A wind field super-resolution method and system based on virtual wind lidar

Abstract

The application discloses a wind field super-resolution method and system based on a virtual wind measurement laser radar, and belongs to the technical field of radar simulation. The method comprises the following steps: simulating a three-dimensional transient turbulent wind field according to a computational fluid dynamics model, obtaining a three-dimensional wake flow field containing the influence of airflow around a building, simulating the radial projection visual angle of a wind measurement laser radar to convert the wind field slice at a preset height into a corresponding radial velocity field, respectively generating simulated scanning wind field data and random sampling wind field data by simulating a structured radar scanning process and random sampling according to a preset radial distance resolution and a scanning step angle, constructing training data based on the random sampling wind field data and the radial velocity field to generate a modern generative denoising model, and inputting the simulated scanning wind field data into the trained modern generative denoising model to obtain a reconstructed radial wind field. The application solves the problems of pulse string interference, spectrum broadening, signal accumulation time extension and poor system stability caused by the prior art in practice.

Description

Technical Field

[0001] This invention relates to a wind field super-resolution method and system based on virtual wind measurement lidar, belonging to the field of radar simulation technology. Background Technology

[0002] Wind-measuring lidar, as a mainstream remote sensing technology for atmospheric wind field detection, retrieves wind field parameters by emitting laser beams and analyzing echo signals. It has been widely applied in meteorological observation and wind energy assessment. In complex urban environments, this technology is still being explored for wind field detection, but its inherent working principle limits its effectiveness. To improve performance, the industry has optimized detection capabilities by improving hardware solutions, such as employing pulse modulation techniques, aiming to enhance spatial resolution and data acquisition efficiency.

[0003] However, existing wind-measuring lidar has significant drawbacks in urban applications. On the one hand, its spatial resolution is insufficient, making it difficult to capture meter-level turbulence characteristics in building wakes. On the other hand, high-rise buildings cause beam blocking, resulting in large data gaps in the core area of ​​building wakes, thus facing challenges of sparse spatial sampling and beam blocking. While hardware improvements such as pulse modulation attempt to improve accuracy to address the resolution issue, they have introduced new problems in practice, including pulse crosstalk, spectral broadening, prolonged signal accumulation time, and poor system stability. Furthermore, solely pursuing hardware performance improvements drastically increases equipment costs and technical complexity, limiting the widespread application and deployment of this technology in fine-grained urban wind field detection. Summary of the Invention

[0004] The purpose of this invention is to provide a wind field super-resolution method and system based on virtual wind lidar. By combining the powerful probabilistic reasoning capability of the diffusion model with the structured spatial information obtained by Doppler wind lidar, the invention aims to solve the problems of pulse crosstalk, spectrum broadening, signal accumulation time extension, and poor system stability caused by existing technologies in practice.

[0005] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution.

[0006] In a first aspect, the present invention provides a wind field super-resolution method based on virtual wind-measuring lidar, comprising:

[0007] The three-dimensional transient turbulent wind field was simulated using a computational fluid dynamics model, resulting in a three-dimensional wake wind field that includes the influence of airflow around the building.

[0008] Based on the radial projection view of the three-dimensional wake wind field simulation wind measurement lidar, the wind field slices at a preset height are converted into the corresponding radial velocity field.

[0009] Based on the radial velocity field, simulated scanning wind field data is generated by simulating a structured radar scanning process according to a preset radial distance resolution and scanning step angle.

[0010] Based on the radial velocity field, random sampling wind field data is generated by random sampling according to the preset radial distance resolution and scanning step angle.

[0011] Based on the randomly sampled wind field data and the radial velocity field, a training data pair is constructed, and a modern generative denoising model is trained using the training data pair.

[0012] The simulated scanning wind field data is input into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a reconstructed radial wind field with consistent physical structure.

[0013] Furthermore, based on the computational fluid dynamics model, a three-dimensional transient turbulent wind field was simulated to obtain a three-dimensional wake wind field that includes the influence of airflow around the building, including:

[0014] The transient turbulent wind field around the target cubic building was simulated using the large eddy simulation method, and a three-dimensional wake wind field containing three Cartesian velocity components was obtained. The three Cartesian velocity components are the velocity components in the x-axis direction, y-axis direction, and z-axis direction, respectively.

[0015] Among them, the side length of the transient turbulent wind field around the target cube building, the computational domain, the atmospheric boundary layer flow field conditions, and the reference wind speed at the inlet boundary are preset.

[0016] Furthermore, based on the radial projection view of the three-dimensional wake wind field simulation wind measurement lidar, the wind field slices at a preset height are converted into corresponding radial velocity fields, including:

[0017] Extract wind field slices at a preset height from the three-dimensional wake wind field;

[0018] The radial projection angle of the wind-measuring lidar is simulated based on wind field slices at a preset height.

[0019] Based on the radial projection view of the wind-measuring lidar, the wind field slice at the preset height is converted into the corresponding radial velocity field according to the preset radar view, radar distance, and radial velocity calculation formula.

[0020] Furthermore, the formula for calculating the radial velocity is expressed as follows:

[0021] ;

[0022] In the formula, Indicates radial velocity, Indicates the radial direction of the wind-measuring lidar scan. Represents the velocity component in the x-axis direction. Represents the cosine function. This indicates the azimuth angle observed by the wind-measuring lidar. Represents the velocity component in the y-axis direction. Represents the sine function. This indicates the elevation angle of the wind-measuring lidar. This represents the velocity component along the z-axis.

[0023] Furthermore, based on the radial velocity field, simulated scanning wind field data is generated through a simulated structured radar scanning process according to a preset radial distance resolution and scanning step angle, including:

[0024] Based on the radial velocity field, a structured sparse scanning process is simulated according to the preset radial distance resolution and scanning step angle, and the occlusion effect generated when the laser beam intersects with the target cubic building is considered to generate simulated scanning wind field data.

[0025] Furthermore, based on the radial velocity field, random sampling wind field data is generated through random sampling according to a preset radial distance resolution and scanning step angle, including:

[0026] Based on the radial velocity field, according to the preset radial distance resolution and scanning step angle, by ignoring the occlusion effect caused when the laser beam intersects with the target cubic building, a different number of points are randomly sampled as sparse inputs, which form training data pairs with the radial velocity field to generate random sampling wind field data.

[0027] Furthermore, the modern generative denoising model is a conditional probability diffusion model.

[0028] Furthermore, a training data pair is constructed based on the randomly sampled wind field data and the radial velocity field, and a modern generative denoising model is trained using the training data pair, including:

[0029] Select a set of radial velocity field and corresponding randomly sampled wind field data from the training data pair;

[0030] A time step is randomly sampled, and noise corresponding to the time step is added to the radial velocity field according to a preset noise level to obtain a noisy wind field sample.

[0031] The noisy wind field sample, the corresponding time step, and the randomly sampled wind field data at the corresponding time step are input into the modern generative denoising model to train the modern generative denoising model to predict the noise added at the specified time step.

[0032] The loss value between predicted noise and real noise is calculated based on the root mean square error loss function, and the parameters of the modern generative denoising model are optimized through backpropagation and gradient descent to obtain the trained modern generative denoising model.

[0033] Furthermore, the simulated scanned wind field data is input into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a physically consistent and complete reconstructed radial wind field, including:

[0034] Repeat the following steps for sparse observation-guided iterative denoising reconstruction until the iteration termination condition is met: In each back-diffusion time step, set the time step to... The perturbation samples at any given time are spliced ​​with the simulated scanned wind field data along the channel dimension to obtain joint input features;

[0035] Using a pre-trained modern generative denoising model, the prediction time step is based on the joint input features. Noise components at any given time;

[0036] From time step Subtract the predicted time step from the perturbation sample at time t. The noise component at time step is used to obtain the time step. Perturbation samples at any given time;

[0037] By iteratively denoising in reverse time step, a reconstructed radial wind field with consistent physical structure is obtained when the iteration terminates.

[0038] Secondly, the present invention provides a wind field super-resolution system based on virtual wind-measuring lidar, comprising:

[0039] The 3D wake wind field generation module is used to simulate the 3D transient turbulent wind field based on the computational fluid dynamics model, and obtain the 3D wake wind field including the influence of the airflow around the building.

[0040] The radial velocity field generation module is used to convert wind field slices at a preset height into corresponding radial velocity fields based on the radial projection view of the three-dimensional wake wind field simulation wind measuring lidar.

[0041] The simulated scanning wind field data generation module is used to generate simulated scanning wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle, through a simulated structured radar scanning process.

[0042] The random sampling wind field data module is used to generate random sampling wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle;

[0043] The model training module is used to construct training data pairs based on the randomly sampled wind field data and the radial velocity field, and to train a modern generative denoising model using the training data pairs.

[0044] The radial wind field reconstruction module is used to input simulated scanned wind field data into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a reconstructed radial wind field with consistent physical structure.

[0045] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0046] 1. This invention combines the use of computational fluid dynamics simulation of three-dimensional wake wind field and further constructs an iterative reconstruction mechanism of training data generated by virtual wind measurement lidar scanning with a generative denoising model. This achieves super-resolution enhancement of the physical structure consistency of sparse radar observation data in complex airflow environments around buildings. The virtual wind measurement lidar scanning simulation of this invention effectively solves the problems of flow field detail loss and structural distortion caused by insufficient scanning resolution and lack of high-fidelity "sparse-complete" physical consistency training samples in traditional wind field reconstruction. Finally, it outputs a reconstructed radial wind field with both high spatial resolution and fluid dynamic physical consistency.

[0047] 2. This invention constructs a three-dimensional wake wind field containing complex turbulent characteristics around buildings using the large eddy simulation method, and strictly presets the building scale, computational domain boundary and atmospheric boundary layer conditions, achieving a high-fidelity simulation of the physical characteristics of the wind field in the actual urban environment. This provides an accurate physical benchmark for subsequent radar data reconstruction and solves the problem of flow field detail distortion caused by the simplified flow field model in traditional methods.

[0048] 3. This invention innovatively considers the occlusion effect of laser beams intersecting with buildings and ignores the occlusion effect to generate simulated scanning wind field data and randomly sampled wind field data. It also maps the three-dimensional wind field to the radar observation space through the radial velocity calculation formula, and constructs a variety of training data pairs that cover the limitations of actual scanning resolution and occlusion effects. This significantly improves the generalization ability and reconstruction robustness of the generative denoising model for complex scanning scenarios such as building occlusion and sparse observation.

[0049] 4. This invention adopts a conditional probability diffusion model as a modern generative denoising framework. It achieves high-resolution reconstruction guided by sparse observations through simulated scanning of wind field data and iterative denoising mechanism. The output is a reconstructed radial wind field with both high spatial resolution and consistency with hydrodynamic physics. This solves the problems of pulse crosstalk, spectrum broadening, signal accumulation time extension and poor system stability caused by existing technologies in practice. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a wind field super-resolution method based on virtual wind-measuring lidar provided in an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of a radial wind field with a viewing angle of 135 degrees provided in an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of a radial wind field with a 90-degree viewing angle provided in an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of a radial wind field with a viewing angle of 45 degrees provided in an embodiment of the present invention;

[0054] Figure 5 This is a schematic diagram of a radial wind field with a viewing angle of 180 degrees provided in an embodiment of the present invention;

[0055] Figure 6 This is a schematic diagram of the initial two-dimensional vector wind field provided in an embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram of the radial wind field with a viewing angle of 0 degrees provided in an embodiment of the present invention;

[0057] Figure 8 This is a schematic diagram of a radial wind field with a viewing angle of 225 degrees provided in an embodiment of the present invention;

[0058] Figure 9 This is a schematic diagram of a radial wind field with a viewing angle of 270 degrees provided in an embodiment of the present invention;

[0059] Figure 10 This is a schematic diagram of a radial wind field with a viewing angle of 315 degrees provided in an embodiment of the present invention;

[0060] Figure 11 This is a schematic diagram of a two-dimensional wind field projected onto a radial wind field according to an embodiment of the present invention;

[0061] Figure 12 This is a schematic diagram of the true values ​​of the radial wind field in the dataset provided in this embodiment of the invention;

[0062] Figure 13 This is a schematic diagram of the structured sampling results at a distance of 30 meters and a viewing angle of 1 degree, provided in an embodiment of the present invention.

[0063] Figure 14 This is a schematic diagram of the structured sampling results at a distance of 60 meters and a viewing angle of 2 degrees, provided in an embodiment of the present invention.

[0064] Figure 15 This is a schematic diagram of the reconstructed network architecture provided in an embodiment of the present invention;

[0065] Figure 16 This is a schematic diagram of the reconstruction results and error analysis provided in the embodiments of the present invention. Detailed Implementation

[0066] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0067] Example 1

[0068] like Figure 1 As shown in the figure, this embodiment introduces a wind field super-resolution method based on virtual wind measurement lidar, including:

[0069] Step 1: Simulate the three-dimensional transient turbulent wind field using a computational fluid dynamics model to obtain the three-dimensional wake wind field that includes the influence of airflow around the building.

[0070] This invention constructs a three-dimensional wake wind field containing complex turbulent characteristics around buildings using the large eddy simulation method. It accurately captures the dynamic evolution of the x / y / z velocity components in transient turbulence, achieving high-fidelity simulation of the physical characteristics of wind fields in urban building clusters, such as turbulence scale, vortex structure, and velocity gradient in the wake region. This provides a physical benchmark that conforms to the laws of fluid mechanics for subsequent radar data reconstruction and solves the problem of flow field detail distortion caused by the simplification of turbulence models in traditional numerical simulations.

[0071] Step 2: Based on the radial projection view of the three-dimensional wake wind field simulation wind measurement lidar, convert the wind field slices at the preset height into the corresponding radial velocity field.

[0072] This invention is based on three-dimensional wake wind field slices. By simulating the radial projection view and radial velocity calculation formula of wind-measuring lidar, the Cartesian velocity field is accurately mapped to the radial velocity field of the radar observation space. This realizes the physical projection conversion from three-dimensional flow field to two-dimensional radar observation, ensuring the physical consistency of subsequent scanning data generation and model training, and avoiding the velocity field distortion caused by neglecting the geometric relationship of the viewpoint in traditional projection methods.

[0073] Step 3: Based on the radial velocity field, simulated scanning wind field data is generated by simulating a structured radar scanning process according to the preset radial distance resolution and scanning step angle.

[0074] This invention generates simulated scanning wind field data that includes the occlusion effect by simulating the structured radar scanning process and quantifying the occlusion effect when the laser beam intersects with the building. This truly reflects the observation sparsity and data missing characteristics caused by building occlusion in actual radar scanning, and improves the generalization and reconstruction capability of the generative denoising model for complex observation scenarios such as partial occlusion and non-uniform sampling.

[0075] Step 4: Based on the radial velocity field, generate random sampling wind field data by random sampling according to the preset radial distance resolution and scanning step angle.

[0076] This invention generates randomly sampled wind field data that ignores occlusion effects through random sampling, and constructs diverse training data pairs with radial velocity fields. By combining preset radial distance resolution and scanning step angle parameters, it achieves joint encoding of radar scanning resolution limitations, sampling randomness, and physical constraints, effectively enhancing the model's robustness to sparse observation data distributions and avoiding the overfitting risk caused by the single data distribution in traditional supervised learning.

[0077] Step 5: Construct training data pairs based on the randomly sampled wind field data and the radial velocity field, and use the training data pairs to train a modern generative denoising model.

[0078] This invention utilizes training data pairs constructed from randomly sampled wind field data and radial velocity fields to train modern generative denoising models such as conditional probability diffusion models. Through deep learning, it implicitly encodes wind field physical characteristics such as turbulent structure, velocity continuity, and noise robustness, thereby achieving explicit constraints and high-dimensional feature extraction of complex wind field physical laws. This lays a model foundation that combines data-driven and physical constraints for subsequent high-precision reconstruction.

[0079] Step 6: Input the simulated scanned wind field data into the trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction to obtain a reconstructed radial wind field with consistent physical structure.

[0080] This invention inputs simulated scanning wind field data into a trained generative denoising model. Through an iterative denoising and reconstruction mechanism guided by sparse observations, it achieves an end-to-end conversion from low-resolution sparse observations to high-resolution, physically consistent, and complete reconstructed radial wind field. This breaks through the radial range resolution limitation of traditional radar scanning, while ensuring that the reconstruction results strictly conform to the laws of fluid dynamics, such as vortex continuity and the rationality of velocity field gradients. It also solves the problems of flow field detail distortion and structural fracture caused by the lack of physical constraints in traditional interpolation methods.

[0081] Example 2

[0082] Step 1: Simulate the three-dimensional transient turbulent wind field using a computational fluid dynamics model to obtain the three-dimensional wake wind field that includes the influence of airflow around the building.

[0083] Step 1.1: Predetermine the side length, computational domain, atmospheric boundary layer flow field conditions, and reference wind speed at the inlet boundary of the transient turbulent wind field around the target cubic building. In this embodiment, the side length of the transient turbulent wind field around the target cubic building is set to 10 meters, the computational domain to be a rectangular space of 500 meters × 500 meters × 50 meters, the atmospheric boundary layer flow field conditions to be neutral stratification, and the reference wind speed at the inlet boundary to be 10 m / s, to ensure the formation of a complex turbulent wake downstream of the obstacle.

[0084] Step 1.2: Simulate the transient turbulent wind field around the target cubic building using the large eddy simulation method to obtain a three-dimensional wake wind field containing three Cartesian velocity components, namely the velocity components in the x-axis direction, y-axis direction, and z-axis direction.

[0085] Step 2: Based on the three-dimensional wake wind field, extract wind field slices at a preset height from the three-dimensional wake wind field to simulate the radial projection angle of the wind-measuring lidar. Based on the preset radar angle and radar distance, convert the wind field slices at the preset height into the corresponding radial velocity field according to the radial velocity calculation formula.

[0086] Step 2.1: Extract wind field slices at a preset height from the three-dimensional wake wind field.

[0087] Step 2.2: Simulate the radial projection angle of the wind-measuring lidar based on the wind field slices at the preset height.

[0088] Step 2.3: Based on the radial projection view of the wind-measuring lidar, the wind field slice at the preset height is converted into the corresponding radial velocity field according to the preset radar view, radar distance and radial velocity calculation formula.

[0089] In this embodiment, the radial velocity calculation results at different azimuth angles observed by the wind-measuring lidar are as follows: Figures 2-11 As shown, where, Figures 2-5 as well as Figures 7-11 These represent the initial two-dimensional vector wind field. Figure 6 Radial wind field projection results at different azimuth angles.

[0090] In this embodiment, the radial velocity calculation formula is expressed as:

[0091] ;

[0092] In the formula, Indicates radial velocity, Indicates the radial direction of the wind-measuring lidar scan. Represents the velocity component in the x-axis direction. Represents the cosine function. This indicates the azimuth angle observed by the wind-measuring lidar. Represents the velocity component in the y-axis direction. Represents the sine function. This indicates the elevation angle of the wind-measuring lidar. This represents the velocity component along the z-axis.

[0093] Step 3: Based on the radial velocity field, simulated scanning wind field data is generated by simulating a structured radar scanning process according to the preset radial distance resolution and scanning step angle.

[0094] Step 3.1: By setting the scanning step angle and radial spatial resolution of the wind-measuring lidar, a structured sparse scanning process is simulated, and the occlusion effect caused by the intersection of the laser beam and the target cubic building is considered to generate occluded sparse observation data for model inference. In this embodiment, the occlusion effect caused by the intersection of the laser beam and the target cubic building is deliberately ignored in this process, which can ensure that the model can learn the complete wake physics characteristics, thereby having a stronger inference ability when facing real occlusion.

[0095] In this embodiment, based on the radial velocity field, a structured sparse scanning process is simulated according to a preset radial distance resolution and scanning step angle. The occlusion effect caused when the laser beam intersects with the target cubic building is also considered to generate simulated scanning wind field data. This embodiment considers the occlusion effect caused when the laser beam intersects with the building during this process, as shown in the attached figure. Figures 12-14 As shown.

[0096] Step 4: Based on the radial velocity field, generate random sampling wind field data by random sampling according to the preset radial distance resolution and scanning step angle.

[0097] In this embodiment, based on the radial velocity field, according to the preset radial distance resolution and scanning step angle, different numbers of points are randomly sampled as sparse inputs by ignoring the occlusion effect caused when the laser beam intersects with the target cubic building. These points form training data pairs with the radial velocity field, generating random sampling wind field data.

[0098] Step 5: Construct training data pairs based on the randomly sampled wind field data and the radial velocity field, and use the training data pairs to train a modern generative denoising model.

[0099] Step 5.1: Select a set of radial velocity field and corresponding random sampled wind field data from the training data pair;

[0100] Step 5.2: Randomly sample a time step and add noise corresponding to the time step to the radial velocity field according to the preset noise to obtain a noisy wind field sample;

[0101] Step 5.3: Input the noisy wind field sample, the corresponding time step, and the random sampled wind field data of the corresponding time step into the modern generative denoising model, and train the modern generative denoising model to predict the noise added at the specified time step;

[0102] Step 5.4: Calculate the loss value between the predicted noise and the real noise based on the root mean square error loss function, and optimize the parameters of the modern generative denoising model through backpropagation and gradient descent to obtain the trained modern generative denoising model.

[0103] In this embodiment, the root mean square error loss function is expressed as:

[0104] ;

[0105] In the formula, This represents the loss value of the loss function. Indicates model parameters, Indicates the expected value. Represents the radial velocity field, v sparse This represents the sparse observation data used as input, including noisy wind field samples, the corresponding time steps, and randomly sampled wind field data for the corresponding time steps. Indicates the corresponding time step. Indicates the preset noise, x t This represents the noisy wind field sample at the corresponding time step. This represents the output of a modern generative denoising model. Indicates a normal distribution. Represents the identity matrix.

[0106] In this embodiment, the modern generative denoising model is a conditional probability diffusion model.

[0107] Step 6: Input the simulated scanned wind field data into the trained modern generative denoising model for sparse observation-guided iterative denoising and reconstruction to obtain a physically consistent and complete reconstructed radial wind field, such as... Figure 15 , Figure 16 As shown, where Figure 15 This is a schematic diagram of the reconstructed network architecture provided in an embodiment of the present invention. Figure 16 This is a schematic diagram of the reconstruction results and error analysis provided in the embodiments of the present invention.

[0108] Step 6.1: Repeat the following steps for sparse observation-guided iterative denoising reconstruction until the iteration termination condition is met: Step 6.1.1: In each backdiffusion time step, set the time step to... The perturbation samples at any given time are spliced ​​with the simulated scanned wind field data along the channel dimension to obtain joint input features;

[0109] Step 6.1.2: Utilize a pre-trained modern generative denoising model to predict the time step based on the joint features. Noise components at any given time;

[0110] Step 6.1.3: From the time step Subtract the predicted time step from the perturbation sample at time t. The noise component at time step is used to obtain the time step. Perturbation samples at any given time;

[0111] Step 6.2: Denoising is achieved by iteratively reversing time steps, resulting in a reconstructed radial wind field with consistent physical structure at the end of the iteration.

[0112] Example 3

[0113] Based on the same inventive concept as other embodiments, this embodiment introduces a wind field super-resolution system based on virtual wind measurement lidar, including:

[0114] The 3D wake wind field generation module is used to simulate the 3D transient turbulent wind field based on the computational fluid dynamics model, and obtain the 3D wake wind field including the influence of the airflow around the building.

[0115] The radial velocity field generation module is used to convert wind field slices at a preset height into corresponding radial velocity fields based on the radial projection view of the three-dimensional wake wind field simulation wind measuring lidar.

[0116] The simulated scanning wind field data generation module is used to generate simulated scanning wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle, through a simulated structured radar scanning process.

[0117] The random sampling wind field data module is used to generate random sampling wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle;

[0118] The model training module is used to construct training data pairs based on the randomly sampled wind field data and the radial velocity field, and to train a modern generative denoising model using the training data pairs.

[0119] The radial wind field reconstruction module is used to input simulated scanned wind field data into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a reconstructed radial wind field with consistent physical structure.

[0120] For the specific functional implementation of each of the above modules, please refer to the relevant content in the method of Embodiment 1 or 2.

[0121] Example 4

[0122] Based on the same inventive concept as other embodiments, this embodiment describes a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods of Embodiment 1 or 2 described above.

[0123] Example 5

[0124] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions that, when executed by a processor, implement the steps of the methods described in Embodiment 1 or 2 above.

[0125] In summary, this invention, by combining the three-dimensional wake wind field simulated by computational fluid dynamics and further constructing an iterative reconstruction mechanism of training data pairs generated by virtual wind measurement lidar scanning and a generative denoising model, achieves super-resolution enhancement of the physical structure consistency of sparse radar observation data in complex airflow environments around buildings. The virtual wind measurement lidar scanning simulation of this invention effectively solves the problems of flow field detail loss and structural distortion caused by insufficient scanning resolution and lack of high-fidelity "sparse-complete" physical consistency training samples in traditional wind field reconstruction, and finally outputs a reconstructed radial wind field with both high spatial resolution and fluid dynamic physical consistency.

[0126] This invention constructs a three-dimensional wake wind field containing complex turbulent characteristics around buildings using the large eddy simulation method. It also strictly pre-sets the building scale, computational domain boundary, and atmospheric boundary layer conditions, achieving a high-fidelity simulation of the physical characteristics of the wind field in the actual urban environment. This provides an accurate physical benchmark for subsequent radar data reconstruction and solves the problem of flow field detail distortion caused by the simplified flow field model in traditional methods.

[0127] This invention innovatively considers the occlusion effect of laser beams intersecting with buildings and ignores the occlusion effect to generate simulated scanning wind field data and randomly sampled wind field data. It also maps the three-dimensional wind field to the radar observation space through the radial velocity calculation formula, and constructs a variety of training data pairs that cover the limitations of actual scanning resolution and occlusion effects. This significantly improves the generalization ability and reconstruction robustness of the generative denoising model for complex scanning scenarios such as building occlusion and sparse observation.

[0128] This invention employs a conditional probability diffusion model as a modern generative denoising framework. It achieves high-resolution reconstruction guided by sparse observations through simulated scanning of wind field data and an iterative denoising mechanism, outputting a reconstructed radial wind field that combines high spatial resolution with hydrodynamic physical consistency. This solves the problems of pulse crosstalk, spectrum broadening, signal accumulation time extension, and poor system stability caused by existing technologies in practice.

[0129] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A wind field super-resolution method based on virtual wind-measuring lidar, characterized in that, include: The three-dimensional transient turbulent wind field was simulated using a computational fluid dynamics model, resulting in a three-dimensional wake wind field that includes the influence of airflow around the building. Based on the radial projection view of the three-dimensional wake wind field simulation wind measurement lidar, the wind field slices at a preset height are converted into the corresponding radial velocity field. Based on the radial velocity field, simulated scanning wind field data is generated by simulating a structured radar scanning process according to a preset radial distance resolution and scanning step angle. Based on the radial velocity field, random sampling wind field data is generated by random sampling according to the preset radial distance resolution and scanning step angle. Based on the randomly sampled wind field data and the radial velocity field, a training data pair is constructed, and a modern generative denoising model is trained using the training data pair. The simulated scanning wind field data is input into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a reconstructed radial wind field with consistent physical structure.

2. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Based on the computational fluid dynamics model, a three-dimensional transient turbulent wind field was simulated, resulting in a three-dimensional wake wind field that includes the influence of airflow around the building, including: The transient turbulent wind field around the target cubic building was simulated using the large eddy simulation method, and a three-dimensional wake wind field containing three Cartesian velocity components was obtained. The three Cartesian velocity components are the velocity components in the x-axis direction, y-axis direction, and z-axis direction, respectively. Among them, the side length of the transient turbulent wind field around the target cube building, the computational domain, the atmospheric boundary layer flow field conditions, and the reference wind speed at the inlet boundary are preset.

3. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Based on the radial projection view of the three-dimensional wake wind field simulation wind measurement lidar, wind field slices at a preset height are converted into corresponding radial velocity fields, including: Extract wind field slices at a preset height from the three-dimensional wake wind field; The radial projection angle of the wind-measuring lidar is simulated based on wind field slices at a preset height. Based on the radial projection view of the wind-measuring lidar, the wind field slice at the preset height is converted into the corresponding radial velocity field according to the preset radar view, radar distance, and radial velocity calculation formula.

4. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 3, characterized in that, The formula for calculating the radial velocity is expressed as follows: ； In the formula, Indicates radial velocity, Indicates the radial direction of the wind-measuring lidar scan. Represents the velocity component in the x-axis direction. Represents the cosine function. This indicates the azimuth angle observed by the wind-measuring lidar. Represents the velocity component in the y-axis direction. Represents the sine function. This indicates the elevation angle of the wind-measuring lidar. This represents the velocity component along the z-axis.

5. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Based on the radial velocity field, simulated scanning wind field data is generated through a simulated structured radar scanning process according to a preset radial distance resolution and scanning step angle, including: Based on the radial velocity field, a structured sparse scanning process is simulated according to the preset radial distance resolution and scanning step angle, and the occlusion effect generated when the laser beam intersects with the target cubic building is considered to generate simulated scanning wind field data.

6. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Based on the radial velocity field, random sampling wind field data is generated through random sampling according to a preset radial distance resolution and scanning step angle, including: Based on the radial velocity field, according to the preset radial distance resolution and scanning step angle, by ignoring the occlusion effect caused when the laser beam intersects with the target cubic building, a different number of points are randomly sampled as sparse inputs, which form training data pairs with the radial velocity field to generate random sampling wind field data.

7. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, The modern generative denoising model is a conditional probability diffusion model.

8. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Training modern generative denoising models using training data includes: Select a set of radial velocity field and corresponding randomly sampled wind field data from the training data pair; A time step is randomly sampled, and noise corresponding to the time step is added to the radial velocity field according to a preset noise level to obtain a noisy wind field sample. The noisy wind field sample, the corresponding time step, and the randomly sampled wind field data at the corresponding time step are input into the modern generative denoising model to train the modern generative denoising model to predict the noise added at the specified time step. The loss value between predicted noise and real noise is calculated based on the root mean square error loss function, and the parameters of the modern generative denoising model are optimized through backpropagation and gradient descent to obtain the trained modern generative denoising model.

9. The wind field super-resolution method based on virtual wind-measuring lidar according to claim 1, characterized in that, Simulated scanned wind field data is input into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a physically consistent and complete reconstructed radial wind field, including: Repeat the following steps for sparse observation-guided iterative denoising reconstruction until the iteration termination condition is met: In each back-diffusion time step, set the time step to... The perturbation samples at any given time are spliced ​​with the simulated scanned wind field data along the channel dimension to obtain joint input features; Using a pre-trained modern generative denoising model, the prediction time step is based on the joint input features. Noise components at any given time; From time step Subtract the predicted time step from the perturbation sample at time t. The noise component at time step is used to obtain the time step. Perturbation samples at any given time; By iteratively denoising in reverse time step, a reconstructed radial wind field with consistent physical structure is obtained when the iteration terminates.

10. A wind field super-resolution system based on virtual wind-measuring lidar using the method described in any one of claims 1-9, characterized in that, include: The 3D wake wind field generation module is used to simulate the 3D transient turbulent wind field based on the computational fluid dynamics model, and obtain the 3D wake wind field including the influence of the airflow around the building. The radial velocity field generation module is used to convert wind field slices at a preset height into corresponding radial velocity fields based on the radial projection view of the three-dimensional wake wind field simulation wind measuring lidar. The simulated scanning wind field data generation module is used to generate simulated scanning wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle, through a simulated structured radar scanning process. The random sampling wind field data module is used to generate random sampling wind field data based on the radial velocity field, according to a preset radial distance resolution and scanning step angle; The model training module is used to construct training data pairs based on the randomly sampled wind field data and the radial velocity field, and to train a modern generative denoising model using the training data pairs. The radial wind field reconstruction module is used to input simulated scanned wind field data into a trained modern generative denoising model for sparse observation-guided iterative denoising reconstruction, resulting in a reconstructed radial wind field with consistent physical structure.

Images