Wind field simulation method and system based on multi-source radar adaptive driving and physical correction

Abstract

The application discloses a wind field simulation method and system based on multi-source radar adaptive driving and physical correction. The method comprises the following steps: acquiring vertical profiles and / or spatial scanning surface data of multiple wind measuring radars; adaptively selecting a multi-source fusion strategy according to the number and type of radars, combining background profiles and virtual constraint points to generate a three-dimensional initial wind field; eliminating divergence through three-dimensional Poisson projection to realize physical consistency correction; sending the corrected wind field into a lattice Boltzmann method solver, and iteratively solving and outputting a high-resolution three-dimensional wind field by means of an embedded sub-grid turbulence model and a wind direction adaptive boundary condition dynamic switching mechanism. The system comprises radar data acquisition and preprocessing, adaptive initial field construction, physical consistency correction, a solver, dynamic similarity conversion, result visualization and periodic scheduling management modules. The system realizes multi-radar adaptive fusion, quality conservation constraint initial field, high-frequency data continuous driving and wind direction boundary fast switching, and is used for real-time sensing and display of urban low-altitude wind fields.

Description

Technical Field

[0001] This invention relates to the field of urban wind field simulation and meteorological data fusion technology, specifically to a wind field simulation method and system based on multi-source radar adaptive driving and physical correction. Background Technology

[0002] With the rapid development of low-altitude economy (such as drone logistics and urban air traffic) and wind power generation, the demand for real-time perception and refined simulation of three-dimensional wind fields in complex urban micro-topography and building complex environments is becoming increasingly urgent. Currently, the technical approaches involved in urban wind field simulation mainly fall into the following categories, but all of them have certain limitations.

[0003] In numerical simulation, mesoscale numerical weather prediction models such as WRF can provide regional-scale wind field information, but their horizontal resolution is typically on the order of hundreds to kilometers, making it difficult to resolve details of wind flow around and wakes within building complexes. To compensate for this insufficient resolution, dynamic or statistical downscaling methods are often used in engineering, but these methods are mostly based on steady-state or empirical assumptions and have limited ability to characterize the instantaneous flow structure of complex urban canopies. Traditional computational fluid dynamics (CFD) methods based on architectures such as RANS and LES can achieve detailed simulations at the building scale, but their complex mesh generation and slow solver convergence mean that a single microscale flow field simulation often takes hours or even days, making it difficult to meet the real-time response requirements for wind field changes. The Lattice Boltzmann Method (LBM), as a mesoscale fluid simulation method, has advantages such as inherent parallelism and simple boundary treatment, and has attracted attention in the field of urban wind field simulation. However, an engineering solution that organically couples high-frequency observation data with the solver has not yet been developed.

[0004] In terms of observational data applications, modern wind-measuring radars can provide second-level three-dimensional wind field observation information, including single-point continuous profiles from vertical profile radars and spatial wind field distributions from scanning radars. However, the industry has not yet developed a complete systematic approach that can automatically adapt corresponding multi-source data fusion and boundary constraint strategies based on the actual radar deployment system, quantity, and spatial layout. Existing practices mostly rely on empirical interpolation methods (such as inverse range weighting) to extrapolate discrete observations to the entire computational domain, or use simplified variational analysis to establish the initial field. These approximations lack a basis for evaluating the reliability of the results when spatial constraint information is insufficient. Especially under the condition of multi-system radar networking, vertical detection and spatial scanning data have fundamental differences in dimensionality and spatiotemporal characteristics. How to construct a standardized fusion framework and establish an adaptive strategy for the initial field under different radar configurations remains a technological gap.

[0005] Regarding physical consistency assurance, existing solutions have the following shortcomings: First, the initial wind field construction lacks mass conservation constraints for incompressible flow. Simple superposition of multi-source data fails to satisfy divergence freedom conditions, leading to non-physical pressure oscillations and numerical adjustment processes in the fluid model during the initial computational phase, prolonging convergence time and even affecting computational stability. Second, the coupling between observational data and the solver largely remains at the static initialization stage, failing to establish a standardized mechanism for continuously integrating discrete observations into the dynamic simulation process. Third, boundary condition updates rely on empirical interpolation or extrapolation, lacking a correction mechanism based on flow physics, making it difficult to maintain momentum and mass balance within the computational domain. These issues collectively limit the application effectiveness of real-time wind field simulation in terms of accuracy and engineering reliability.

[0006] In summary, the existing technology has the following main drawbacks: 1. Mesoscale numerical models have insufficient spatial resolution, and traditional downscaling methods are mostly based on empirical assumptions, making it difficult to capture instantaneous flow characteristics at the building scale; 2. Traditional CFD methods are computationally time-consuming and lack real-time sensing and rapid response capabilities; 3. Existing LBM simulation studies have failed to solve the problem of effective integration with measured high-frequency wind field data, and lack adaptive initial field construction and boundary update strategies for multi-system and multi-radar deployment configurations; 4. Existing observation fusion methods mostly rely on empirical interpolation and approximation, lacking an initial field correction mechanism based on fluid dynamics physical constraints, making it difficult to guarantee the numerical stability of real-time calculations under complex environments. Summary of the Invention

[0007] Based on an in-depth analysis of the limitations of existing technologies, the core problem this invention aims to solve is: how to utilize a limited amount of wind-measuring radar observation data to achieve real-time, continuous, and physically reliable lightweight simulation and display of three-dimensional wind fields under complex urban terrain. Specifically, the following sub-problems need to be addressed: 1. Adaptive transformation of multi-array radar data: In practical engineering, given the limited number of wind-measuring radars and their varying deployment locations, how can discrete observation data be adaptively interpolated or mapped into a global initial flow field and the boundary conditions required by the model based on different radar spatial layouts?

[0008] 2. Problem of ensuring the physical consistency of the initial flow field: The initial wind field generated by pure mathematical interpolation usually does not conform to the laws of fluid mechanics. How to introduce physical constraints such as mass conservation to alleviate the numerical oscillation or divergence of the model in the early stage of calculation.

[0009] 3. Continuous dynamic driving problem of high-frequency measured data: The wind radar data is updated at a high frequency (seconds or minutes). How to establish a reasonable boundary condition update mechanism in the calculation process, smoothly introduce the continuous measured data into the fluid dynamics solver, and realize the continuous rolling simulation of the wind field.

[0010] 4. Computational efficiency and visualization issues under limited computing power: Due to the massive amount of fluid dynamics mesh data, how to optimize the algorithm logic or utilize local computing resources (such as the CPU / GPU of ordinary computers) to meet the efficiency requirements of real-time wind field calculation and 3D visualization while ensuring building-level microscale resolution.

[0011] To solve the above problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a wind field simulation method based on multi-source radar adaptive driving and physical correction, comprising: Acquire real-time observation data from at least one wind-measuring radar within the target area, the observation data including vertical profile data and / or spatial scan surface data; Based on the number and type of the wind-measuring radar, an appropriate multi-source data fusion strategy is adaptively selected to weight and fuse the real-time observation data with background profile data and virtual boundary constraint point data to generate a three-dimensional initial wind field covering the target computational domain. The physical consistency correction is performed on the three-dimensional initial wind field. The divergence is eliminated by the three-dimensional Poisson projection method to make it satisfy the mass conservation condition of incompressible fluid, and the corrected initial wind field is obtained. The corrected initial wind field is input into the lattice Boltzmann method solver, which incorporates a sub-lattice turbulence model and a dynamic switching mechanism for wind-direction adaptive boundary conditions. The solver is used for iterative calculations to output the three-dimensional wind field simulation results of the target area.

[0012] In one alternative approach, the adaptive selection of a corresponding multi-source data fusion strategy based on the number and type of the wind-measuring radars specifically includes: When there is only one vertical wind profiler radar, extrapolation is performed along the vertical layers based on the single-point wind profile, while the horizontal direction depends on the background field filling. When there is only one space scanning radar, its horizontal wind field sensing capability is combined with inversion methods to obtain the vertical distribution. When there are two radars, one is a vertical wind profiler radar and the other is a space scanning radar, the former is used to constrain the reference wind speed at each height level and the latter is used to correct the horizontal distribution. When there are three or more radars, the spatial interpolation accuracy is enhanced and the wind field gradient and divergence information is analyzed based on the spatial coverage capability of the hybrid system combination or the same system combination.

[0013] In one optional approach, the physical consistency correction of the three-dimensional initial wind field specifically includes: The horizontal wind components U and V of the three-dimensional initial wind field are subjected to spectral domain low-pass filtering. During filtering, the horizontal wind vector is first decomposed into two scalar fields, velocity and direction angle, and processed separately. Then, phase expansion is performed to eliminate jump noise. The filtered three-dimensional divergence field is calculated, and the Poisson equation is solved in wavenumber space by three-dimensional fast Fourier transform to obtain the velocity potential, where the DC component is forcibly set to zero. The computational domain is zero-filled in the horizontal direction and a cosine conical window function is applied to suppress periodic boundary effects; Subtracting the gradient of the velocity potential from the original velocity field yields a modified wind field that strictly satisfies the divergence freedom condition.

[0014] In one alternative approach, the lattice Boltzmann method solver employs a multi-relaxation time scheme and embeds a sub-lattice turbulence model. It estimates the local turbulence intensity using the norm of the non-equilibrium stress tensor and dynamically adjusts the relaxation rate corresponding to the shear mode.

[0015] In one alternative approach, the wind-direction adaptive boundary condition dynamic switching mechanism specifically includes: During the initialization phase, complete boundary type marking matrices are pre-calculated and stored for multiple preset wind direction sectors. The marking matrices distinguish between inlet boundaries, outlet boundaries, side boundaries, and solid wall boundaries. At each time step of the simulation, the sector to which the simulation belongs is determined based on the current observed wind direction; When the wind direction changes beyond a preset threshold, the active boundary marker matrix is ​​switched directly through memory overwrite without re-initializing the computational domain.

[0016] In one alternative approach, the node velocities at the inlet boundary are obtained by preprocessing the time-series wind speed data into a multidimensional array with time and height as the two axes and storing it in the GPU memory; in the GPU kernel function of the solver at each time step, bilinear interpolation is performed on the current simulation time and the height of the inlet node to obtain the velocity vector of each inlet boundary node in real time.

[0017] In one alternative approach, before inputting the modified initial wind field into the lattice Boltzmann method solver, a dynamic similarity transformation step is included: determining a geometric scaling factor based on the target feature length and the model feature length, while maintaining a consistent Reynolds number; and mapping the parameters of the actual physical space to the simulated lattice space based on the geometric scaling factor.

[0018] In one alternative approach, acquiring real-time observation data from at least one wind-measuring radar within the target area further includes: Timestamp alignment is performed on observation data from multiple radars to retain valid observation times shared by all stations; A uniform projection transformation is performed on the spatial coordinates, mapping all observation points from the geographic coordinate frame to a projected coordinate system in meters.

[0019] In one alternative approach, a periodic scheduling step is also included: the simulation calculation task is triggered periodically according to a configurable time interval, and the process automatically decides whether to sleep and wait or resume immediately based on the relationship between the previous calculation time and the time interval. Multiple instances are prevented from running concurrently through process status identifiers, and residual states are automatically cleaned up and the process is restarted when an abnormal exit occurs.

[0020] Secondly, the present invention also provides a wind field simulation system based on multi-source radar adaptive driving and physical correction, comprising: The radar data acquisition and preprocessing module is used to receive real-time observation data from multiple wind-measuring radars and perform quality control and spatiotemporal alignment. An adaptive initial field construction module is used to automatically select a fusion strategy based on the number and type of the wind measuring radars, and to weightedly fuse the observation data with the background profile and virtual boundary constraint points to generate a three-dimensional initial wind field. The physical consistency correction module is used to eliminate the divergence of the three-dimensional initial wind field through the three-dimensional Poisson projection method and output the corrected initial wind field. The lattice Boltzmann method solver has a built-in sublattice turbulence model and a wind-direction adaptive boundary condition dynamic switching mechanism for iterative flow field calculation based on the corrected initial wind field. The dynamic similarity conversion module is used to map actual physical space parameters to the simulated lattice space while maintaining the consistency of the Reynolds number. The results post-processing and visualization module is used to restore the physical units and perform lightweight processing on the solution results; The periodic scheduling manager is used to coordinate the periodic task connections of the radar data acquisition and preprocessing module, the adaptive initial field construction module, the physical consistency correction module, the lattice Boltzmann method solver, the dynamic similarity conversion module, and the result post-processing and visualization module, and has process protection, log management, and exception recovery functions.

[0021] In one alternative scheme, an urban geographic modeling module is also included, which reads digital elevation model data and building vector data of the target area, and after coordinate transformation and vector rasterization, superimposes them to generate a comprehensive terrain elevation matrix, and finally converts it into a Boolean distribution matrix of three-dimensional solid nodes and fluid nodes, which serves as the computational domain input of the solver.

[0022] In one alternative approach, the adaptive initial field construction module employs the following fusion strategy for different numbers and types of radars: With a single radar: vertical wind profiler radar relies on the background field, while space scanning radar combines inversion methods; When there are two radars: when there is one radar of each of the two systems, the vertical profile constrains the reference wind speed and the horizontal distribution of the scanning surface is corrected; when both are scanning systems, the distance weights of multiple stations are jointly constrained. When there are three or more radars: spatial interpolation is enhanced, supporting divergence and vorticity diagnosis and cross-validation.

[0023] In one alternative scheme, the lattice Boltzmann method solver adopts the D3Q19 multi-relaxation time scheme, and the sub-lattice turbulence model dynamically adjusts the relaxation rate of the shear stress-related modes by local turbulence intensity; the boundary condition dynamic switching mechanism pre-generates the boundary marker matrix corresponding to the four quadrant wind directions during initialization and keeps it in memory, and switches it in real time when the wind direction changes.

[0024] In one alternative approach, the system supports registration as a background service of the operating system, enabling automatic startup and crash recovery to adapt to long-term unattended operation scenarios.

[0025] The beneficial effects of this invention are as follows: (1) Adaptive Initial Wind Field Construction for Multiple Radars: This invention designs initial wind field construction strategies applicable to single or multiple radar deployment scenarios. Different numbers of radars correspond to different spatial interpolation and constraint schemes, and the boundary condition update strategy is also adjusted accordingly with the radar layout. This enables the system to flexibly adapt to radar configurations at different project sites.

[0026] (2) Physical consistency guarantee of the initial field: The initial field construction process introduces a mass conservation constraint mechanism that combines spectral domain low-pass filtering with three-dimensional Poisson projection to ensure that the three-dimensional wind field retains observation information while satisfying the divergence freedom condition. This step ensures that the initial field has physical reliability before being sent to the fluid dynamics solver.

[0027] (3) Real-time local operation: Through computational acceleration optimization of the fluid dynamics solver—including spatial mapping methods based on fluid similarity criteria, enhanced collision schemes, and subgrid-scale turbulence models—the system can complete a simulation calculation on a regular local computer in minutes, without relying on high-performance computing clusters or cloud computing resources. The solver supports both CPU and GPU computing modes and can be independently deployed and run in operating system environments such as Windows and Linux, thereby reducing the hardware and environment requirements in practical applications. The solver also supports heterogeneous computing platforms and cross-operating system deployment, significantly reducing the hardware and environment threshold for practical applications.

[0028] (4) Adaptive boundary condition switching based on wind direction: The fluid dynamics solver has a built-in mechanism for pre-calculation and real-time switching of four-quadrant boundary conditions based on wind direction. When the observed wind direction changes significantly, the system automatically switches the configuration of the inlet, outlet, and side boundaries on the CPU and GPU without re-initializing the computational domain, thus achieving seamless dynamic updating of boundary conditions.

[0029] (5) Second-level radar data drive: The wind-measuring radar provides observation data with a time resolution of seconds. The data processing flow of this invention can make full use of this high-frequency observation capability to provide the simulator with frequently updated input data.

[0030] (6) Urban geographic modeling and solver work closely together: The supporting urban geographic modeling tool transforms geographic information such as building clusters and terrain into the grid form required by the fluid dynamics calculation domain, realizing customized modeling for each simulation case area.

[0031] (7) Full-process automated scheduling: The system includes a complete cycle scheduling manager to realize automatic looping of data acquisition, initial field construction, solution calculation, result post-processing, and visualization. The scheduler has engineering features such as process locking, log rotation, exception recovery, and service-oriented deployment to ensure the stability of long-term unattended operation.

[0032] (8) Simulation accuracy assurance: The simulation results have been verified by comparison with classical computational fluid dynamics examples, by comparison with wind tunnel experimental data, and by consistency assessment of field observations to ensure physical reliability; at the same time, the simulation results are based on radar quality control observation data to ensure consistency with actual measurements. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the content of the embodiments of the present invention and these drawings without creative effort.

[0034] Figure 1 This is a flowchart of the wind field simulation method based on multi-source radar adaptive driving and physical correction provided in the embodiments of the present invention; Figure 2 This is a structural block diagram of the wind field simulation system based on multi-source radar adaptive driving and physical correction provided in the embodiments of the present invention; Figure 3 This is the verification result of the three-dimensional top cover driving square cavity flow calculation example provided in the embodiment of the present invention; Figure 4This is a comparison between simulated and wind tunnel measurement data of flow velocity profiles at three downstream locations (x / H = 1.0, 2.0, 4.0) provided in the embodiments of the present invention; Figure 5 This invention provides an embodiment of urban wind field simulation turbulent energy spectrum analysis. Detailed Implementation

[0035] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0036] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0037] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0038] In the description of this embodiment, the terms "upper," "lower," "left," and "right," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used solely for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In the description of the present invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, the terms "first" and "second" are merely used for descriptive distinction and have no special meaning.

[0039] This invention provides a wind field simulation method and system based on multi-source radar adaptive driving and physical correction. The method is as follows: Figure 1 As shown, it includes the following steps: S1. Acquire real-time observation data from at least one wind-measuring radar within the target area. The observation data includes vertical profile data and / or spatial scanning surface data. S2. Based on the number and type of wind measuring radar, an appropriate multi-source data fusion strategy is adaptively selected to weight and fuse real-time observation data with background profile data and virtual boundary constraint point data to generate a three-dimensional initial wind field covering the target computational domain. S3. Physical consistency correction is performed on the three-dimensional initial wind field. The divergence is eliminated by the three-dimensional Poisson projection method to make it satisfy the mass conservation condition of incompressible fluid, and the corrected initial wind field is obtained. S4. Input the corrected initial wind field into the lattice Boltzmann method solver. The solver has a built-in sub-lattice turbulence model and a dynamic switching mechanism for wind-direction adaptive boundary conditions. S5. Use the solver to perform iterative calculations and output the three-dimensional wind field simulation results of the target area.

[0040] System such as Figure 2 As shown, it includes: radar data acquisition and preprocessing module, urban geographic modeling module, adaptive initial field construction module, physical consistency correction module, lattice Boltzmann method solver, dynamic similarity conversion module, result post-processing and visualization module, and periodic scheduling manager. The modules rely on standardized data exchange interfaces to realize the flow and interaction of information.

[0041] Specifically, in the data source and environment preparation phase, the radar data acquisition and preprocessing module serves as the data entry point, responsible for receiving observation data from various types of wind-measuring radars. After quality control, coordinate transformation, and format standardization, it generates standard driver files that can be read by the underlying solver. Simultaneously, the urban geographic modeling module extracts the topographic elevation and building vector information of the target area, transforming it into a spatial distribution representation of solid regions within the computational domain, thereby completing customized computational space construction for different business scenarios.

[0042] In the preprocessing stage, the adaptive initial field construction module and the physical consistency correction module automatically match the corresponding multi-source data fusion strategy to generate an initial three-dimensional wind field based on the number and type of radars actually deployed on-site. Then, the initial field is adjusted through spectral domain constraints and divergence correction techniques to ensure that the global velocity field satisfies the mass conservation condition for incompressible flow. To balance computational accuracy and resource consumption, the dynamic similarity conversion module maps the actual physical space parameters to the internal simulation space according to the fluid similarity criterion, effectively reducing the mesh size and computational overhead while ensuring flow similarity.

[0043] The system's computational engine is based on a solver core using the lattice Boltzmann method. This solver employs a multi-dimensional discrete velocity model, coupled with an enhanced collision scheme and a sub-lattice turbulence model, and incorporates a dynamic boundary condition switching mechanism adapted to wind direction. The solver's underlying architecture ensures compatibility with mainstream operating systems and heterogeneous computing platforms, guaranteeing efficient iterative flow field calculations while reducing hardware and environmental requirements in practical applications. After computation, the result post-processing and visualization module performs dimensionality reduction and feature extraction on massive grid data, supporting functions such as flow field slicing at specified heights, coordinate translation, and trajectory analysis. Finally, the results are encapsulated into lightweight structured data and pushed to the front end via a standard interface for real-time rendering of the 3D wind field.

[0044] Specifically, heterogeneous computing platforms primarily refer to solvers that support operation on different hardware architectures, including: GPU-accelerated computing mode, such as using NVIDIA GPUs to execute the solution process in parallel; and CPU computing mode, which directly utilizes multi-core CPUs to complete calculations even without a dedicated GPU. In other words, the system can be deployed on workstations equipped with GPUs or on ordinary CPU computers, without relying on high-performance computing clusters or cloud resources. Mainstream operating systems include Windows and Linux environments, and the solver can be deployed and run independently on different operating systems without requiring a specific server environment.

[0045] Preferably, to ensure the closed-loop operation of the above-mentioned links, the system is equipped with an independent periodic scheduling manager. This module manages the periodic task connection between data input and simulation calculation, and integrates control logic such as process protection, log management, service-oriented deployment, and automatic anomaly recovery, ensuring the long-term, highly reliable operation of the entire process in an unattended environment.

[0046] Preferably, in the radar data acquisition and preprocessing stage, the system mainly processes two types of multi-source observation data: vertical profile and scanning surface, and generates unified solver input conditions through a spatiotemporal alignment mechanism. For vertical profile observations, the system extracts core metadata such as station geographic location, elevation, and time labels from the data files through an automatic parsing mechanism. The wind direction and speed information from the original observations is then converted into a unified analytical coordinate form to facilitate integration with subsequent fluid dynamics models. For the two-dimensional observation field provided by the space-scanning radar, the system similarly identifies parameters such as spatial range, resolution, and reference coordinates from the data files, and transforms the discrete sampling points acquired by the radar in a specific scanning mode into a regularized spatial grid representation.

[0047] To address the asynchronous nature of multi-source heterogeneous data, the system employs a strict timestamp alignment mechanism. First, the time stamps of the multi-station profile data are intersected and filtered to retain the valid observation times shared by all stations, forming a baseline time series. Then, this series is aligned and matched with the time stamps of the scanned surface data. A valid match is considered achieved when the time deviation falls within a configurable tolerance range, and the merged data sets thus constitute a standardized time section for a single global calculation. Regarding the spatial reference, to meet the model's requirement for measuring physical distances, the system performs a unified projection transformation on the spatial coordinates of all input data, mapping discrete observation points and grid cells from the original geographic coordinate framework to a meter-based projected coordinate system, thereby achieving a high degree of integration of multi-source radar data in both spatiotemporal dimensions.

[0048] Preferably, regarding urban geographic modeling and computational domain generation, the system first automatically determines the computational domain range based on the spatial distribution of all observation points: taking the bounding rectangle of the observation points as the basis, it extends outwards by a preset buffer distance to form a complete boundary, divides the grid in the horizontal direction at equal intervals according to configurable resolution, and independently sets the grid spacing and computational height upper limit in the vertical direction.

[0049] In terms of terrain processing, the system reads the digital elevation model (DEM) raster data covering the computational domain, and after coordinate reprojection and resampling, generates a ground elevation matrix aligned with the horizontal grid, automatically filtering out abnormal elevation values. Simultaneously, the system reads the building vector data of the target area, and after projection transformation and clipping, automatically identifies the height attribute field. Through vector rasterization, it transforms the building outlines and heights into a two-dimensional height matrix aligned with the horizontal grid.

[0050] The DEM elevation matrix and building height matrix are superimposed to form a composite terrain, which is then smoothed to suppress numerical artifacts. This smoothed terrain is then used for initial field construction and solver boundary marking. Grid nodes above ground level in the composite terrain are marked as solid nodes, ultimately generating a three-dimensional solid and fluid node distribution matrix, which serves as the computational domain input for the fluid dynamics solver.

[0051] Preferably, the present invention automatically selects the corresponding three-dimensional initial wind field construction strategy based on the number and spatial distribution of wind radars in the actual deployment scenario, thereby achieving adaptive initial field construction. The construction process consists of the following key steps: S210, multi-source weighted fusion The initial 3D wind field was constructed using a multi-source weighted fusion method. For each 3D grid node, its initial wind speed value was determined by dynamically weighting multiple data sources: Wind profile observation data: Wind profile information provided by each vertical sounding station is vertically interpolated, and spatial weights are calculated based on the horizontal distance between the station and the grid nodes. The weights decrease smoothly with distance, and their influence range can be adjusted by parameters to control the horizontal scale of the information from a single station.

[0052] Scanning surface observation data: Two-dimensional wind field distributions acquired by space-scanning radar, after appropriate spatial interpolation and noise suppression, are extended to the entire horizontal analysis area. The vertical influence of this type of data is controlled by an attenuation function related to the scanning height: it has the highest contribution near the scanning surface and gradually weakens with increasing vertical deviation distance. For terrain-obstructed or near-ground-obstructed areas, the vertical weights of the corresponding data are reasonably suppressed.

[0053] Background profile: A representative wind profile constructed by fusing vertical observation information from multiple stations, providing background constraints for areas with insufficient observation coverage. Its overall contribution intensity can be configured through parameters.

[0054] Virtual boundary constraint points: Virtual reference points are set at appropriate locations outside the analysis domain. The wind speed of these points is extrapolated from the background field and assigned a low confidence weight. This strategy is used to stabilize the interpolation behavior in sparse data regions and avoid non-physical extrapolation results near the boundaries.

[0055] S220, Vertical Structure Construction Strategy In the near-surface layer, when the target height is below the lowest height that reliable observations can cover, the boundary layer wind profile is used to extrapolate downwards to maintain the physical rationality of the near-surface wind speed. Within the effective observation height range, an appropriate interpolation method is used to achieve vertical continuity. Near the scanning surface height layer, the system extracts the horizontal wind field of the corresponding height layer as a spatial distribution reference, and combines it with the background profile to obtain the scale factor of each height layer relative to this reference surface, thereby implementing vertically consistent scaling of the horizontal wind field. This strategy ensures that the horizontal wind field at different height layers is spatially consistent with the scanning surface observations, while satisfying the wind field continuity constraint across the entire layer.

[0056] S230, terrain-induced vertical wind speed The vertical wind speed component is constructed through correlation calculations between topographic variation features and the horizontal wind speed component. The topographic information source is preprocessed topographic data. By leveraging the interaction between topographic gradients and local horizontal wind fields, an estimate of the vertical motion induced by the topography is obtained. The topographic-induced vertical wind speed primarily affects the near-surface layer and attenuates according to a set pattern with increasing altitude. This attenuation characteristic is controlled by configurable parameters, ensuring that the topographic dynamic effects are concentrated on correcting the low-level wind field and do not excessively affect the upper-level wind field structure.

[0057] S240, Mass Conservation Constraint Due to the influence of multi-source data superposition, the initial three-dimensional wind field after multi-source fusion and vertical structure construction usually does not satisfy the continuity equation (divergence freedom condition). This invention uses a three-dimensional Poisson projection method based on Fourier transform to perform mass conservation correction on the initial field. The specific steps are as follows: S241 Pre-filtering: Before projection, physically consistent spectral domain low-pass filtering is performed on the layer-by-layer horizontal wind components U and V. During filtering, the horizontal wind vector is decomposed into two scalar fields: velocity and direction angle, which are filtered separately to avoid direction artifacts caused by independent filtering of the U and V components. The direction angle needs to be phase-expanded before filtering to prevent the abrupt transition between 360° and 0° from introducing high-frequency noise.

[0058] S242 Three-Dimensional Divergence Calculation: The three-dimensional divergence is calculated using the pre-filtered U and V fields and the vertical velocity W field.

[0059]

[0060] Solving the Poisson equation S243: for the divergence field Perform a three-dimensional fast Fourier transform to solve the Poisson equation in wavenumber space:

[0061] Obtaining velocity potential The Poisson equation in wavenumber space then becomes:

[0062]

[0063] in and The spectral coefficients of the velocity potential and divergence are respectively. The wavenumber vector is used. The DC component (zero wavenumber) is forced to zero to ensure the uniqueness of the velocity potential. In addition, to mitigate the effects of periodic boundary effects, the computational domain is extended with adjustable zero-filling in the horizontal direction, and a cosine conical window function is applied for edge attenuation.

[0064] S244 Velocity Correction: The corrected velocity field is obtained by subtracting the velocity potential gradient from the original velocity field.

[0065] The corrected field strictly satisfies The corrected portion is then clipped back to the original computational domain. This mass-conserving projection is applied both globally to the 3D initial field and independently to the 2D preprocessing of the scanned surface data. The 2D projection uses the same spectral domain Poisson solution framework, forcibly satisfying the 2D incompressibility condition on the horizontal plane:

[0066] Two-dimensional projection requires solving the horizontal Poisson equation and correcting the U and V components accordingly.

[0067]

[0068] S250, strategic differences with varying numbers of radars The differences in the number and type of radars mainly lie in the dimension and richness of spatial constraint information. In actual network deployment, radars can be divided into two basic systems according to their detection modes: one type primarily aims to obtain the vertical distribution of wind fields over a single point, while the other has the ability to scan a certain airspace range. Each type has its own emphasis in terms of information dimension, thus requiring differentiated fusion constraint strategies under different configurations.

[0069] 1 radar: For vertical wind profile radar: only wind profile information is available above a single point, and the spatial constraint dimension exists only in the vertical direction. The initial field is constructed mainly by extrapolating the profile along the vertical layers of this single point. There is a lack of gradient constraints in the horizontal direction, which must be filled in by the background field or empirical profiles. The boundary conditions adopt a relatively simplified time update or steady assumption, and the background field has a relatively high weight.

[0070] If a scanning radar is used: Although there is only one site, its detection method gives it a certain ability to perceive the horizontal wind field in the near-site area. It can be combined with appropriate inversion methods to obtain the vertical wind field distribution, providing richer constraints for the initial field and boundary conditions than pure vertical detection. Overall, the background field still has a relatively high weight in this configuration.

[0071] 2 radars: When one radar of each system is used, the vertical wind profiler radar provides a continuous, high-temporal-resolution vertical structure, while the scanning radar provides horizontal wind field distribution characteristics within a spatial range. Combining the two systems achieves a balance between vertical stratification constraints and horizontal gradient estimation. During initial field construction, the vertical profiler constrains the baseline wind speed at each altitude level, while the scanning data corrects the horizontal distribution. During boundary updates, the spatial information from the scanning radar can be used to adjust boundary conditions in real time. When both systems are spatial scanning and the spatial distribution is reasonable, wind field information from two locations can be obtained, enabling estimation of the dominant wind field gradient within the effective coverage area. The reliability of initial field and boundary updates is improved compared to single-station systems. In spatial interpolation, the range weights of the two radars jointly constrain the horizontal distribution of the field, and the redundancy information in the overlapping area can be used for data quality control.

[0072] 3 radars: Under the hybrid system, the temporal continuity of vertical sensing information and the spatial coverage of spatial scanning information are superimposed, further enhancing the reconstruction effect of the three-dimensional wind field. If all systems are spatial scanning systems, the increase in sampling points allows for more refined spatial interpolation of the analysis area. In this case, diagnostic quantities such as divergence and vorticity of the horizontal wind field can be partially analyzed, and the constraint capability of boundary conditions is correspondingly enhanced. The overlapping area of ​​the three-station scans can provide cross-validation evidence for the quality assessment of wind field inversion.

[0073] 4 radars: A well-planned layout of the four stations ensures sufficient spatial constraints in all directions of the computational domain. When a spatial scanning system is predominant, multi-point collaboration effectively covers the horizontal gradient distribution across the entire field, resulting in more reliable initial field construction and better reflection of spatial wind field changes in boundary condition updates. This further reduces reliance on prior background information or virtual constraints. If the network includes a vertically sensing radar, its continuous vertical observation advantage can supplement the temporal constraints, further enhancing the spatiotemporal consistency of wind field fusion.

[0074] In summary, different radar systems have different strengths in vertical precision and spatial coverage. Vertical wind profiler radars excel at capturing the fine vertical structure and rapid evolution of wind fields, while scanning radars have advantages in horizontal spatial sampling and gradient constraints. Multi-radar networking strategies require careful consideration of the actual type and number of radars deployed, and the appropriate weighting of the fusion of vertical profile constraints and spatial scanning constraints to achieve optimal 3D wind field construction.

[0075] Preferably, the construction of a fluid dynamics solver based on the lattice Boltzmann method mainly includes the following: Lattice Boltzmann Model and Sublattice Turbulence Model: This invention employs a three-dimensional LBM model (D3Q19) in the multiple relaxation time (MRT) scheme as the core fluid dynamics solver. Compared to other three-dimensional fluid dynamics models, it significantly reduces computational resource requirements while maintaining sufficient accuracy, making it suitable for real-time calculations of large-scale urban wind fields. The model discretizes the momentum space into a uniform Cartesian grid and defines 19 discrete velocity directions (including the stationary direction) in three-dimensional space, each with a specific weighting coefficient. The model utilizes a particle distribution function to alternately perform collision and migration processes between grid points to reconstruct the velocity and mass changes of macroscopic flow. Its evolution equation is as follows:

[0076] in, Let be the particle distribution function along the i-th discrete velocity direction. It is a discrete velocity vector. Δt is the collision operator; x is the spatial position vector, representing the coordinates of the grid point in the three-dimensional Cartesian grid; Δt is the time step, representing the time interval between each iteration.

[0077] In the MRT scheme, the collision process transforms the velocity space distribution function to the moment space via a transformation matrix M. In the moment space, independent relaxation rates are applied to each physical mode (density, momentum, stress tensor, etc.), and then the process is reversed back to the velocity space. The relaxation rate corresponding to shear stress is determined by fluid viscosity, while the relaxation rates of other moments can be independently adjusted to optimize numerical stability. The post-collision distribution function... for:

[0078] in, The distribution function before the collision. Represented as a moment space, For equilibrium moment, This is a diagonal relaxation rate matrix, where each diagonal element corresponds to the relaxation rate of a different physical mode.

[0079] Urban wind field simulations involve high Reynolds number turbulence, and the grid resolution cannot resolve all turbulence scales. To resolve subgrid-scale eddies, this invention embeds a subgrid-scale turbulence model (SGS) into the MRT scheme. Local turbulence intensity is estimated using the norm of the non-equilibrium stress tensor, and the relaxation rate corresponding to the shear mode is dynamically adjusted.

[0080] in The relaxation time corresponds to the molecular viscosity. The additional relaxation time, determined by the local turbulence intensity, applies only to shear stress-related modes and has no effect on the relaxation of conserved quantities.

[0081] Ultimately, macroscopic density With speed Reconstruct from the zeroth and first moments of the distribution function:

[0082] Dynamic Similarity Unit Conversion System: The actual computational domain of urban wind fields can reach thousands of meters in size. Directly simulating the data at the actual physical scale using a grid would require an enormous number of grids and time steps, making computational resources unsustainable. Therefore, this invention designs a complete dynamic similarity unit conversion system to dynamically scale the actual physical space (prototype scale) and the simulation space (model scale), effectively reducing the computational scale while maintaining flow similarity. The system follows similarity criteria, determining the geometric scaling factor based on the specified prototype and model feature lengths to ensure consistent Reynolds numbers between the model and prototype. In discretization mapping, the grid spacing is determined by the feature length and grid resolution, the time step is determined by the target grid velocity and the feature velocity, and the relaxation time is derived from the viscosity coefficient in the grid space. The system pre-calculates a bidirectional conversion factor between grid units and physical units, used for restoring physical quantities in the output results and importing observational data for boundary conditions. Furthermore, during the initialization phase, the system automatically checks whether the key feature numbers are within a stable range; if they exceed a preset threshold, appropriate processing is performed to ensure the numerical reliability of the solution process.

[0083] Adaptive boundary condition mechanism: The system configures the following boundary conditions according to different interfaces of the computational domain: the solid wall adopts no-slip treatment, strictly satisfying the constraint that the velocity at the wall is zero; the wind inlet direction is set as the velocity inlet, and the unknown distribution function components on the boundary nodes are calculated by back-calculating the external profile; the top space adopts free slip condition, allowing horizontal flow to be unconstrained by shear; the outlet direction adopts an open boundary based on internal flow information, effectively suppressing non-physical numerical reflections, so that the internal flow field flows smoothly out of the computational domain.

[0084] When the prevailing wind direction changes during the simulation period, the system achieves a rapid response to boundary conditions through the following mechanism: (1) Pre-calculation stage: During initialization, a complete boundary type marking matrix is ​​pre-calculated for each preset wind direction sector, including boundary types such as inlet, outlet, side, and wall. The pre-calculation results reside in the video memory.

[0085] (2) Real-time switching phase: Each time step determines the sector to which it belongs based on the current wind direction. When the wind direction changes beyond the set threshold, the active boundary marker matrix is ​​switched by memory overwriting without recalculation.

[0086] (3) Dynamic interpolation of inlet profile: The time-series wind speed data is preprocessed into an array with time and height as the two axes. At each time step, bilinear interpolation is performed on the current simulation time and the height of the inlet node to obtain the velocity vector of each inlet node. At the same time, a configurable progressive acceleration mechanism is supported to smoothly transition during the simulation start-up phase to avoid initial transient impact.

[0087] Furthermore, after the fluid dynamics solution is completed, the system performs post-processing operations on the velocity field data, including solid region marking, physical unit restoration, and necessary coordinate transformations. The processed results are then output in multiple formats as needed: a self-describing scientific data format for storage and analysis, and a lightweight structured data format for front-end visualization. The latter supports extracting two-dimensional slices at a specified height from the three-dimensional field, adapting them, and encapsulating them into wind field distribution information in standard geographic coordinates for real-time rendering and interactive display.

[0088] Preferably, the present invention designs a dedicated periodic scheduling manager, the core mechanism of which includes: (1) Adaptive scheduling strategy: The manager starts the simulation calculation at a configurable time interval and automatically decides to sleep or resume immediately based on the relationship between the time taken by the previous calculation and the interval, so as to ensure the timeliness of the simulation.

[0089] (2) Concurrency control and status protection: The process status identification mechanism prevents multiple instances of the manager from running concurrently. After abnormal exit, the residual status is automatically cleaned up to ensure the reliable execution of the scheduling logic.

[0090] (3) Log and resource management: The simulation process output is recorded in real time and supports automatic rotation maintenance to prevent storage resources from getting out of control; it supports timeout control to prevent the computing process from getting stuck abnormally.

[0091] (4) Service-oriented deployment and self-recovery: The system supports registration as a background service of the operating system, enabling automatic startup and automatic restart after abnormal crashes, ensuring long-term unattended continuous operation.

[0092] Preferably, the simulation accuracy of the present invention is confirmed by a three-level verification system, which covers classical benchmark test, wind tunnel experiment comparison and field observation consistency evaluation.

[0093] (1) Verification by benchmark example To verify the computational accuracy of a high-performance 3D CFD solver developed based on the Lattice Boltzmann Method (LBM), a 3D top-cover driven square cavity flow with a Reynolds number Re=1000 was selected as a standard example. The simulation results were compared with the near-machine accuracy benchmark solution obtained by Albensoeder and Kuhlmann (2005) based on a high-precision spectral method. This benchmark solution is widely used as a reference standard in academia. Figure 3As shown, the left figure illustrates the distribution of dimensionless velocity along the z-direction on the vertical centerline of the three-dimensional square cavity flow under the condition of Re=1000, while the right figure shows the distribution of dimensionless vertical velocity along the x-direction on the horizontal centerline. The blue solid line in the figures represents the numerical solution of the solver developed in this invention based on the lattice Boltzmann method (LBM), and the red hollow dots represent the benchmark solution obtained by Albensoeder and Kuhlmann (2005) using the high-precision spectral method. Figure 3 As shown in Table 1, under steady-state convergence conditions, the minimum vertical velocity on the horizontal centerline is -0.435002, with a relative error of only 0.18% compared to the benchmark value of -0.434230, accurately capturing the intensity characteristics of the main vortex; the maximum vertical velocity is 0.248565, with a relative error of 1.84% compared to the benchmark value of 0.244074, effectively reflecting the flow characteristics of the secondary vortex system. The prediction errors of all key physical quantities are controlled within 2%, and the lateral position deviation of the main vortex is approximately 0.8%. The high consistency of the overall flow field structure indicates that this solver has reliable performance in complex three-dimensional turbulent field calculations, laying a solid computational foundation for subsequent research.

[0094]

[0095] Table 1. Comparison of solver simulation values ​​and baseline values ​​for key characteristic parameters of the three-dimensional top cover driven square cavity flow (Re=1000) (2) Verification of wind tunnel test data To further evaluate the applicability of the solver in simulating complex flows at the urban scale, the classic wind tunnel test conducted by Meng & Hibi (1998) was selected as the validation benchmark. This experiment systematically measured the mean flow velocity and turbulence characteristics around a cubic building under controlled conditions; the data is fully published and widely used for CFD model validation. This invention rigorously reproduces the geometric configuration and mean incoming flow conditions of the wind tunnel experiment, conducting simulations at the same Reynolds number (Re = 24,000). Figure 4As shown, the simulated and wind tunnel measured velocity profiles at three downstream locations (x / H = 1.0, 2.0, 4.0) are compared. The solid blue line represents the simulation results (average of 64 files) based on the LBM method of this invention, the hollow circles represent the experimental measurement data of Meng & Hibi (1998), and the red dashed line represents the building height (H). The normalized flow velocity profiles at the three downstream locations (x / H = 1.0, 2.0, 4.0) were compared with the wind tunnel measured data. The simulation results successfully captured key flow characteristics such as the backflow zone (negative velocity region) behind the building, the velocity recovery process, the downstream boundary layer reconstruction, and the overall morphology of the velocity profile in the wake region. The Pearson correlation coefficients at the three downstream locations reached 0.970, 0.986, and 0.995, respectively, indicating that the numerical results reproduced the trend and shape of the measured velocity distribution well. There is a certain deviation in the absolute value of the flow velocity, mainly due to the generation method of the inlet turbulence or the parameterization of the subgrid-scale model, which does not affect the consistency of the overall flow field structure.

[0096] (3) Consistency of field observations: verification of urban wind field inversion and turbulent energy spectrum Based on the verified solver described above, this invention constructs a 3D urban model based on real geographic information and deploys multiple laser wind radars for field observation. Based on the acquired high-frequency wind profile data, an observation data fusion initialization algorithm for LBM (Leadership in Wind Model) is developed, and a dynamic inflow boundary condition adaptive adjustment mechanism is established, realizing real-time 3D wind field inversion and updating driven by boundary observation. Figure 5 As shown, the left figure shows the variation of turbulent energy spectrum E(k) with wavenumber k for horizontal wind speed at different height levels (z = 30 m to 300 m). The black dashed line is the slope of the Kolmogorov-5 / 3 power law, and the black dotted line is the slope of -3. The right figure shows the distribution of building coverage at different height levels, reflecting the vertical stratification characteristics of urban underlying surface complexity. Figure 3 The simulated urban wind field turbulent energy spectrum analysis shows that the inverted wind field exhibits a physically consistent turbulent vortex structure in the vertical profile, demonstrating good spatial continuity and dynamic rationality. The horizontal wind speed turbulent energy spectra at each height level (z = 30 m to 300 m) agree well with the classical Kolmogorov-5 / 3 power law (inertial sub-region), proving the existence of a fully developed turbulent inertial sub-region in the flow field and verifying the physical self-consistency of the integrated inversion system. The mass-conserved Poisson projection applied in the initial field further ensures the physical consistency of the reconstructed wind field.

[0097] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0098] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A wind field simulation method based on multi-source radar adaptive driving and physical correction, characterized in that, include: Acquire real-time observation data from at least one wind-measuring radar within the target area, wherein the real-time observation data includes vertical profile data and / or spatial scanning surface data; Based on the number and type of the wind-measuring radar, an appropriate multi-source data fusion strategy is adaptively selected to weight and fuse the real-time observation data with background profile data and virtual boundary constraint point data to generate a three-dimensional initial wind field covering the target computational domain. The physical consistency correction is performed on the three-dimensional initial wind field. The divergence is eliminated by the three-dimensional Poisson projection method to make it satisfy the mass conservation condition of incompressible fluid, and the corrected initial wind field is obtained. The modified initial wind field is input into the lattice Boltzmann method solver, which has a built-in sub-lattice turbulence model and a dynamic switching mechanism for wind-direction adaptive boundary conditions; The solver is used for iterative calculations to output the three-dimensional wind field simulation results of the target area.

2. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, The adaptive selection of a corresponding multi-source data fusion strategy based on the number and type of wind-measuring radars specifically includes: When there is only one vertical wind profiler radar, extrapolation is performed along the vertical layers based on the single-point wind profile, while the horizontal direction depends on the background field filling. When there is only one space scanning radar, its horizontal wind field sensing capability is combined with inversion methods to obtain the vertical distribution. When there are two radars, one is a vertical wind profiler radar and the other is a space scanning radar, the former is used to constrain the reference wind speed at each height level and the latter is used to correct the horizontal distribution. When there are three or more radars, the spatial interpolation accuracy is enhanced and the wind field gradient and divergence information is analyzed based on the spatial coverage capability of the hybrid system combination or the same system combination.

3. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, The physical consistency correction of the three-dimensional initial wind field specifically includes: The horizontal wind components U and V of the three-dimensional initial wind field are subjected to spectral domain low-pass filtering. During filtering, the horizontal wind vector is first decomposed into two scalar fields, velocity and direction angle, and processed separately. Then, phase expansion is performed to eliminate jump noise. The filtered three-dimensional divergence field is calculated, and the Poisson equation is solved in wavenumber space by three-dimensional fast Fourier transform to obtain the velocity potential, where the DC component is forcibly set to zero. The computational domain is zero-filled in the horizontal direction and a cosine conical window function is applied to suppress periodic boundary effects; Subtracting the gradient of the velocity potential from the original velocity field yields a modified wind field that strictly satisfies the divergence freedom condition.

4. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, The lattice Boltzmann method solver employs a multi-relaxation time scheme and embeds a sub-lattice turbulence model. It estimates the local turbulence intensity through the norm of the non-equilibrium stress tensor and dynamically adjusts the relaxation rate corresponding to the shear mode.

5. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, The wind direction adaptive boundary condition dynamic switching mechanism specifically includes: During the initialization phase, complete boundary type marking matrices are pre-calculated and stored for multiple preset wind direction sectors. The marking matrices distinguish between inlet boundaries, outlet boundaries, side boundaries, and solid wall boundaries. At each time step of the simulation, the sector to which the simulation belongs is determined based on the current observed wind direction; When the wind direction changes beyond a preset threshold, the active boundary marker matrix is ​​switched directly through memory overwrite without re-initializing the computational domain.

6. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 5, characterized in that, The node velocities on the inlet boundary are obtained as follows: the time-series wind speed data is preprocessed into a multidimensional array with time and height as the two axes and stored in the GPU memory; in the GPU kernel function of each time step of the solver, bilinear interpolation is performed on the current simulation time and the height of the inlet node to obtain the velocity vector of each inlet boundary node in real time.

7. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, Before inputting the corrected initial wind field into the lattice Boltzmann method solver, a dynamic similarity transformation step is also included: The geometric scaling factor is determined based on the target feature length and the model feature length, while maintaining a consistent Reynolds number; the parameters of the actual physical space are mapped to the simulation lattice space based on the geometric scaling factor.

8. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to claim 1, characterized in that, The acquisition of real-time observation data from at least one wind-measuring radar within the target area further includes: Timestamp alignment is performed on observation data from multiple radars to retain valid observation times shared by all stations; A uniform projection transformation is performed on the spatial coordinates, mapping all observation points from the geographic coordinate frame to a projected coordinate system in meters.

9. The wind field simulation method based on multi-source radar adaptive driving and physical correction according to any one of claims 1 to 8, characterized in that, It also includes a periodic scheduling step: periodically triggering simulation calculation tasks according to a configurable time interval, automatically deciding whether to sleep and wait or immediately continue based on the relationship between the previous calculation time and the time interval, preventing multiple instances from running concurrently through process status identifiers, and automatically cleaning up residual states and restarting when exiting abnormally.

10. A wind field simulation system based on multi-source radar adaptive driving and physical correction, characterized in that, include: The radar data acquisition and preprocessing module is used to receive real-time observation data from multiple wind-measuring radars and perform quality control and spatiotemporal alignment. An adaptive initial field construction module is used to automatically select a fusion strategy based on the number and type of the wind measuring radars, and to weightedly fuse the observation data with the background profile and virtual boundary constraint points to generate a three-dimensional initial wind field. The physical consistency correction module is used to eliminate the divergence of the three-dimensional initial wind field through the three-dimensional Poisson projection method and output the corrected initial wind field. The lattice Boltzmann method solver has a built-in sublattice turbulence model and a wind-direction adaptive boundary condition dynamic switching mechanism for iterative flow field calculation based on the corrected initial wind field. The dynamic similarity conversion module is used to map actual physical space parameters to the simulated lattice space while maintaining the consistency of the Reynolds number. The results post-processing and visualization module is used to restore the physical units and perform lightweight processing on the solution results; The periodic scheduling manager is used to coordinate the periodic task connections of the radar data acquisition and preprocessing module, the adaptive initial field construction module, the physical consistency correction module, the lattice Boltzmann method solver, the dynamic similarity conversion module, and the result post-processing and visualization module, and has process protection, log management, and exception recovery functions.

11. The wind field simulation system based on multi-source radar adaptive driving and physical correction according to claim 10, characterized in that, It also includes an urban geographic modeling module, which reads digital elevation model data and building vector data of the target area, and after coordinate transformation and vector rasterization, superimposes them to generate a comprehensive terrain elevation matrix, and finally converts it into a Boolean distribution matrix of three-dimensional solid nodes and fluid nodes, which serves as the computational domain input of the solver.

12. The wind field simulation system based on multi-source radar adaptive driving and physical correction according to claim 10, characterized in that, The adaptive initial field construction module employs the following fusion strategies for different numbers and types of radars: With a single radar: vertical wind profiler radar relies on the background field, while space scanning radar combines inversion methods; When there are two radars: when there is one radar of each of the two systems, the vertical profile constrains the reference wind speed and the horizontal distribution of the scanning surface is corrected; when both are scanning systems, the distance weights of multiple stations are jointly constrained. When there are three or more radars: spatial interpolation is enhanced, supporting divergence and vorticity diagnosis and cross-validation.

13. The wind field simulation system based on multi-source radar adaptive driving and physical correction according to claim 10, characterized in that, The lattice Boltzmann method solver uses the D3Q19 multi-relaxation time scheme. The sublattice turbulence model dynamically adjusts the relaxation rate of the shear stress-related modes by adjusting the local turbulence intensity. The boundary condition dynamic switching mechanism pre-generates the boundary marker matrix corresponding to the four quadrant wind directions during initialization and stores it in memory, switching it instantly when the wind direction changes.

14. The wind field simulation system based on multi-source radar adaptive driving and physical correction according to claim 10, characterized in that, The system supports registration as a background service of the operating system, enabling automatic startup and automatic recovery from crashes, in order to adapt to long-term unattended operation scenarios.

Images