 for a schematic diagram of the grid):

(2.2.1) According to the Delaunay triangulation algorithm, first find out the composition of the three points around the interpolation point

Triangles list triangles (pseudocode below):

(2.2.2) Call the griddata function to perform data interpolation to obtain the dust velocity field, concentration field, temperature field, pressure field, and humidity field represented by a uniform grid.

3) Model pre-training:

(3.1) Initial simulation data: Take the dust data at 23:00 on the evening of July 20, 2021 as the initial simulation data at time T ₀ .

(3.2) Forward calculation by traditional method of physics: The algorithm that the present invention mainly relies on when performing the forward calculation by traditional method is Navier-Stokes equations, or N-S equation for short.

(3.2.1) Body force term: update the body force term f=ρg (g is the acceleration of gravity, ρ is the density).

网格点上粒子对应的速度场横向分量为u，纵向分量为v,求解1个时间步长Δt后网格点物理量场值

的步骤为：(3.2.2) Convective term (refer to Figure 4 for the convection process): Given a grid point position (x, y), the physical quantity field value of this point is

The lateral component of the velocity field corresponding to the particle on the grid point is u, and the vertical component is v. After solving one time step Δt, the physical quantity field value of the grid point is

The steps are:

(1) Track backward for 1 time step to obtain the grid position (x-uΔt, y-vΔt) of the particle;

(2) Constructing the Catmull-Rom spline interpolation function according to the physical field grid to calculate the mean value of the variation of the particle physical quantity field

的计算公式为：

(3) Final

The calculation formula is:

已知，现需要求解第n时刻的物理量

其具体计算步骤为：(3.2.3) Diffusion term: The main purpose of this project is to solve the viscosity term a _v = μ▽ ² u of the fluid, where a _v is the acceleration of the particle under the action of the fluid viscous force, and μ is the viscosity coefficient of the fluid. Assuming the n+1th moment, the physical quantity at the grid coordinates (x, y)

Known, now we need to solve the physical quantity at the nth time

The specific calculation steps are:

其中Δt为时间步长；(1) Establish the backward Euler equation

where Δt is the time step;

(2) The central difference method is used to calculate the Laplace operator:

为

网格四周的物理量，Δx为网格宽度，Δy为网格高度，本专利使用正方形网格，固有Δx＝Δy；in

for

Physical quantities around the grid, Δx is the width of the grid, Δy is the height of the grid, this patent uses a square grid, inherent Δx=Δy;

将常数项提出得到：(3) Order

Taking the constant term out, we get:

(4) Convert it into matrix form, and use the conjugate gradient method to solve the large-scale positive definite sparse matrix.

进而由增广数据集

训练得到预训练模型U，增广数据集中的速度场和浓度场可以参照附图5和附图6。(3.2.4) According to the body force term, the convection term, the diffusion term, and the traditional method, the physical field data of the dust at the subsequent time nodes T ₁ , T ₂ , T ₃ , ..., T _n , T _n+1 are obtained

then augmented by the dataset

The pre-training model U is obtained by training, and the velocity field and the concentration field in the augmented data set can be referred to Fig. 5 and Fig. 6 .

(3.3) Pre-training process:

(3.3.1) Network selection:

Based on the characteristics of dust drift movement (spatially, dust data of adjacent sites are strongly correlated), the present invention introduces the concept of grid and field when studying dust data. Data set D is essentially a set of expressions that express various The spatial distribution data of the physical variable field changes with time, so it can be analogized to the pixel distribution data of the "image". Therefore, the U-net model with excellent performance in the field of image segmentation prediction is more suitable for the dust physical field data in the overall structure than other networks. Therefore, the present invention selects the U-net model as the pre-training model (the specific U-net model). The structure diagram can refer to Figure 7).

(3.3.2) Training details:

(1) The U-Net model is selected for pre-training, the value of the parameter epoch is set to 240, the value of the parameter batchsize is set to 8, the size of the data set is 2000 groups, and the ratio of training set, test set and validation set is 12: 3:1, the regression indicator selects the L2 loss function.

(2) The input channel size is 32×32×6×2: 32×32 is the grid size, 6×2 represents 6 physical field data (transverse component of velocity field, longitudinal component of velocity field, concentration field, temperature field, pressure field, humidity field) and two adjacent time fields. The output channel is 32×32×3×1: 32×32 is the grid size, and 3×1 represents the predicted generated 3 physical fields (transverse component of velocity field, longitudinal component of velocity field and concentration field) and 1 time field .

4) Model migration and fine-tuning:

(4.1) Model migration: retain all the parameters in the pre-trained U-Net model and use it as the training model for the real data set;

(4.2) Data training: the input channel size is 32×32×4×2, of which 32×32 is the grid size, and 4×2 represents 4 real physical field data (concentration field, temperature field, pressure field, humidity field) and two adjacent time fields. The output channel is 32×32×3×1, where 32×32 is the grid size, and 3×1 represents the predicted generated 3 physical fields (transverse component of velocity field, longitudinal component of velocity field and concentration field) and 1 time field ;

(4.3) Model fine-tuning: adjust the value of the parameter epoch to 240, adjust the value of the training batch parameter batchsize to 8, set L2 as the loss function, and train to obtain the final dust drift trajectory. Among them, the L2 loss function is the Least Square Error function. The purpose of using LSE is to minimize the square sum of the difference between the real value _yi and the training result f(x _i ) D _L2 , and its function formula is as follows :

Aiming at the problems that the traditional method has too large amount of calculation, poor accuracy and stability, and the neural network method lacks interpretability, the present invention combines the codec neural network and the optimized traditional method, and proposes a dust particle drift. Trajectory model and its modeling method.

The content described in the embodiments of the present specification is only an enumeration of the realization forms of the inventive concept, and the protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments, and the protection scope of the present invention also extends to those skilled in the art. Equivalent technical means that can be conceived by a person based on the inventive concept.

Claims (5)

1. a dust particle drift trajectory model and its modeling method, is characterized in that, comprises the steps: 1) Data preprocessing stage: preprocessing data; 2) Model pre-training stage: divide the grid according to the geographical location of the site equipment; 3) Model pre-training stage: model pre-training through simulated data and traditional physical methods; 4) Transfer learning fine-tuning stage: use the trained model to predict the real drift trajectory of dust particles. 2. A dust particle drift trajectory model and its modeling method as claimed in claim 1, wherein the data preprocessing stage comprises the following steps: 1.1) Data duplication check stage: through table query, filter out the dust data repeatedly uploaded by the site equipment; 1.2) Wind speed processing stage: Calculate the horizontal and vertical components of wind speed according to the wind speed and wind direction parameters in the dust data. The specific execution steps are as follows: 1.2.1) Calculate the angle θ between it and the coordinate axis of the geodetic plane according to the wind direction; 1.2.2) Calculate the wind speed V: u=V cos(θ), v=V sin(θ); where u represents the horizontal component and v represents the vertical component; 1.3) Geographical coordinate Gaussian projection stage: The latitude and longitude coordinates (L, B) of the site are converted into Gaussian coordinates (x, y) through Gaussian projection calculation, where L is the geodetic longitude and B is the geodetic latitude; the specific execution steps are as follows: 1.3.1) Calculate the central meridian longitude, belt number and longitude difference according to the latitude and longitude. The model selects the 3° belt central meridian. The steps are as follows: 1.3.1.1) Calculate the 3° band number n ₃ : n ₃ =int(L/3+0.5); 1.3.1.2) Calculate the central meridian L ₀ with the number n ₃ of 3°: L ₀ =3n ₃ ; 1.3.1.3) Calculate the longitude difference l (the distance between the station and the central meridian): l=LL ₀ ; 1.3.2) Calculate the meridian arc length X according to the selection of the ellipsoid; the model selects the Krasovsky ellipsoid for calculation: X=111134.861B-16036.480sin2B+16.828sin4B-0.022sin6B; 1.3.3) Calculate the value of each parameter in the positive calculation formula of Gaussian coordinates, wherein, B is the geodetic latitude, t is the tangent of the geodetic dimension, e is the eccentricity of the ellipse, N is the radius of curvature of the Maoxi circle, and a is the long radius of the ellipsoid parameter:

1.3.4) Use the Gaussian coordinate formula to solve the coordinates (x, y) in the Gaussian coordinate system: x=X+N sinB cosBl ² +N sinB cos ³ B(5-t ² -9e ² cos ² B)l ⁴ y=N cosBl+N cos ³ B(1-t ² +e ² cos ² B)l ³ +Ncos ⁵ B(5-18t ² +t ⁴ )l ⁵ . 3. A dust particle drift trajectory model and its modeling method as claimed in claim 1, wherein the grid processing stage comprises the following steps: 2.1) Cluster analysis stage: Before dividing the grid, perform DBScan cluster analysis according to the location of the site equipment, and obtain the site equipment cluster cluster under Gaussian coordinates; 2.2) Grid division stage: After obtaining a cluster with better effect in step 2.1), divide a uniform square grid according to the geographical location of the site equipment in the cluster; 2.3) Data interpolation stage: perform linear interpolation according to the distance between equipment sites and dust data values, and finally save the dust data in the grid in the form of velocity field, concentration field, temperature field, pressure field, and humidity field. 4. A kind of dust particle drift trajectory model and modeling method thereof as claimed in claim 1, is characterized in that, described network training stage comprises the following steps: 3.1) Simulation data stage: simulate and generate dust simulation data at time T0 in the divided grid, including the dust velocity field, concentration field, temperature field, pressure field and humidity field; 3.2) The forward calculation stage of the traditional physical method: On the basis of the simulated data, according to the traditional physical method, that is, according to the NS Navier-Stokes equation, the calculation of the body force term f, the convection term and the diffusion term is updated, and the forward calculation is carried out. To predict, get the dust physics data set of subsequent time nodes T1, T2, T3, ..., Tn, Tn+1

The specific steps are as follows: 3.2.1) Calculation of the body force term f: f=ρg, where g is the acceleration of gravity, and ρ is the fluid density; 3.2.2) Convective term calculation: Given a grid point position (x, y), the physical quantity field value of this point is

The lateral component of the velocity field corresponding to the particle on the grid point is u, and the vertical component is v. After solving one time step Δt, the physical quantity field value of the grid point is

The relevant steps are: 3.2.2.1) Track backward for 1 time step to get the grid position of the particle (x-uΔt, y-vΔt); 3.2.2.2) The Catmull-Rom spline interpolation function is constructed according to the physical field grid to calculate the mean value of the change of the particle physical quantity field

3.2.2.3) Final

The calculation formula is:

3.2.3) Calculation of diffusion term: The main purpose of this step is to solve the viscosity term a _v = μ▽ ² u of the fluid, where a _v is the acceleration of the particle under the action of the fluid viscous force, and μ is the viscosity coefficient of the fluid; suppose the nth +1 time, physical quantity at grid coordinates (x, y)

Known, now we need to solve the physical quantity at the nth time

The specific calculation steps are: 3.2.3.1) Establish the backward Euler equation

where Δt is the time step; 3.2.3.2) Use the central difference method to calculate the Laplace operator:

in

for

Physical quantities around the grid, Δx is the grid width, Δy is the grid height; 3.2.3.3) Order

Taking the constant term out, we get:

3.2.3.4) Convert it to matrix form, and use the conjugate gradient method to solve the large-scale positive definite sparse matrix:

3.3) Data set generation stage: According to the update and calculation of the body force term, the convection term, and the diffusion term, the dust physical field data at the subsequent time nodes T ₁ , T ₂ , ..., T _n , T _n+1 are obtained.

Then by the dataset

pre-training; 3.4) Pre-training stage: use the dust physical information of time nodes T ₁ , T ₂ , T ₃ , . . . , T _n , and T _n+1 calculated by traditional physical methods in step C3).

Pre-training the neural network model U as a dataset: 3.4.1) Model selection: select U-Net as the training network; 3.4.2) Training details: The input channel size is 32×32×6×2, of which 32×32 is the grid size, 6×2 represents 6 physical field data, 2 adjacent time fields, and 6 physical field data They are the transverse component of the velocity field, the longitudinal component of the velocity field, the concentration field, the temperature field, the pressure field, and the humidity field; the output channel is 32×32×3×1, of which 32×32 is the grid size, and 3×1 represents the prediction generation There are 3 physical fields and 1 time field, and the 3 physical fields are the transverse component of the velocity field, the longitudinal component of the velocity field and the concentration field. 5. A dust particle drift trajectory model and its modeling method as claimed in claim 4, wherein the migration training and fine-tuning stage comprises the following steps: 4.1) Model migration stage: retain all the parameters in the U-Net model pre-trained in step 3.4), and use it as the training model of the real data set; 4.2) Data training phase: the input channel size is 32×32×4×2, of which 32×32 is the grid size, 4×2 represents 4 real physical field data and 2 adjacent time fields, 4 real physical fields The field data are concentration field, temperature field, pressure field, and humidity field; the output channel is 32×32×3×1, of which 32×32 is the grid size, 3×1 represents the 3 physical fields and 1 generated by prediction Time field, the three physical fields are the transverse component of the velocity field, the longitudinal component of the velocity field and the concentration field; 4.3) Model fine-tuning stage: adjust the value of the parameter epoch to 240, set the value of the training batch parameter batchsize to 8, set L2 as the loss function, and finally get the prediction result. Among them, the loss function L2 adopts the minimized square error (LSE) function. The purpose of using LSE is to minimize the square sum of the difference between the real value y _i and the training result f( _xi ) D _L2 , and its function formula is

Images