A low-altitude meteorological digital twin reconstruction method based on three-dimensional entity data modeling

Abstract

The application relates to the technical field of low-altitude atmospheric monitoring, and particularly discloses a low-altitude meteorological digital twin reconstruction method based on three-dimensional entity data modeling, which realizes seamless connection from a regional macro wind field to a city micro flow field through the combination of a mesoscale meteorological model and a computational fluid dynamics model, directly considers the physical action of a ground surface feature on airflow based on real city three-dimensional data modeling, continuously corrects the error of a CFD grid model through real-time observation data, makes up for the uncertainty of pure numerical simulation, improves the capturing capability of spatiotemporal mutation phenomena, and further supplements the prediction of complex turbulent flow phenomena that are difficult to analyze by using a computational fluid dynamics model through an intelligent prediction model to mine hidden rules in data, so that the precision of city boundary layer airflow simulation can be significantly improved, and more accurate meteorological guarantee services can be provided for the fields of atmospheric environment monitoring, city planning, aviation safety and the like.

Description

Technical Field

[0001] This invention belongs to the field of low-altitude atmospheric monitoring technology, specifically relating to a method for reconstructing low-altitude meteorological digital twins based on three-dimensional entity data modeling. Background Technology

[0002] Typical scenarios such as unmanned urban delivery, urban governance, science and technology tourism, and border control mainly rely on low-altitude aerial activities (including drones, small aircraft, etc.). Since low-altitude operations are highly dependent on weather conditions, accurate and timely meteorological information services through low-altitude meteorological digital twin reconstruction are essential to ensure flight safety, improve delivery efficiency, and meet regulatory requirements.

[0003] Currently, low-altitude meteorological digital twin reconstruction involves constructing a virtual mirror of the low-altitude atmospheric environment, integrating multi-source meteorological data (such as radar, satellites, and ground observation stations), and combining IoT, AI, and big data technologies to achieve real-time monitoring, prediction, and simulation of low-altitude meteorological conditions. Its core technologies include data acquisition and fusion, 3D modeling and simulation, real-time updating and feedback mechanisms, and intelligent analysis and prediction based on historical and real-time data.

[0004] However, during data collection and integration, the airflow in the urban boundary layer is affected by complex surface features such as buildings, streets, canyons, and green spaces, resulting in micro-meteorological phenomena such as turbulence and eddies. Furthermore, the spatiotemporal resolution limitations of radar, satellites, and ground meteorological stations make it difficult to capture the spatial heterogeneity and spatiotemporal abrupt changes in the urban boundary layer from meteorological data. Spatial heterogeneity includes calm wind zones formed on the leeward side of buildings and accelerated airflows at the top of tall buildings, while spatiotemporal abrupt changes include sudden gusts and thermal convection bubbles. These factors affect the accuracy of simulating and predicting the complex and variable airflow conditions within the urban boundary layer, hindering the provision of precise atmospheric environmental protection services. Therefore, we need to propose a low-altitude meteorological digital twin reconstruction method based on three-dimensional entity data modeling to address the aforementioned problems, enabling accurate simulation and prediction of the complex and variable airflow conditions within the urban boundary layer, thereby providing precise atmospheric environmental protection services. Summary of the Invention

[0005] The purpose of this invention is to provide a low-altitude meteorological digital twin reconstruction method based on three-dimensional entity data modeling, which can accurately simulate and predict the complex and ever-changing airflow conditions within the urban boundary layer, thereby providing precise atmospheric environmental protection services to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for reconstructing low-altitude meteorological digital twins based on 3D entity data modeling includes the following steps:

[0008] S1. Obtain initial wind field data for the urban area and its surrounding areas using a mesoscale meteorological model;

[0009] S2. Using the wind field output by the mesoscale meteorological model as the background, a computational fluid dynamics model is introduced to refine the spatial grid of the urban area, generating a refined background flow field for the city.

[0010] S3. Collect three-dimensional entity data of the city through a sensor array and construct a three-dimensional geographic information model to reflect the surface roughness characteristics;

[0011] S4. Use the three-dimensional geographic information model as the boundary condition of the computational fluid dynamics model to simulate airflow and construct a CFD mesh model;

[0012] S5. Real-time acquisition of high-frequency wind field data in the 0-1000 meter boundary layer through low-altitude monitoring network to obtain multi-source real-time data;

[0013] S6. Integrate multi-source real-time data with the refined urban background flow field to correct the error of the CFD grid model;

[0014] S7. Based on real-time observation data, the corrected CFD mesh model is updated frequently to form a low-altitude three-dimensional twin model that includes spatial heterogeneity and spatiotemporal abrupt changes.

[0015] S8. Establish an intelligent prediction model for the characteristics of urban boundary layer airflow and learn the laws of micro-meteorological evolution.

[0016] S9. Combines real-time updated low-altitude three-dimensional twin model and intelligent prediction model to predict future airflow changes, and continuously optimizes the parameters of intelligent prediction model by comparing with new observation data.

[0017] Preferably, the mesoscale meteorological model used is the WRF mesoscale meteorological model, and the process for obtaining the initial wind field data is as follows:

[0018] A1. Download the global forecast system to obtain global forecast data, and process the topographic and underlying surface data through the WPS module in the mesoscale meteorological model;

[0019] A2. Input global forecast data into the mesoscale meteorological model, set the simulation area to cover the city and its surrounding area within a range of 100-200km, and run a rapid cyclic assimilation for 3 hours;

[0020] A3. By integrating observation data from ground automatic weather stations and radiosonde stations, a three-dimensional initial wind field containing wind speed, wind direction, and air pressure is generated.

[0021] Preferably, the process of refining the spatial grid of an urban area using a computational fluid dynamics model is as follows:

[0022] B1. The calculation scope is the area extending outward from the city center at a height of 5-10 times the building height.

[0023] B2. Set boundary conditions according to the calculation range. Boundary conditions include entrance boundary, surface boundary, exit boundary and top boundary.

[0024] B3. Use computational fluid dynamics models to refine the grid and perform fluid dynamics calculations on urban areas to generate a refined background flow field for the city.

[0025] The preferred process for constructing a 3D geographic information model is as follows:

[0026] C1. Filter the data collected by the sensor group to remove noise points in the data, and classify the processed data into point clouds, which are then classified into ground points, building points, and vegetation points.

[0027] C2. Eliminate the spatial deviation between the lidar and UAV aerial images by matching the coordinate system of the classified point cloud with feature points;

[0028] C3. Based on the point cloud after spatial deviation elimination, a continuous triangular mesh model is generated using a triangulation algorithm to form the surface contours of buildings and terrain.

[0029] C4. Optimize the parameters of high-rise buildings and complex terrain using modeling software to ensure that the error between the building and terrain data in the triangular mesh model and the actual measurement data is less than 5%, thereby obtaining a three-dimensional geographic information model.

[0030] Preferably, the process for constructing a CFD mesh model is as follows:

[0031] G1. Convert the parameters of surface roughness and building morphology in the 3D geographic information model into wall function boundary conditions of the computational fluid dynamics model.

[0032] G2. Discrete equations based on computational fluid dynamics models are solved using the finite volume method.

[0033] G3. Select a suitable turbulence model and couple it with the discrete equations to handle complex airflows;

[0034] G4 updates the parameters of the turbulence model and the wall function boundary conditions of the computational fluid dynamics model in real time, iteratively solves the coupled discrete equations until the flow field parameters converge, and forms a CFD mesh model.

[0035] Preferably, during the acquisition of multi-source real-time data, the sensor group is combined with the low-altitude monitoring network to obtain real-time data of the boundary layer high-frequency wind field. Then, the real-time data is processed by outlier removal and moving average filtering. The processed data is transmitted to the cloud server in real time through the 5G network to build a spatiotemporal database.

[0036] Preferably, the error correction process for the CFD mesh model is as follows:

[0037] H1. Real-time data is interpolated onto the grid nodes of the CFD grid model using the inverse distance weighting algorithm;

[0038] H2. Calculate the deviation between the measured value and the simulated value of the CFD mesh model, and use the Kalman filter algorithm to feed the error back to the boundary conditions of the CFD mesh model;

[0039] H3. Evaluate the confidence level of the corrected CFD mesh model. If the confidence level is within the allowable range, proceed to S7. If the confidence level exceeds the allowable range, re-execute H1-H2.

[0040] Preferably, the process of updating the CFD grid model includes: incorporating the latest real-time observation data into the CFD grid model at fixed time steps, capturing airflow differences of different surface types through the Kriging interpolation algorithm, identifying sudden changes in the flow field caused by sudden weather by using the moving window standard deviation, and finally coupling the updated CFD grid model with the three-dimensional geographic information model to generate a low-altitude three-dimensional twin model.

[0041] The preferred process for establishing an intelligent prediction model is as follows:

[0042] K1. Extract flow field features and meteorological elements from the low-altitude three-dimensional twin model. The flow field features include wind speed gradient and vortex intensity, and the meteorological elements include temperature and humidity.

[0043] K2 uses long short-term memory neural networks to build models using spatiotemporal sequence data as input.

[0044] K3 utilizes historical twin data and observational data for supervised learning, and optimizes hyperparameters through cross-validation. The optimized model is the intelligent prediction model.

[0045] Preferably, when predicting future airflow changes, a real-time low-altitude three-dimensional twin model is used as the initial condition. The flow field changes in the next 1-6 hours are extrapolated through the intelligent prediction model. New measured data are acquired in real time, and the deviation between the predicted value and the measured data is compared. Based on the deviation, the parameters of the intelligent prediction model are adjusted in reverse to form a closed loop of prediction, verification and optimization.

[0046] The low-altitude meteorological digital twin reconstruction method based on three-dimensional entity data modeling proposed in this invention has the following advantages compared with existing technologies:

[0047] This invention achieves a seamless connection from regional macroscopic wind fields to urban microscopic flow fields by combining mesoscale meteorological models and computational fluid dynamics (CFD) models. Based on real urban 3D data modeling, it directly considers the physical effects of surface features on airflow. It continuously corrects the errors of the CFD grid model through real-time observation data, making up for the uncertainties of pure numerical simulation and improving the ability to capture spatiotemporal abrupt phenomena. Furthermore, it uses intelligent prediction models to uncover hidden patterns in the data and supplements the prediction of complex turbulent phenomena that are difficult to analyze by CFD models. This method can significantly improve the accuracy of urban boundary layer airflow simulation and provide more accurate meteorological support services for atmospheric environmental monitoring, urban planning, aviation safety and other fields. Attached Figure Description

[0048] Figure 1 A flowchart of a low-altitude meteorological digital twin reconstruction method according to an embodiment of the present invention is shown;

[0049] Figure 2 A flowchart illustrating the process of spatial mesh refinement for an urban area using a computational fluid dynamics model according to an embodiment of the present invention is shown.

[0050] Figure 3 A flowchart illustrating the construction of a three-dimensional geographic information model according to an embodiment of the present invention is shown;

[0051] Figure 4 A flowchart illustrating the error correction process for a CFD mesh model according to an embodiment of the present invention is shown. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] This invention provides, for example Figure 1-4 The method for reconstructing a low-altitude meteorological digital twin based on 3D entity data modeling, as shown, includes the following steps:

[0054] S1. Obtain initial wind field data for the urban area and its surrounding areas using a mesoscale meteorological model. The initial wind field data includes wind speed, wind direction, and air pressure.

[0055] The mesoscale meteorological model used is the WRF mesoscale meteorological model, which includes a WPS module and a WRF-ARW module. The WPS module is responsible for data preprocessing, such as geographic interpolation and meteorological field extraction and interpolation; the WRF-ARW module is responsible for data integration calculation, such as dynamic framework and physical process parameterization.

[0056] The process for acquiring initial wind field data is as follows:

[0057] A1. Download the Global Forecast System (GFS) to obtain global forecast data. Process the topographic and underlying surface data using the WPS module in the mesoscale meteorological model. The topographic data is ASTER GDEM 30m data, and the underlying surface data is MODIS 21 land use data. The ASTER GDEM 30m data is a global digital elevation model generated based on optical stereo image pairs from the ASTER sensor on the Terra satellite. 1.5 million scenes of ASTER archived data are processed using stereo correlation technology to generate elevation information for 1.26 million independent scenes. After cloud removal and outlier correction, a global dataset is generated by dividing the data into 1×1 degree sections. The MODIS 21 land use data is based on MODIS sensor observation data from the Terra and Aqua satellites. A global land cover classification map is generated using a supervised classification algorithm. The data adopts the 21-class classification system of the International Geosphere-Biosphere Programme (IGBP). Combining MODIS's 36 spectral bands (including visible light, near-infrared, thermal infrared, etc.) and annual synthesis algorithms, a global land cover product with a spatial resolution of 500 meters and a temporal resolution of 1 year is generated.

[0058] A2. Input global forecast data into the mesoscale meteorological model, set the simulation area to cover the city and its surrounding area within a range of 100-200km, set the vertical top of the layer to 500hPa and the horizontal resolution to 1-3km, and run for 3 hours for rapid cyclic assimilation.

[0059] Rapid cyclic assimilation is a process in numerical weather prediction that continuously integrates new observational data into a mesoscale meteorological model every 10-30 minutes, constantly updating the initial field. 3-hour cyclic assimilation is a closed-loop process that completes data assimilation, model operation and result output every 3 hours. The high frequency of updates improves the forecast's ability to respond to sudden weather events.

[0060] A3. Integrate observation data from ground automatic weather stations and radiosonde stations to generate a three-dimensional initial wind field including wind speed, wind direction, and air pressure;

[0061] By obtaining regional-scale wind field background through mesoscale meteorological models, it is possible to capture large-scale weather systems around cities, providing macroscopic drivers for subsequent refined simulations.

[0062] S2. Using the wind field output by the mesoscale meteorological model as the background, a computational fluid dynamics model is introduced to refine the spatial grid of the urban area, generating a refined background flow field for the city.

[0063] The process of refining the spatial grid of an urban area using a computational fluid dynamics model is as follows:

[0064] B1. The calculation scope is the area extending outward from the city center at a height of 5-10 times the building height.

[0065] B2. Set boundary conditions according to the calculation range. Boundary conditions include inlet boundary, surface boundary, outlet boundary, and top boundary. The inlet boundary is the wind field data output by the mesoscale meteorological model, such as wind speed profile and turbulence intensity. The surface boundary is set as a no-slip wall considering the influence of roughness. The outlet boundary is set as a pressure outlet with a pressure value of atmospheric pressure. The top boundary adopts a free-slip boundary.

[0066] B3. Use computational fluid dynamics models to refine the grid and perform fluid dynamics calculations on urban areas to generate a refined background flow field for the city.

[0067] During mesh refinement, tetrahedral meshes are used to capture boundary layer flows around buildings and in near-ground areas. For tetrahedral elements within a specified area, the midpoint of the longest edge is taken as a new node, and the original element is divided in two. The mesh is gradually refined to the target density. Inconsistencies between the refined mesh and adjacent meshes are addressed, and multiple iterations are used to ensure a smooth mesh transition and avoid distortion. Local meshes are refined to 10-100m. Hexahedral meshes are used to capture boundary layer flows in the far field. While maintaining the ability to capture flow field characteristics, a gradual transition strategy is adopted to avoid calculation errors caused by abrupt changes in mesh size.

[0068] Computational fluid dynamics models perform fluid dynamics calculations using discrete equations. The formulas for these discrete equations are as follows:

[0069] ,

[0070] ,in, For cell face index, For unit volume, air density, For noodles Center velocity, For noodles Area vector The velocity at the previous time step. For time step, For noodles Central pressure, It is the acceleration due to gravity. For cell center index, Aerodynamic viscosity, For control surface The velocity gradient tensor;

[0071] Computational fluid dynamics (CFD) models are high-resolution numerical simulation tools that can describe complex fluid flow conditions in detail. In urban wind field analysis, CFD models can accurately capture the local effects of urban buildings, street layouts, and terrain undulations on the wind field by simulating on a small-scale grid. This high-resolution data can make up for the lack of detail in mesoscale models and provide more refined background flow field information.

[0072] S3. Collect three-dimensional entity data of the city through a sensor array and construct a three-dimensional geographic information model to reflect the surface roughness characteristics;

[0073] The sensor group includes a lidar and a drone equipped with a camera. The lidar acquires three-dimensional point clouds of the ground surface through laser pulse scanning with centimeter-level accuracy, and directly generates geometric data of building outlines and terrain height. The drone equipped with a camera can acquire images from multiple angles. Through multi-angle image matching, it can reconstruct details of building facades and roof shapes, supplementing the lidar's blind spots in the vertical plane.

[0074] The process of constructing a 3D geographic information model is as follows:

[0075] C1. Filter the data collected by the sensor group to remove noise points in the data, and classify the processed data into point clouds, which are then classified into ground points, building points, and vegetation points.

[0076] C2. The coordinate system of the classified point cloud is used to eliminate the spatial deviation between the lidar and the UAV aerial imagery by feature point matching; feature points can be selected from building corners, road intersections, etc.

[0077] C3. Based on the point cloud after spatial deviation elimination, a continuous triangular mesh model is generated using a triangulation algorithm to form the surface contours of buildings and terrain.

[0078] C4. Optimize the parameters of high-rise buildings and complex terrains using modeling software to make the error between the building and terrain data in the triangular mesh model and the actual measurement data less than 5%, thereby obtaining a three-dimensional geographic information model.

[0079] S4. Use the three-dimensional geographic information model as the boundary condition of the computational fluid dynamics model to simulate airflow and construct a CFD mesh model;

[0080] The process of constructing a CFD mesh model is as follows:

[0081] G1. Convert the parameters of surface roughness and building morphology in the 3D geographic information model into wall function boundary conditions of the computational fluid dynamics model.

[0082] The formula for the wall function is:

[0083] ,

[0084] , , ,

[0085] in, For dimensionless velocity, Distance from the wall Flow velocity at that point For friction speed, For wall shear stress, is the von Kármán constant, approximately 0.41. The distance is a dimensionless wall distance. Distance from the wall Aerodynamic viscosity, air density, This is an empirical constant, approximately 5.0;

[0086] G2. Discrete equations based on computational fluid dynamics models are solved using the finite volume method.

[0087] During discretization, the computational domain is discretized into polyhedral control volumes, with the center of each control volume serving as a computational node. A second-order upwind scheme is used to suppress numerical diffusion, and Gaussian linear interpolation is employed to ensure flux conservation. The second-order upwind scheme is a numerical discretization method primarily used to solve for the convection term. It calculates the value of the current node based on information from upstream nodes, and improves computational accuracy by considering information from two upstream nodes. Specifically, the second-order upwind scheme not only utilizes the node values ​​immediately upstream of the current node but also considers the value of another upstream node, thus enabling a more accurate capture of the flow field characteristics.

[0088] In Gaussian linear interpolation, a certain number of Gaussian integration points are selected on each boundary of the control volume. The positions and weights of these points are determined according to the Gaussian integration formula to ensure the accuracy of the interpolation.

[0089] Then, based on the position and weight of the Gaussian integration points and the value of the control volume center node, interpolation coefficients are calculated. These coefficients are used to interpolate the value of the control volume center node to the Gaussian integration points on the boundary.

[0090] Finally, using the interpolation coefficients and information at the Gaussian integration points (such as velocity and density), the flux values ​​at the control volume boundary are calculated. These flux values ​​are then used to solve the conservation equations of the control volume.

[0091] G3. Select a suitable turbulence model and couple it with the discrete equations to handle complex airflows; the turbulence model includes... Model, Model and LES model, with Taking the model as an example, the discrete equations after coupling with the discrete equations are as follows:

[0092] ,

[0093] ,

[0094] in, Let k be the derivative of the turbulent kinetic energy k with respect to time t. For turbulent kinetic energy k in relation to spatial coordinate components The derivative, Aerodynamic viscosity, Let j be the j-direction component of the air velocity. The turbulent viscosity coefficient, is the Prandtl number for the turbulent kinetic energy equation, with a value of 1.0. The turbulent kinetic energy generation term, generated by the interaction between the average velocity gradient and turbulent viscosity, reflects the energy generation mechanism of turbulence. This is the turbulent kinetic energy generation term caused by buoyancy, which needs to be considered when there is density stratification / buoyancy effect. air density, The Prandtl number for the dissipation rate equation is 1.3. , and These are the empirical constants of the dissipation rate equation. 44. , Related to the buoyancy effect, Turbulent dissipation rate The derivative with respect to time t, Turbulent dissipation rate For spatial coordinate components The derivative,

[0095] G4. Update the parameters of the turbulence model and the wall function boundary conditions of the computational fluid dynamics model in real time, and iteratively solve the coupled discrete equations until the flow field parameters converge to form a CFD mesh model.

[0096] By refining the boundary conditions of geographic information models, the computational fluid dynamics model can be improved to depict airflow disturbances on urban underlying surfaces (such as high-rise buildings and terrain).

[0097] S5. Real-time acquisition of high-frequency wind field data in the 0-1000 meter boundary layer through low-altitude monitoring network to obtain multi-source real-time data;

[0098] The low-altitude monitoring network mainly consists of environmental monitoring sensors and terrain and space sensors. The environmental monitoring sensors include anemometers, wind direction sensors, temperature and humidity sensors, and barometric pressure sensors. These sensors are installed at meteorological stations to collect wind speed, wind direction, temperature, and humidity data at a frequency of 10-60 seconds per time. The terrain and space sensors are drones equipped with thermal imagers, used for dynamic inspections.

[0099] When collecting multi-source real-time data, the sensor group is combined with the low-altitude monitoring network to collect real-time data of the high-frequency wind field in the boundary layer. Then, the real-time data is processed by outlier removal and moving average filtering. The processed data is transmitted to the cloud server in real time through the 5G network to build a spatiotemporal database.

[0100] S6. Integrate multi-source real-time data with the refined urban background flow field to correct the error of the CFD grid model;

[0101] The error correction process for CFD mesh models is as follows:

[0102] H1. Real-time data is interpolated onto the grid nodes of the CFD mesh model using the inverse distance weighting algorithm. The formula for the inverse distance weighting algorithm is:

[0103] ,in, These are the estimated values ​​for the points to be interpolated. For the first Real-time data, For the interpolation point and the first Euclidean distance between real-time data points This is a power parameter used to control the rate at which the weights decay with Euclidean distance. The total amount of real-time data involved in the calculation; The result is the weight value, the power parameter. The larger the value, the faster the weight decays with increasing distance, and the more significant the impact on neighboring points.

[0104] H2. Calculate the deviation between the measured values ​​and the simulated values ​​of the CFD mesh model, and use the Kalman filter algorithm to feed the error back to the boundary conditions of the CFD mesh model. The Kalman filter algorithm formula is:

[0105] ,

[0106] ,

[0107] in, For the first State estimate at time step For the first Real-time data at any given moment For the first The state transition matrix at each time step, For the first Time-of-flight observation matrix For Kalman gain, For the first Time error covariance matrix, For process noise, For the first State estimate at time 10:00 For the first The error covariance matrix at time t. It is the identity matrix. for The transpose of the matrix;

[0108] H3. Evaluate the confidence level of the corrected CFD mesh model. If the confidence level is within the allowable range, proceed to S7. If the confidence level exceeds the allowable range, re-execute H1-H2.

[0109] S7. Based on real-time observation data, the corrected CFD mesh model is updated frequently to form a low-altitude three-dimensional twin model that includes spatial heterogeneity and spatiotemporal abrupt changes.

[0110] The process of updating the CFD grid model includes: incorporating the latest real-time observation data into the CFD grid model at a fixed time step, capturing the airflow differences of different land surface types through the Kriging interpolation algorithm, identifying the sudden changes in the flow field caused by sudden weather by using the moving window standard deviation, and finally coupling the updated CFD grid model with the three-dimensional geographic information model to generate a low-altitude three-dimensional twin model.

[0111] The process of integrating the latest real-time observation data with the CFD grid model is as follows:

[0112] D1. Based on the coordinates of the real-time data observation points, determine the corresponding grid cell in the CFD grid model. If the observation point is located at the grid boundary, use the weighting method to associate it with the surrounding cells. If there are multiple observation points in each grid cell, select the coordinates corresponding to the average value of the data from multiple observation points as the initial attribute.

[0113] D2. Real-time data is updated every 5-10 minutes. Each time it is updated, the latest observation data is superimposed on the current state of the CFD grid model, and the flow field parameters of the CFD grid model are recalculated.

[0114] D3. Compare the CFD mesh model output before and after the update with the measured data, and calculate the error between the model output value and the measured data. If the error is within the allowable range, it is determined that the real-time observation data has been successfully integrated. Otherwise, the real-time observation data needs to be integrated again until the real-time observation data is successfully integrated.

[0115] The process of using the Kriging interpolation algorithm to capture differences in surface airflow is as follows:

[0116] E1. Group the observation data according to land surface type, and calculate the semivariogram for each group. The formula for calculating the semivariogram is:

[0117] ,

[0118] in, The value is the semivariogram, representing the squared mean of the attribute differences for sample pairs with an interval of h. For spatial lag distance, For data The attribute value, The number of data pairs with a distance of h;

[0119] E2. Use a spherical model or an exponential model to fit the measured semivariogram to obtain the spatial variability of airflow under different surface types;

[0120] E3. Determine the interpolation points based on the spatial variability of land surface types. Calculate the weighting coefficients using the interpolation formula, incorporating known points within the neighborhood. Then, interpolate the points based on these weighting coefficients. The interpolation formula is as follows:

[0121] ,

[0122] in, Given points For a known point Weighting coefficients It is a Lagrange multiplier. The point to be interpolated. and For known points in the neighborhood, The semivariogram value represents the known points. and Spatial variation characteristics between them This represents the number of known points in the neighborhood.

[0123] E4. The predicted airflow parameters at the interpolation points are obtained by weighted summation. The formula for calculating the predicted airflow parameters is as follows:

[0124] ,

[0125] in, interpolation point Predicted airflow parameters Given points For interpolation points The weighting coefficients, where m is the number of known points. Given points Parameter values;

[0126] E5. Based on the physical characteristics of different land surface types, the Kriging interpolation results are coupled with the land surface type correction factor to obtain the airflow parameter distribution under different land surface types. The coupling formula is as follows:

[0127] ,in, interpolation point Predicted airflow parameters These are the corrected airflow parameters. The correction function is based on surface characteristics. The correction function selects one or more of the following methods according to the actual surface characteristics: C-correction method, Minnaert correction model, SCS model, statistical correction method, and foundation bearing capacity characteristic value correction formula. The above correction functions are all common knowledge in the field, so the correction function itself will not be described in detail here.

[0128] When coupling the updated CFD mesh model with the 3D geographic information model, it is recommended to first establish a real-time data exchange protocol between the CFD mesh model and the 3D geographic information model. This will enable the updated flow field data of the CFD mesh model to be automatically synchronized to the 3D geographic information model and trigger a 3D scene refresh. Then, a spatiotemporal database should be used to store the historical update data of the CFD mesh model and the dynamic change data of the 3D geographic information model. A 3D visualization engine should be used to overlay and render the flow field data (such as wind speed vector field) calculated by the CFD mesh model with the terrain and building models of the 3D geographic information model. The coupling effect between the physical field and the geographic environment should be displayed through transparency adjustment and dynamic textures. The 3D visualization engine should use Cesium or Unity.

[0129] S8. Establish an intelligent prediction model for the characteristics of urban boundary layer airflow and learn the laws of micro-meteorological evolution.

[0130] The process of building an intelligent prediction model is as follows:

[0131] K1. Extract flow field features and meteorological elements from the low-altitude three-dimensional twin model. The flow field features include wind speed gradient and vortex intensity, and the meteorological elements include temperature and humidity.

[0132] K2 uses long short-term memory neural networks to build models using spatiotemporal sequence data as input.

[0133] K3. Supervised learning is performed using historical twin data and observational data. Hyperparameters are optimized through cross-validation. The optimized model is the intelligent prediction model.

[0134] For example, taking time series data as an example, the hyperparameter optimization process is as follows: Divide the historical twin data and observation data into L non-overlapping subsets in chronological order, ensuring that each subset contains similar spatiotemporal distribution features. Each time, select L-1 subsets as the training set and 1 subset as the validation set. Train the model on the training set using the current hyperparameter combination, calculate the loss value and evaluation index on the validation set, repeat L times, and take the average loss as the performance index of the hyperparameter combination to train the model cyclically. Model the relationship between hyperparameters and performance through Gaussian process, and select the hyperparameter combination with the minimum average loss in cross-validation for the final model training.

[0135] S9. Combines real-time updated low-altitude three-dimensional twin model and intelligent prediction model to predict future airflow changes, and continuously optimizes the parameters of intelligent prediction model by comparing with new observation data.

[0136] When predicting future airflow changes, a real-time low-altitude three-dimensional twin model is used as the initial condition. The flow field changes in the next 1-6 hours are extrapolated through the intelligent prediction model. New measured data are acquired in real time. The deviation between the predicted value and the measured data is compared. Based on the deviation, the parameters of the intelligent prediction model are adjusted in reverse to form a closed loop of prediction, verification and optimization, and continuously improve the prediction accuracy of the intelligent prediction model.

[0137] The intelligent prediction model parameter back-adjustment method is to use the deviation as the input of the loss function, calculate the gradient of each layer parameter through the backpropagation algorithm, and use the optimizer to update the parameters according to the gradient (the optimizer can use Adam) to reduce the deviation between the predicted value and the measured value. The parameter update range of the intelligent prediction model can be selected to update all model parameters or update local parameters related to the current spatiotemporal features in order to improve the update efficiency.

[0138] By combining mesoscale meteorological models with computational fluid dynamics (CFD) models, a seamless connection is achieved from regional macroscopic wind fields to urban microscopic flow fields. Based on real urban 3D data modeling, the physical effects of surface features on airflow are directly considered. Errors in the CFD grid model are continuously corrected through real-time observation data, compensating for the uncertainties of pure numerical simulation and improving the ability to capture spatiotemporal abrupt phenomena. Furthermore, hidden patterns in the data are mined through intelligent prediction models to supplement the prediction of complex turbulent phenomena that are difficult to analyze by CFD models. This method can significantly improve the accuracy of urban boundary layer airflow simulation and provide more precise meteorological support services for atmospheric environmental monitoring, urban planning, aviation safety, and other fields.

[0139] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for reconstructing low-altitude meteorological digital twins based on three-dimensional entity data modeling, characterized in that: Includes the following steps: S1. Obtain initial wind field data for the urban area and its surrounding areas using a mesoscale meteorological model; S2. Using the wind field output by the mesoscale meteorological model as the background, a computational fluid dynamics model is introduced to refine the spatial grid of the urban area, generating a refined background flow field for the city. S3. Collect three-dimensional entity data of the city through a sensor array and construct a three-dimensional geographic information model to reflect the surface roughness characteristics; S4. Use the three-dimensional geographic information model as the boundary condition of the computational fluid dynamics model to simulate airflow and construct a CFD mesh model; The process of constructing a CFD mesh model is as follows: G1. Convert the parameters of surface roughness and building morphology in the 3D geographic information model into wall function boundary conditions of the computational fluid dynamics model. G2. Discrete equations based on computational fluid dynamics models are solved using the finite volume method. G3. Select a suitable turbulence model and couple it with the discrete equations to handle complex airflows; G4. Update the parameters of the turbulence model and the wall function boundary conditions of the computational fluid dynamics model in real time, and iteratively solve the coupled discrete equations until the flow field parameters converge to form a CFD mesh model. S5. Real-time acquisition of high-frequency wind field data in the 0-1000 meter boundary layer through low-altitude monitoring network to obtain multi-source real-time data; S6. Integrate multi-source real-time data with the refined urban background flow field to correct the error of the CFD grid model; S7. Based on real-time observation data, the corrected CFD mesh model is updated frequently to form a low-altitude three-dimensional twin model that includes spatial heterogeneity and spatiotemporal abrupt changes. S8. Establish an intelligent prediction model for the characteristics of urban boundary layer airflow and learn the laws of micro-meteorological evolution. S9. Combines real-time updated low-altitude three-dimensional twin model and intelligent prediction model to predict future airflow changes, and continuously optimizes the parameters of intelligent prediction model by comparing with new observation data.

2. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 1, characterized in that: The mesoscale meteorological model used is the WRF mesoscale meteorological model, and the process for obtaining the initial wind field data is as follows: A1. Download the global forecast system to obtain global forecast data, and process the topographic and underlying surface data through the WPS module in the mesoscale meteorological model; A2. Input global forecast data into the mesoscale meteorological model, set the simulation area to cover the city and its surrounding area within a range of 100-200km, and run a rapid cyclic assimilation for 3 hours; A3. By integrating observation data from ground automatic weather stations and radiosonde stations, a three-dimensional initial wind field containing wind speed, wind direction, and air pressure is generated.

3. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 2, characterized in that: The process of refining the spatial grid of an urban area using a computational fluid dynamics model is as follows: B1. The calculation scope is the area extending outward from the city center at a height of 5-10 times the building height. B2. Set boundary conditions according to the calculation range. Boundary conditions include entrance boundary, surface boundary, exit boundary and top boundary. B3. Use computational fluid dynamics models to refine the grid and perform fluid dynamics calculations on urban areas to generate a refined background flow field for the city.

4. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 3, characterized in that: The process of constructing a 3D geographic information model is as follows: C1. Filter the data collected by the sensor group to remove noise points in the data, and classify the processed data into point clouds, which are then classified into ground points, building points, and vegetation points. C2. Eliminate the spatial deviation between the lidar and UAV aerial images by matching the coordinate system of the classified point cloud with feature points; C3. Based on the point cloud after spatial deviation elimination, a continuous triangular mesh model is generated using a triangulation algorithm to form the surface contours of buildings and terrain. C4. Optimize the parameters of high-rise buildings and complex terrain using modeling software to ensure that the error between the building and terrain data in the triangular mesh model and the actual measurement data is less than 5%, thereby obtaining a three-dimensional geographic information model.

5. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 4, characterized in that: During the acquisition of multi-source real-time data, the sensor group is combined with the low-altitude monitoring network to obtain real-time data of the high-frequency wind field in the boundary layer. The real-time data is then processed by outlier removal and moving average filtering. The processed data is transmitted to the cloud server in real time via the 5G network to build a spatiotemporal database.

6. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 5, characterized in that: The error correction process for CFD mesh models is as follows: H1. Real-time data is interpolated onto the grid nodes of the CFD grid model using the inverse distance weighting algorithm; H2. Calculate the deviation between the measured value and the simulated value of the CFD mesh model, and use the Kalman filter algorithm to feed the error back to the boundary conditions of the CFD mesh model; H3. Evaluate the confidence level of the corrected CFD mesh model. If the confidence level is within the allowable range, proceed to S7. If the confidence level exceeds the allowable range, re-execute H1-H2.

7. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 6, characterized in that: The process of updating the CFD grid model includes: incorporating the latest real-time observation data into the CFD grid model at fixed time steps, capturing airflow differences between different land surface types through the Kriging interpolation algorithm, identifying sudden changes in the flow field caused by sudden weather by using the moving window standard deviation, and finally coupling the updated CFD grid model with the three-dimensional geographic information model to generate a low-altitude three-dimensional twin model.

8. The method for low-altitude meteorological digital twin reconstruction based on three-dimensional entity data modeling according to claim 7, characterized in that: The process of building an intelligent prediction model is as follows: K1. Extract flow field features and meteorological elements from the low-altitude three-dimensional twin model. The flow field features include wind speed gradient and vortex intensity, and the meteorological elements include temperature and humidity. K2 uses long short-term memory neural networks to build models using spatiotemporal sequence data as input. K3 utilizes historical twin data and observational data for supervised learning, and optimizes hyperparameters through cross-validation. The optimized model is the intelligent prediction model.

9. A method for reconstructing low-altitude meteorological digital twins based on three-dimensional entity data modeling according to claim 8, characterized in that: When predicting future airflow changes, a real-time low-altitude three-dimensional twin model is used as the initial condition. The flow field changes in the next 1-6 hours are extrapolated through the intelligent prediction model. New measured data are acquired in real time, and the deviation between the predicted value and the measured data is compared. Based on the deviation, the parameters of the intelligent prediction model are adjusted in reverse to form a closed loop of prediction, verification and optimization.

Images