Sparse reconstruction method for minute-level urban microclimate wind field based on laser radar data
By combining Doppler wind lidar and ground meteorological stations with numerical weather prediction models, and employing an adaptive sparse scanning and alternating direction multiplier method optimization framework, we have achieved minute-level real-time reconstruction of urban micro-meteorological wind fields. This solves the problems of insufficient timeliness and precision in traditional methods and provides high-precision three-dimensional wind field data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- PEKING UNIV SHENZHEN GRADUATE SCHOOL
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-12
AI Technical Summary
Traditional urban micro-meteorological wind field sparse reconstruction methods suffer from limitations such as a single sparse prior model, simple multi-source data fusion methods, and insufficient real-time performance of algorithms, making it difficult to meet real-time monitoring needs and unable to effectively capture complex wind field distortions and eddy phenomena.
By employing Doppler wind lidar combined with ground meteorological stations and numerical weather prediction models, data is acquired through an adaptive sparse scanning strategy. An optimization framework based on alternating direction multipliers is constructed to perform efficient minute-level three-dimensional wind field reconstruction. By combining data fidelity terms and background field constraints, deep collaborative optimization of multi-source data is achieved.
It achieves real-time inversion of urban 3D wind fields with minute-level time updates and meter-level spatial resolution, resolving the contradiction between timeliness and precision in traditional methods, and providing high-precision 3D wind field data that is spatially continuous and physically consistent.
Smart Images
Figure CN122017793A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of urban meteorological monitoring and data processing technology, specifically a minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data. Background Technology
[0002] With the acceleration of urbanization, the impact of urban micrometeorological effects on residents' lives, air quality, and low-altitude flight safety is becoming increasingly prominent. The spatiotemporal evolution of complex wind field structures such as canyon winds and corner winds within urban building complexes is the core and challenge of micrometeorological research. Accurately and in real-time acquiring three-dimensional urban wind field information is of crucial strategic significance for smart city management, disaster early warning, and the development of the low-altitude economy. Currently, urban wind field information is mainly obtained through ground meteorological station networks and computational fluid dynamics numerical simulations. However, these traditional methods still have the following problems in practical applications:
[0003] Traditional ground-based meteorological station networks are constrained by the cost and physical space of the stations, resulting in a sparse network of stations. While such networks can provide meteorological data for individual points, they cannot effectively capture complex local phenomena such as wind field distortions, eddies, and abrupt changes caused by dense building clusters. The data update frequency is low, and the representativeness is insufficient. Although computational fluid dynamics numerical simulations can provide high-fidelity three-dimensional wind field distributions based on physical equations, they require complex geometric modeling, high-quality mesh generation, and other computational tasks, resulting in enormous computational time and making it difficult to meet the needs of real-time monitoring.
[0004] In recent years, wind-measuring lidar technology, with its advantages of high precision and high spatiotemporal resolution, has become an important tool for atmospheric wind field detection. It actively emits a laser beam and inverts the wind speed along the line-of-sight by detecting the Doppler shift of the backscattered signal from aerosol particles in the atmosphere, thus providing observations of sparse wind fields along the laser path. However, the inversion method typically assumes that the wind field is horizontally uniform within the scanned volume, which is not valid in urban environments with complex terrain and buildings. To obtain a three-dimensional wind field, a time-consuming volumetric scan is usually required; completing a full, high-density scan of a large urban area often takes tens of minutes or even longer.
[0005] In summary, traditional methods for sparse reconstruction of urban micrometeorological wind fields suffer from several problems: a single sparse prior model that fails to fully integrate the multiple sparsity characteristics of urban wind field segmentation smoothing and structured turbulence; a simplistic multi-source data fusion method that fails to achieve deep collaboration between lidar, ground station, and NWP data within a unified optimization framework; insufficient consideration of scanning strategy optimization to maximize information acquisition efficiency; and inadequate real-time performance of the algorithm, making it difficult to meet operational requirements. Summary of the Invention
[0006] The purpose of this invention is to provide a minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data, so as to solve the problems of sparse observation and low computational efficiency in traditional wind field monitoring technology mentioned in the background.
[0007] Therefore, this invention provides a minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data, comprising the following steps:
[0008] S1. Multi-source data acquisition: Deploy at least one Doppler wind lidar within the target urban area to acquire sparse radial wind speed observation data of the wind field within the target urban area. Simultaneously, it acquires fixed-point three-dimensional wind speed and direction observation data obtained from ground meteorological stations within the region. Low-resolution background wind field data covering the target area obtained by the sum-number weather prediction model ;
[0009] S2. Data Processing: Unify the acquired multi-source data to the same three-dimensional Cartesian coordinate system, perform time alignment and noise filtering, and process low-resolution background wind field data. Interpolate to the same high-resolution grid as the wind field to be reconstructed to obtain a high-resolution background wind field. ;
[0010] S3. Constructing an optimization model: Let u be the high-resolution three-dimensional wind field vector to be reconstructed, and construct the objective optimization function;
[0011] S4. Optimization Solution: The objective optimization function is iteratively solved using the alternating direction multiplier method. By introducing auxiliary variables to separate the regularized terms with the L1 norm, the original problem is decomposed into multiple subproblems for alternating updates until the convergence condition is met, thus obtaining the optimal high-resolution three-dimensional wind field solution. ;
[0012] S5. Output Results: Output and store the optimal high-resolution three-dimensional wind field solution. .
[0013] Preferably, in step S1, the Doppler wind-measuring lidar performs a preset adaptive sparse scan, and the scanning strategy is as follows:
[0014] Based on the building geographic information system data of the target area and prior wind field knowledge, key areas with drastic wind field changes are identified. Within a single scanning cycle, more scanning time and rays are allocated to perform intensive sampling of the key areas, while sparse sampling is performed on areas with relatively stable wind fields.
[0015] Preferably, the specific steps of data processing in step S2 are as follows:
[0016] S201. Establish a coordinate system: With the center point of the target city area as the origin O, the X-axis points due east, the Y-axis points due north, and the Z-axis points vertically upward, construct a three-dimensional Cartesian coordinate system as the carrier for the wind field u to be solved.
[0017] S202, Coordinate Transformation Unification: Utilizing the known coordinates of the Doppler wind lidar station site ( Convert it to a three-dimensional Cartesian coordinate system. The expression for the conversion is: In the formula The slant range of the radar. It is the azimuth angle. Angle of elevation;
[0018] The fixed geographic coordinates of each ground station are converted into coordinates in the target coordinate system through map projection calculations.
[0019] All grid points of the numerical weather prediction model data are unified into the target Cartesian coordinate system through projection and height conversion, forming a low-resolution three-dimensional background field;
[0020] S203, Time Unification: Set a time window, retain all radial wind speed data points collected by lidar within the time window as the observation set of the window, select the minute average observation value corresponding to the center time of the time window from the ground station data as the representative value of the window, and select the forecast that is closest in time to the start or center time of the reconstruction window from the numerical weather prediction model data as the background field of the window.
[0021] S204. Noise Removal and Data Cleaning: Calculate the signal-to-noise ratio (SNR) for each observation point, set a SNR threshold of -20 dB, and discard all observation points below this threshold to remove regions with weak signals and dominated by noise.
[0022] S205, Low-resolution background wind field data interpolation: Input low-resolution background wind field data Each grid point has three components: u, v, and w. The output is a high-resolution reconstruction of the wind field on the grid.
[0023] Preferably, in step S205, the specific steps for interpolating the low-resolution background wind field data are as follows:
[0024] Using trilinear interpolation, for each target point on the high-resolution target grid, the smallest cubic cell containing that point is found in the low-resolution background wind field grid.
[0025] Independent scalar interpolation is performed on the three wind speed components u, v, and w respectively. Specifically, in the u direction, two bilinear interpolations are first performed on the vertical layer of the low-resolution grid to obtain the u values at two virtual horizontal positions directly above and below the target point. Then, linear interpolation is performed between these two values to obtain the u component value at the target point. This process is repeated for the v and w components.
[0026] For points in the target grid that exceed the range of the low-resolution background wind field grid, nearest neighbor interpolation or extrapolation based on boundary layer theory is used. After interpolation, a high-resolution background wind field prior that is completely consistent with the u-dimensional wind field to be reconstructed is obtained. .
[0027] Preferably, in step S3, the specific steps for constructing the optimization model are as follows:
[0028] S301. Construct the lidar data fidelity term, the expression of which is: In the formula The L2 norm squared represents the least squares fitting under Gaussian noise. It represents the observation operator that projects the three-dimensional wind field onto the laser radar scanning direction. It is a large sparse matrix that sets the projection relationships of all observed points in the entire wind field u.
[0029] S302. Construct a data fidelity term for ground weather stations to ensure that the reconstructed wind field at the location of the ground weather station matches the actual observation value. The expression is: In the formula The sampling operator is a matrix composed of wind speed values extracted from grid points corresponding to the weather station locations across the entire wind field u. These are weighting coefficients, set based on the measurement accuracy and representativeness of the ground stations;
[0030] S303. Construct the fidelity term for numerical weather prediction model data, the expression of which is: This provides a physically reasonable default guess in areas where data is extremely sparse or nonexistent.
[0031] S304. Construct a multiple sparse regularization term, expressed as follows: In the formula This is a linear difference operator used to calculate the spatial differences of the wind field u in the x, y, and z directions, representing the piecewise smoothness of the wind field. This represents the change in wind speed in three directions at each grid point. This is a linear operator that transforms the wind field from physical space to characteristic space, representing the structured characteristics of turbulence. and These represent the regularization parameters that control the intensity of these two sparsities, respectively.
[0032] S305. The objective optimization function is obtained by integration, and its expression is: .
[0033] Preferably, the observation operator The construction method is as follows: for each radial wind speed observation by lidar, determine all high-resolution grid cells through which its scanning ray passes;
[0034] For each grid cell traversed, based on its center coordinates and the azimuth angle of the scanning ray and elevation angle Calculate the radial projection factor As the corresponding wind field component in the observation equation The coefficients of all observations are arranged in rows to form a sparse matrix. .
[0035] Preferably, the linear operator It is trained from historical high-fidelity computational fluid dynamics simulation data or historical high-density lidar scan data.
[0036] Preferably, in step S4, the specific steps for optimization are as follows:
[0037] S401, Introducing auxiliary variables and The objective function is equivalently transformed into the following constrained optimization problem, expressed as follows: , , ;
[0038] S402. Construct the augmented Lagrangian function for the constrained optimization problem and initialize the wind field variables. Auxiliary variables , and dual variables and ;
[0039] S403, Perform iterative updates, in the... In the next iteration:
[0040] fixed , , and The wind field variables are updated by solving a system of linear equations. ;
[0041] fixed , and Update auxiliary variables using a soft threshold function , ;
[0042] Update dual variables , ;
[0043] S404. Check if the original residual and dual residual are less than the preset threshold. If so, stop the iteration and output the result. As Otherwise Then return to step S403 to continue the iteration.
[0044] Preferably, in step S403, the coefficient matrix of the linear equation system is: , In the formula and The penalty parameters for the alternating direction multiplier method are solved using the preprocessed conjugate gradient method.
[0045] Preferably, in step S5, after the iteration converges, This means that the best high-resolution 3D wind field reconstructed within the current 2-minute period is archived by the server, marked with a timestamp, and pushed to authorized subscribers through a RESTful API interface.
[0046] The present invention proposes a minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data, the advantages of which are as follows:
[0047] This invention employs an adaptive sparse scanning strategy to perform intelligent data collection at the source, prioritizing the capture of key area information, which can effectively shorten the scanning time. Combined with a subsequent efficient sparse reconstruction algorithm, it can achieve real-time inversion of urban-scale three-dimensional wind fields with minute-level time updates and meter-level spatial resolution, thus solving the core contradiction of traditional reconstruction methods in that they cannot balance timeliness and precision.
[0048] A unified multi-source collaborative optimization framework was created, which deeply integrates the wide-area coverage capability of lidar, the accuracy of ground meteorological stations, and the macroscopic physical consistency of fluid dynamics numerical simulation models. By constructing data fidelity terms and background field constraint terms, complementary advantages are achieved, and a complete three-dimensional wind field with spatial continuity, physical consistency, and significantly improved accuracy is obtained in the entire target area, with strong anti-interference capability.
[0049] For high-dimensional, non-smooth optimization problems, the robust and efficient solution framework of Alternating Directional Multiplier Method (ADMM) is adopted. ADMM decomposes the original complex problem into a series of subproblems that can be solved in parallel or even have analytical solutions through variable splitting. It has good convergence. Combined with the aforementioned sparse scanning, it reduces the amount of data from the source, ensuring that the entire process from data collection, transmission, reconstruction to publication can be completed within a minute-level time window, which fully meets the stringent timeliness requirements of business applications such as meteorology and traffic management. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 This is a flowchart of the wind field sparse reconstruction method of the present invention;
[0052] Figure 2 A flowchart illustrating the construction of the optimization model for this invention;
[0053] Figure 3 This is a flowchart illustrating the optimization solution of this invention. Detailed Implementation
[0054] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described in this specification are merely for explaining the invention and are not intended to limit the invention.
[0055] Example:
[0056] Please see Figure 1-3 This invention provides a minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data. The method involves multi-source data acquisition: deploying at least one Doppler wind-measuring lidar within the target urban area to acquire sparse radial wind speed observation data of the wind field within the target urban area. Simultaneously, it acquires fixed-point three-dimensional wind speed and direction observation data obtained from ground meteorological stations within the region. Low-resolution background wind field data covering the target area obtained by the sum-number weather prediction model ;
[0057] In practical use: Select an area with a ground-based meteorological station as the target area. Within the target area, select three high-rise buildings approximately 1.5-2 kilometers apart. Deploy a long-range coherent Doppler wind lidar on the rooftop of each building. Specific parameters are set as follows: wavelength 1550nm, maximum detection distance 8km, radial velocity measurement range ±70m / s, accuracy ±0.1m / s, distance resolution 10m, equipped with a high-speed two-dimensional scanning galvanometer. The three lidars are connected to a central data server via a fiber optic network. The ground-based meteorological station provides 1-minute average three-dimensional wind speed, wind direction, temperature, humidity, and other data, which are transmitted to the server in real time via a 4G / 5G network. The central data server is equipped with two high-performance multi-core CPUs and at least two NVIDIA A100 or H100 GPU accelerator cards, ≥512GB of memory, a high-speed NVMe SSD array for real-time data processing, a large-capacity HDD array for historical data archiving, a high-speed internet interface for receiving numerical weather forecast data, and a Linux operating system, with CUDA toolkit, Python scientific computing stack, and parallel computing framework installed. Configure a web server and database, provide a RESTful API interface, and provide a professional terminal for departments such as urban management, environmental monitoring, and drone operation, supporting WebGL 3D visualization.
[0058] At the start of a 2-minute reconstruction cycle, the central server controls three lidar units to synchronously execute a pre-defined adaptive sparse scanning task: The total scanning time is allocated as follows: approximately 60% is used for multi-angle, intersecting planar position indicator and distance / height indicator scans of key areas, densely populated high-rise areas, main roads, and canyons to obtain rich wind field structure information; the remaining 30% is used for large-angle, low-density overview scans of other relatively open areas; and the final 10% is used for vertical profile detection of the sky to obtain the background wind profile at high altitudes. The scanning paths of the three lidar units are collaboratively optimized to ensure effective cross-coverage of their line-of-sight in space, especially in key areas, thereby improving the geometric accuracy of 3D wind speed inversion.
[0059] Three lidar units completed the scan and transmitted the massive amounts of radial velocity data, including time, location, radial velocity values, and signal-to-noise ratio, to the server in real time. Simultaneously, the server acquired average wind speed data from ground stations over minute intervals, as well as the latest numerical weather prediction model forecasts.
[0060] Data processing: The acquired multi-source data were unified to the same three-dimensional Cartesian coordinate system, and time alignment and noise filtering were performed. Low-resolution background wind field data were also processed. Interpolate to the same high-resolution grid as the wind field to be reconstructed to obtain a high-resolution background wind field. ;
[0061] In practical use: The server first unifies all data into a three-dimensional Cartesian coordinate system with the center of the study area as the origin. (This refers to the spherical coordinate data of the LiDAR.) Transform into The coordinates, including the latitude and longitude coordinates of the numerical weather prediction model, are also converted to these Cartesian coordinates.
[0062] For lidar data, invalid data points and isolated noise points are removed based on the signal-to-noise ratio threshold. A three-dimensional grid with a resolution of 10m x 10m x 10m is defined, covering an area of 4km x 4km x 0.5km. The three-dimensional wind field u to be reconstructed is defined as this huge vector with 400 × 400 × 50 = 8,000,000 grid points. The preprocessed lidar radial velocity observations are then... Three-dimensional wind speed and direction observation data obtained from ground meteorological stations Low-resolution background wind field data covering the target area obtained by the sum-number weather prediction model Associated with the nearest grid point or grid cell.
[0063] S3. Constructing an optimization model: Let u be the high-resolution three-dimensional wind field vector to be reconstructed, and construct the objective optimization function;
[0064] In practical use: Iterate through all valid LiDAR data points. For each data point, determine which 3D mesh cells the laser beam passed through. Then, based on its center coordinates... and the azimuth angle of the scanning ray and elevation angle Calculate the radial projection factor As the corresponding wind field component in the observation equation The coefficients of all observations are arranged in rows to form a sparse matrix. For ground weather stations, find their corresponding reconstructed grid index and construct a 15x1200000 sparse matrix S, where each row has only one element of 1, to extract the corresponding wind speed component of the grid point.
[0065] Construct a representation of a three-dimensional gradient operator The sparse difference matrix, This is the sum of the absolute values of the velocity differences between adjacent grid points in the x, y, and z directions. A high-precision CFD software is first run to perform large eddy simulations on the target area under several typical background wind directions (such as prevailing wind direction, summer wind direction, and winter wind direction), generating several high-fidelity wind field snapshots. These wind field snapshots are used as training data, and the K-SVD algorithm is used to train an overcomplete dictionary containing 2048 atoms. It is pre-stored on the server;
[0066] Based on experience and cross-validation, the initial weighting coefficients were set. , , as well as , The larger value is due to the high accuracy of the ground station data. The smaller value is because NWP is a weaker prior. and The parameters can be adaptively adjusted during system operation to balance data fidelity and solution sparsity.
[0067] Optimization Solution: The objective function is iteratively solved using the alternating direction multiplier method. Auxiliary variables are introduced to separate the L1 norm regularization term, decomposing the original problem into multiple subproblems for alternating updates until the convergence condition is met, yielding the optimal high-resolution three-dimensional wind field solution. ;
[0068] Introducing auxiliary variables and The objective function is equivalently transformed into the following constrained optimization problem, expressed as follows: , , ;
[0069] Construct the augmented Lagrangian function for the constrained optimization problem and initialize the wind field variables. Auxiliary variables , and dual variables and ;
[0070] Perform iterative updates, at the... In the next iteration:
[0071] fixed , , and The wind field variables are updated by solving a system of linear equations. ;
[0072] fixed , and Update auxiliary variables using a soft threshold function , ;
[0073] Update dual variables , ;
[0074] Check if the original residual and the dual residual are less than a preset threshold. If so, stop the iteration and output the result. As Otherwise Then return to the previous step and continue iterating.
[0075] The coefficient matrix of the linear equation system is In the formula and The penalty parameters for the alternating direction multiplier method are solved using the preprocessed conjugate gradient method.
[0076] Output and store the optimal high-resolution three-dimensional wind field solution. .
[0077] The optimal solution Reconstructed into a 400x400x50x3 four-dimensional array, with timestamps and metadata added, stored in a time-series database, and displayed on the client as a horizontal wind speed cloud map at a specified height with wind vector arrows overlaid, showing the vertical wind field structure along main streets or specific profiles, and using GPU rendering to generate three-dimensional streamline animations to intuitively display key vortex structures.
[0078] Through the RESTful API of the data server, the authorized system can be provided with services such as querying real-time wind speed and direction by latitude, longitude and altitude, subscribing to wind field change alarms for specific areas, and obtaining historical wind field data for analysis.
[0079] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data, characterized in that: Includes the following steps: S1. Multi-source data acquisition: Deploy at least one Doppler wind lidar within the target urban area to acquire sparse radial wind speed observation data of the wind field within the target urban area. Simultaneously, it acquires fixed-point three-dimensional wind speed and direction observation data obtained from ground meteorological stations within the region. Low-resolution background wind field data covering the target area obtained by the sum-number weather prediction model ; S2. Data Processing: Unify the acquired multi-source data to the same three-dimensional Cartesian coordinate system, perform time alignment and noise filtering, and process low-resolution background wind field data. Interpolate to the same high-resolution grid as the wind field to be reconstructed to obtain a high-resolution background wind field. ; S3. Constructing an optimization model: Let u be the high-resolution three-dimensional wind field vector to be reconstructed, and construct the objective optimization function; S4. Optimization Solution: The objective optimization function is iteratively solved using the alternating direction multiplier method. By introducing auxiliary variables to separate the regularized terms with the L1 norm, the original problem is decomposed into multiple subproblems for alternating updates until the convergence condition is met, thus obtaining the optimal high-resolution three-dimensional wind field solution. ; S5. Output Results: Output and store the optimal high-resolution three-dimensional wind field solution. .
2. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 1, characterized in that: In step S1, the Doppler wind-measuring lidar performs a preset adaptive sparse scan, and the scanning strategy is as follows: Based on the building geographic information system data of the target area and prior wind field knowledge, key areas with drastic wind field changes are identified. Within a single scanning cycle, more scanning time and rays are allocated to perform intensive sampling of the key areas, while sparse sampling is performed on areas with relatively stable wind fields.
3. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 1, characterized in that: The specific steps of data processing in step S2 are as follows: S201. Establish a coordinate system: With the center point of the target city area as the origin O, the X-axis points due east, the Y-axis points due north, and the Z-axis points vertically upward, construct a three-dimensional Cartesian coordinate system as the carrier for the wind field u to be solved. S202, Coordinate Transformation Unification: Utilizing the known coordinates of the Doppler wind lidar station site ( Convert it to a three-dimensional Cartesian coordinate system. The expression for the conversion is: In the formula The slant range of the radar. It is the azimuth angle. Angle of elevation; The fixed geographic coordinates of each ground station are converted into coordinates in the target coordinate system through map projection calculations. All grid points of the numerical weather prediction model data are unified into the target Cartesian coordinate system through projection and height conversion, forming a low-resolution three-dimensional background field; S203, Time Unification: Set a time window, retain all radial wind speed data points collected by lidar within the time window as the observation set of the window, select the minute average observation value corresponding to the center time of the time window from the ground station data as the representative value of the window, and select the forecast that is closest in time to the start or center time of the reconstruction window from the numerical weather prediction model data as the background field of the window. S204. Noise Removal and Data Cleaning: Calculate the signal-to-noise ratio (SNR) for each observation point, set a SNR threshold of -20 dB, and discard all observation points below this threshold to remove regions with weak signals and dominated by noise. S205, Low-resolution background wind field data interpolation: Input low-resolution background wind field data Each grid point has three components: u, v, and w. The output is a high-resolution reconstruction of the wind field on the grid.
4. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 3, characterized in that: In step S205, the specific steps for interpolating low-resolution background wind field data are as follows: Using trilinear interpolation, for each target point on the high-resolution target grid, the smallest cubic cell containing that point is found in the low-resolution background wind field grid. Independent scalar interpolation is performed on the three wind speed components u, v, and w respectively. Specifically, in the u direction, two bilinear interpolations are first performed on the vertical layer of the low-resolution grid to obtain the u values at two virtual horizontal positions directly above and below the target point. Then, linear interpolation is performed between these two values to obtain the u component value at the target point. This process is repeated for the v and w components. For points in the target grid that exceed the range of the low-resolution background wind field grid, nearest neighbor interpolation or extrapolation based on boundary layer theory is used. After interpolation, a high-resolution background wind field prior that is completely consistent with the u-dimensional wind field to be reconstructed is obtained. .
5. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 1, characterized in that: In step S3, the specific steps for constructing the optimization model are as follows: S301. Construct the lidar data fidelity term, the expression of which is: In the formula The L2 norm squared represents the least squares fitting under Gaussian noise. It represents the observation operator that projects the three-dimensional wind field onto the laser radar scanning direction. It is a large sparse matrix that sets the projection relationships of all observed points in the entire wind field u. S302. Construct a data fidelity term for ground weather stations to ensure that the reconstructed wind field at the location of the ground weather station matches the actual observation value. The expression is: In the formula The sampling operator is a matrix composed of wind speed values extracted from grid points corresponding to the weather station locations across the entire wind field u. These are weighting coefficients, set based on the measurement accuracy and representativeness of the ground stations; S303. Construct the fidelity term for numerical weather prediction model data, the expression of which is: This provides a physically reasonable default guess in areas where data is extremely sparse or nonexistent. S304. Construct a multiple sparse regularization term, expressed as follows: In the formula This is a linear difference operator used to calculate the spatial differences of the wind field u in the x, y, and z directions, representing the piecewise smoothness of the wind field. This represents the change in wind speed in three directions at each grid point. This is a linear operator that transforms the wind field from physical space to characteristic space, representing the structured characteristics of turbulence. and These represent the regularization parameters that control the intensity of these two sparsities, respectively. S305. The objective optimization function is obtained by integration, and its expression is: 。 6. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 5, characterized in that: The observation operator The construction method is as follows: for each radial wind speed observation by lidar, determine all high-resolution grid cells through which its scanning ray passes; For each grid cell traversed, based on its center coordinates and the azimuth angle of the scanning ray and elevation angle Calculate the radial projection factor As the corresponding wind field component in the observation equation The coefficients of all observations are arranged in rows to form a sparse matrix. .
7. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 5, characterized in that: The linear operator It is trained from historical high-fidelity computational fluid dynamics simulation data or historical high-density lidar scan data.
8. The minute-level sparse reconstruction method for urban micro-meteorological wind fields based on lidar data according to claim 5, characterized in that: In step S4, the specific steps for optimization are as follows: S401, Introducing auxiliary variables and The objective function is equivalently transformed into the following constrained optimization problem, expressed as follows: , ， ； S402. Construct the augmented Lagrangian function for the constrained optimization problem and initialize the wind field variables. Auxiliary variables , and dual variables and ; S403, Perform iterative updates, in the... In the next iteration: fixed , , and The wind field variables are updated by solving a system of linear equations. ; fixed , and Update auxiliary variables using a soft threshold function , ; Update dual variables , ; S404. Check if the original residual and dual residual are less than the preset threshold. If so, stop the iteration and output the result. As Otherwise Then return to step S403 to continue the iteration.
9. A method for sparse reconstruction of minute-level urban micro-meteorological wind fields based on lidar data according to claim 8, characterized in that: In step S403, the coefficient matrix of the linear equation system is: In the formula and The penalty parameters for the alternating direction multiplier method are solved using the preprocessed conjugate gradient method.
10. A method for sparse reconstruction of minute-level urban micro-meteorological wind fields based on lidar data according to claim 9, characterized in that: In step S5, after the iteration converges... This means that the best high-resolution 3D wind field reconstructed within the current 2-minute period is archived by the server, marked with a timestamp, and pushed to authorized subscribers through a RESTful API interface.