Method for reconstructing fine-scale turbulence using gan- and latent diffusion-based super-resolution models for urban air mobility operation support

Abstract

The present invention relates to a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations. More specifically, the invention relates to a method for reconstructing an ultra-high-resolution micro-scale turbulence field performed on a computer for supporting urban air mobility (UAM) operations, comprising the steps of: generating low-resolution urban wind field data of 100m by applying an area interpolation method to high-resolution urban wind field data of 5m resolution generated using PALM (Parallelized Large-Eddy Simulation Model); and training an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) based on a generative adversarial network using the low-resolution urban wind field data and the high-resolution urban wind field data as training data. The method is characterized by the following steps: inputting urban wind field data with a resolution of 100m into a trained Generative Adversarial Network-based ESRGAN or a pre-trained Latent Diffusion Model (LDM) to generate an ultra-high resolution micro-scale turbulent field with a resolution of 5m through a 20-fold improvement in spatial resolution; calculating turbulence intensity (TI) using u, v, and w components from the ultra-high resolution micro-scale turbulent field; classifying the turbulence into one of low turbulence conditions (TI<0.1), medium turbulence conditions (0.1≤TI<0.3), and high turbulence conditions (TI≥0.3) based on the turbulence intensity; and evaluating physical consistency by comparing the turbulent kinetic energy spectrum of the generated ultra-high resolution micro-scale turbulent field with the turbulent kinetic energy spectrum of the PALM-based high-resolution urban wind field data for each of the classified multiple turbulence conditions. According to the method for reconstructing ultra-high-resolution micro-scale turbulence fields based on generative adversarial networks and latent diffusion models for supporting urban air mobility operations proposed in this invention, by reconstructing ultra-high-resolution micro-scale turbulence fields from low-resolution urban wind field data using generative adversarial networks and latent diffusion models, the micro-scale turbulence structure of urban areas can be effectively reproduced without repeatedly performing high-resolution numerical simulations based on large vortex simulations, which are computationally expensive.

Description

Technology Field

[0001] The present invention relates to a method for reconstructing a microscale turbulence field, and more specifically, to a method for reconstructing an ultra-high-resolution microscale turbulence field based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations. Background Technology

[0002] The content described in this section merely provides background information regarding an embodiment of the present invention and does not constitute prior art.

[0004] Urban Air Mobility (UAM) is attracting attention as a next-generation mode of transportation that transports people or cargo using aircraft, such as electric vertical take-off and landing (eVTOL) vehicles, in low-altitude airspace over urban areas. For the safe operation of such urban air mobility, it is crucial to accurately predict atmospheric phenomena, such as wind speed variations, turbulence, and wind shear, that occur in the low-altitude urban regions where aircraft operate.

[0006] However, urban areas feature a complex distribution of buildings, roads, and terrain, leading to the formation of turbulent structures of various scales; these microscale turbulent fields significantly impact the stability and operational safety of aircraft. Therefore, high-resolution wind field data including microscale turbulent structures is required to support urban air mobility operations.

[0008] Generally, high-resolution wind field data for urban atmospheric environments can be generated through computational fluid dynamics (CFD) simulations or large eddy simulations (LES). For example, large eddy simulation models such as PALM (Parallelized Large-Eddy Simulation Model) can simulate complex turbulent structures in urban areas with high spatial resolution. However, such high-resolution numerical simulations require massive computational resources and long computation times, which limits their application to real-time or near-real-time navigation support systems.

[0010] Meanwhile, super-resolution technology for generating high-resolution data from low-resolution data is currently being researched in various fields. In particular, Generative Adversarial Network (GAN)-based super-resolution models and Diffusion Model-based super-resolution models are demonstrating excellent performance in the field of image processing, and their application is also being attempted in the meteorology and fluid dynamics sectors.

[0012] However, conventional super-resolution technologies focus primarily on improving image quality, which limits their direct application in the field of meteorology. Therefore, there is a need to develop a technology capable of effectively reconstructing high-resolution micro-scale turbulence fields from low-resolution urban wind field data, while also evaluating whether the reconstructed turbulence field maintains the characteristics of the actual turbulence field.

[0014] The aforementioned background technology is technical information that the inventor possessed for the derivation of the present invention or acquired during the process of deriving the present invention, and it cannot be considered as publicly known technology disclosed to the general public prior to the filing of the present invention. Prior art literature

[0015] Wang, 0-0.Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B., 2022. High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE / CVF conference on computer vision and pattern recognition, pp. 10684-10695. The problem to be solved

[0016] The present invention is proposed to solve the aforementioned problems of previously proposed methods, and aims to provide a method for reconstructing ultra-high-resolution micro-scale turbulence fields based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations, which can effectively reproduce the micro-scale turbulence structure of urban areas without repeatedly performing computationally expensive high-resolution numerical simulations based on large vortex simulations by reconstructing ultra-high-resolution micro-scale turbulence fields from low-resolution urban wind field data using a generative adversarial network and a latent diffusion model.

[0018] In addition, another objective of the present invention is to provide a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations, which can verify whether the generated turbulence field maintains the physical characteristics of an actual urban turbulence field by performing a physical consistency evaluation using turbulence intensity and turbulence kinetic energy spectra on the reconstructed turbulence field.

[0020] In addition, another objective of the present invention is to provide a method for reconstructing ultra-high-resolution micro-scale turbulence fields based on a generative adversarial network and a latent diffusion model for supporting urban air mobility (UAM) operations, which can rapidly generate high-resolution wind field information required for UAM operations and utilize it for operational support.

[0022] However, the technical problem that the present invention aims to solve is not limited to the technical problem described above, and other technical problems may exist. It goes without saying that objectives or effects that can be understood from the means of solving the problem or the embodiments, even if not explicitly mentioned, are also included. means of solving the problem

[0023] A method for reconstructing an ultra-high-resolution microscale turbulence field based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to the features of the present invention for achieving the above-mentioned purpose is,

[0024] In a method for reconstructing ultra-high-resolution micro-scale turbulence fields performed on a computer to support Urban Air Mobility (UAM) operations,

[0025] A step of generating low-resolution urban wind field data of 100m by applying an area interpolation method to high-resolution urban wind field data of 5m resolution generated using PALM (Parallelized Large-Eddy Simulation Model);

[0026] A step of training an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) based on a Generative Adversarial Network, using low-resolution urban wind field data and high-resolution urban wind field data as training data;

[0027] A step of inputting urban wind field data with a resolution of 100m into a trained Generative Adversarial Network-based ESRGAN or a pre-trained Latent Diffusion Model (LDM) to generate an ultra-high-resolution micro-scale turbulent field with a resolution of 5m through a 20-fold improvement in spatial resolution;

[0028] A step of calculating Turbulence Intensity (TI) using u, v, and w components from an ultra-high resolution micro-scale turbulent field;

[0029] A step of classifying into one of low turbulence conditions (TI<0.1), medium turbulence conditions (0.1≤TI<0.3) and high turbulence conditions (TI≥0.3) based on turbulence intensity; and

[0030] The method is characterized by a configuration that includes a step of evaluating physical consistency by comparing the turbulent kinetic energy spectrum of a generated ultra-high-resolution micro-scale turbulent field with the turbulent kinetic energy spectrum of PALM-based high-resolution urban wind field data for each of the multiple classified turbulent conditions.

[0032] Preferably,

[0033] The input data for ESRGAN is normalized using the mean and standard deviation for each of the u, v, and w components of the wind speed, and the generated turbulent field is denormalized using the same mean and standard deviation.

[0034] The input data of the LDM is normalized using the minimum and maximum values ​​for each wind speed component, and the generated turbulent field can be denormalized using the same minimum and maximum values.

[0036] Preferably, in the step of evaluating physical consistency,

[0037] By analyzing the energy distribution of the turbulent kinetic energy spectrum in the low, medium, and high wavenumber regions, it is possible to evaluate whether the energy spectrum of the generated turbulent field is preserved. Effects of the invention

[0038] According to the method for reconstructing ultra-high-resolution micro-scale turbulence fields based on generative adversarial networks and latent diffusion models for supporting urban air mobility operations proposed in this invention, by reconstructing ultra-high-resolution micro-scale turbulence fields from low-resolution urban wind field data using generative adversarial networks and latent diffusion models, the micro-scale turbulence structure of urban areas can be effectively reproduced without repeatedly performing high-resolution numerical simulations based on large vortex simulations, which are computationally expensive.

[0040] In addition, according to the present invention, by performing a physical consistency evaluation using turbulence intensity and turbulence kinetic energy spectra on the reconstructed turbulence field, it is possible to verify whether the generated turbulence field maintains the physical characteristics of an actual urban turbulence field.

[0042] In addition, according to the present invention, high-resolution wind field information required for urban air mobility (UAM) operation can be rapidly generated and utilized for operation support.

[0044] Furthermore, the various and beneficial advantages and effects of the present invention are not limited to those described above and may be more easily understood in the process of explaining specific embodiments of the present invention. Brief explanation of the drawing

[0045] FIG. 1 is a diagram illustrating the configuration of an apparatus for implementing a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. FIG. 2 is a diagram illustrating the flow of a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. Figure 3 is a diagram showing an example of a simulation domain of PALM. FIG. 4 is a diagram illustrating the structure of ESRGAN used in a method for ultra-high-resolution micro-scale turbulence field reconstruction based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. FIG. 5 is a diagram illustrating an LDM structure used in a super-high-resolution micro-scale turbulence field reconstruction method based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. Specific details for implementing the invention

[0046] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0048] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "indirectly connected" with other elements interposed between them. Furthermore, terms such as "include," "have," or "have" described below should be interpreted as indicating the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Additionally, singular expressions used in the present invention include plural expressions unless the context clearly indicates otherwise.

[0050] In addition, each component, process, procedure, or method included in each embodiment of the present invention may be shared within a scope that is not technically contradictory to one another.

[0052] Additionally, terms such as “…part,” “…unit,” and “module” described in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware, software, or a combination of hardware and software.

[0054] In addition, some of the operations or functions described as being performed by a terminal, device, or device in the present invention may instead be performed by a server connected to said terminal, device, or device. Likewise, some of the operations or functions described as being performed by a server may also be performed by a terminal, device, or device connected to said server.

[0056] In particular, the means for executing the system according to each embodiment of the present invention may be an application or a web server, and the terminal, which is the means for reading the recording medium on which the application or web server is recorded, may include not only general PCs such as general desktops or laptops, but also mobile terminals such as smartphones and tablet PCs.

[0058] The following examples are detailed descriptions to aid in understanding the present invention and are not intended to limit the scope of the present invention. Accordingly, inventions within the same scope that perform the same function as the present invention will also fall within the scope of the present invention.

[0060] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0062] FIG. 1 is a diagram illustrating the configuration of an apparatus for implementing a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. As shown in FIG. 1, the method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention can be performed in a system including a processor and memory.

[0064] Here, instructions for performing the method of the present invention may be stored in memory, and the processor may perform low-resolution urban wind field data generation, ESRGAN learning, ultra-high-resolution micro-scale turbulence field generation, turbulence intensity calculation, turbulence condition classification, and physical consistency evaluation by executing the instructions stored in memory.

[0066] FIG. 2 is a diagram illustrating the flow of a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial neural network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. As illustrated in FIG. 2, a method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a Generative Adversarial Network and a Latent Diffusion Model for supporting urban air mobility operations according to an embodiment of the present invention is a method for reconstructing an ultra-high-resolution micro-scale turbulence field performed on a computer for supporting urban air mobility (UAM) operations, comprising the steps of: generating low-resolution urban wind field data of 100m by applying an area interpolation method to high-resolution urban wind field data of 5m resolution generated using PALM (S100); training a Generative Adversarial Network-based ESRGAN using the low-resolution urban wind field data and the high-resolution urban wind field data as training data (S200); inputting the 100m resolution urban wind field data into the trained Generative Adversarial Network-based ESRGAN or a pre-trained LDM to generate an ultra-high-resolution micro-scale turbulence field of 5m resolution through a 20-fold improvement in spatial resolution (S300); and ultra-high-resolution micro-scale The method may be implemented by including the step (S400) of calculating turbulence intensity (TI) using u, v, and w components from a turbulence field; the step (S500) of classifying the turbulence into one of low turbulence conditions (TI<0.1), medium turbulence conditions (0.1≤TI<0.3), and high turbulence conditions (TI≥0.3) based on the turbulence intensity; and the step (S600) of evaluating physical consistency by comparing the turbulence kinetic energy spectrum of the generated ultra-high resolution micro-scale turbulence field with the turbulence kinetic energy spectrum of PALM-based high-resolution urban wind field data for each of the classified multiple turbulence conditions.

[0068] In step S100, low-resolution urban wind field data with a horizontal resolution of 100m can be generated by applying area interpolation to high-resolution urban wind field data with a horizontal resolution of 5m generated using PALM (Parallelized Large-Eddy Simulation Model). Here, PALM may be a model that simulates the micro-scale wind field and turbulence structure of an urban area based on Large Eddy Simulation (LES).

[0070] Figure 3 is a diagram showing an example of a simulation domain of PALM. As shown in Figure 3, the PALM simulation domain may be a 4 km × 4 km area including an urban area, and the urban area may be a complex urban environment including buildings, roads, and terrain structures.

[0072] High-resolution urban wind field data may have a grid resolution of 5m in the horizontal direction and a grid resolution of 3m near the surface in the vertical direction, and the grid spacing may be set to increase with altitude. In addition, high-resolution urban wind field data may include u, v, and w components of wind speed. Here, the u and v components correspond to horizontal wind speed components, and the w component corresponds to vertical wind speed components.

[0074] In step S100, the processor can generate low-resolution urban wind field data with a horizontal resolution of 100m by applying area interpolation to high-resolution urban wind field data with a horizontal resolution of 5m. Accordingly, high-resolution urban wind field data with a resolution of 5m and low-resolution urban wind field data with a resolution of 100m can be configured as corresponding input-correct pairs. The low-resolution urban wind field data may be data with reduced high-frequency components and microscale turbulent structures, and the high-resolution urban wind field data may be used as correct data or reference data for a super-resolution reconstruction model.

[0076] In one embodiment, PALM-based high-resolution urban wind field data may be generated from the results of a simulation performed for 24 hours and may include wind field data stored at predetermined time intervals. Input-correct-spot pairs may consist of a total of 23,328 samples, of which 18,565 are used for training and 4,763 are used for testing or evaluation.

[0078] In step S200, an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) based on a Generative Adversarial Network can be trained using low-resolution urban wind field data and high-resolution urban wind field data as training data. That is, ESRGAN can be trained using low-resolution urban wind field data as input training data and high-resolution urban wind field data as ground truth training data.

[0080] ESRGAN may be a super-resolution reconstruction model for generating a high-resolution micro-scale turbulent field from low-resolution urban wind field data. In one embodiment, ESRGAN may include a generator and a discriminator. Here, the generator may be configured to receive low-resolution urban wind field data as input and generate a high-resolution turbulent field, and the discriminator may be configured to compare the generated turbulent field with PALM-based high-resolution urban wind field data to determine the realism of the generation result.

[0082] FIG. 4 is a diagram illustrating the structure of ESRGAN used in a method for reconstructing an ultra-high-resolution micro-scale turbulent field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. As shown in FIG. 4, in the method for reconstructing an ultra-high-resolution micro-scale turbulent field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention, the generator of ESRGAN may include a plurality of Residual-in-Residual Dense Blocks (RRDBs). The RRDBs may include multiple residual connections and dense connections, thereby enabling the learning of multi-scale flow structures and fine wind speed fluctuations within the turbulent field. In one embodiment, the generator of ESRGAN may include an initial convolutional layer, a plurality of RRDBs, an upsampling layer, and an output layer.

[0084] In the training process of ESRGAN, at least one of pixel-level loss, perceptual loss, and adversarial loss may be used. For example, pixel-level loss may be used to reduce the difference between the generated turbulent field and high-resolution urban wind field data, perceptual loss may be used to ensure similarity of spatial structure, and adversarial loss may be used to induce the generated turbulent field to have a fine structure similar to high-resolution reference data. Additionally, ESRGAN may be further trained using a loss function that includes perceptual loss and adversarial loss after pre-training on ESRNet based on L1 loss.

[0086] Meanwhile, wind speed data input to ESRGAN can be normalized by wind speed component. Specifically, the input data to ESRGAN can be z-normalized using the mean and standard deviation for each of the u, v, and w components of the wind speed. In this case, the mean and standard deviation for each wind speed component can be calculated from high-resolution training data, and the calculated mean and standard deviation can be commonly used for the inverse normalization of low-resolution input data, high-resolution ground truth data, and the generated turbulence field. Accordingly, wind speed components with different dynamic ranges can be stably processed in the same training space.

[0088] In step S300, urban wind field data with a horizontal resolution of 100m can be input into a trained Generative Adversarial Network-based ESRGAN or a pre-trained Latent Diffusion Model (LDM) to generate an ultra-high-resolution micro-scale turbulent field with a horizontal resolution of 5m through a 20-fold improvement in spatial resolution. In one embodiment, ESRGAN can directly receive low-resolution urban wind field data with a horizontal resolution of 100m and generate an ultra-high-resolution micro-scale turbulent field with a horizontal resolution of 5m. That is, ESRGAN may be trained to directly perform a 20-fold super-resolution reconstruction. Similar to the training step of step S200, the input data of ESRGAN is normalized using the mean value and standard deviation for each of the u, v, and w components of wind speed, and the generated turbulent field can be denormalized using the same mean value and standard deviation. In one embodiment, the ESRGAN generator may include 23 RRDB blocks and 64 feature channels, and the discriminator may have a VGG-style discriminator structure.

[0090] FIG. 5 is a diagram illustrating the LDM structure used in the method for reconstructing a super-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention. As shown in FIG. 5, in the method for reconstructing a super-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations according to an embodiment of the present invention, the LDM may be configured to encode input data into a latent space, perform diffusion and reconstruction processes in the latent space, and then decode the reconstructed latent representation to generate a high-resolution result. The LDM may be a model pre-trained on a natural image dataset, and in an embodiment of the present invention, it may be applied to the super-resolution reconstruction of low-resolution urban wind field data without retraining on separate turbulence field data.

[0092] In one embodiment, the LDM may be a model pre-trained to perform 4x super-resolution restoration, and the processor may align the input and output resolutions of the LDM and then perform a 20x spatial resolution enhancement corresponding to a 100m horizontal resolution to a 5m horizontal resolution through interpolation or projection. In this case, the input data of the LDM may be normalized using minimum and maximum values ​​for each wind speed component, and the generated turbulent field may be denormalized using the same minimum and maximum values. More specifically, to match the input characteristics of the pre-trained LDM, the input urban wind field data may be normalized to the range [0, 1] using minimum and maximum values ​​for each wind speed component. Subsequently, the result generated by the LDM may be denormalized using the same minimum and maximum values ​​used for normalizing the input data. Accordingly, it is possible to restore wind speed values ​​corresponding to the original physical units while applying the LDM pre-trained based on natural images to the wind speed field data.

[0094] Ultra-high resolution microscale turbulent fields can be generated for the u, v, and w components of wind speed, respectively. The turbulent field generated by ESRGAN may contain relatively sharp micro-turbulent structures, while the turbulent field generated by LDM may have a relatively gentle spatial distribution and retain large-scale flow structures.

[0096] In step S400, turbulence intensity (TI) can be calculated using the u, v, and w components from an ultra-high resolution micro-scale turbulence field. Here, turbulence intensity can be calculated based on the variability of the horizontal wind speed component and the average horizontal wind speed. For example, turbulence intensity can be calculated based on the ratio of the variation of the u and v components and the average horizontal wind speed over a predetermined time interval.

[0098] In step S500, based on the turbulence intensity, the turbulence field can be classified into one of low turbulence conditions (TI < 0.1), medium turbulence conditions (0.1 ≤ TI < 0.3), and high turbulence conditions (TI ≥ 0.3). That is, in one embodiment, the processor classifies the turbulence field as a low turbulence condition when the turbulence intensity TI is less than 0.1, classifies the turbulence field as a medium turbulence condition when the turbulence intensity TI is 0.1 or more and less than 0.3, and classifies the turbulence field as a high turbulence condition when the turbulence intensity TI is 0.3 or more.

[0100] The low-turbulence, medium-turbulence, and high-turbulence conditions classified in this manner in step S500 can be used as criteria for evaluating the degree of preservation of physical characteristics of the generated ultra-high-resolution micro-scale turbulence field according to turbulence intensity. Generally, as turbulence intensity increases, the proportion of micro-scale turbulence structures and high-frequency energy components increases; therefore, the performance of the super-resolution reconstruction model may also vary depending on the turbulence conditions. Accordingly, in step S600, which will be described in detail below, the turbulence kinetic energy spectrum is calculated for each of the low-turbulence, medium-turbulence, and high-turbulence conditions, and the physical consistency according to turbulence intensity can be evaluated by comparing it with the turbulence kinetic energy spectrum of PALM-based high-resolution urban wind field data.

[0102] In step S600, physical consistency can be evaluated for a generated ultra-high-resolution micro-scale turbulent field by comparing the Turbulent Kinetic Energy Spectrum with the Turbulent Kinetic Energy Spectrum of PALM-based high-resolution urban wind field data for each of the classified multiple turbulent conditions. More specifically, in step S600, the energy distribution in the low-wavenumber, mid-wavenumber, and high-wavenumber regions of the Turbulent Kinetic Energy Spectrum can be analyzed to evaluate whether the energy spectrum of the generated turbulent field is preserved.

[0104] The turbulent kinetic energy spectrum may be an indicator of how well the generated turbulent field preserves the energy distribution at a spatial scale. In one embodiment, the turbulent kinetic energy spectrum may be calculated as a one-dimensional power spectral density by performing a discrete Fourier transform in the y-direction on each of the u, v, and w components of wind speed, and averaging the spectra calculated for each column. Subsequently, the processor may compare the energy spectrum of the generated turbulent field with the energy spectrum of PALM-based high-resolution urban wind field data to evaluate whether the generated turbulent field maintains the physical characteristics of the reference high-resolution turbulent field.

[0106] Physical consistency evaluation may include analyzing energy distributions in low-wavenumber, mid-wavenumber, and high-wavenumber regions. The low-wavenumber region may be associated with large-scale flow structures, the mid-wavenumber region with intermediate-scale turbulent structures, and the high-wavenumber region with micro-scale turbulent structures and high-frequency fluctuations. The processor can evaluate whether the energy spectrum of the generated turbulent field is preserved by analyzing the extent to which the energy distribution of the turbulent field generated in each wavenumber region matches the energy distribution of PALM-based high-resolution urban wind field data.

[0108] In one embodiment, the processor can calculate a ratio between the turbulent kinetic energy spectrum of the generated super-high resolution micro-scale turbulent field and the turbulent kinetic energy spectrum of PALM-based high-resolution urban wind field data. This ratio may represent the degree of energy conservation per wavenumber domain and ideally may have a value close to 1. By calculating the ratio for low turbulence, medium turbulence, and high turbulence conditions, the processor can evaluate how accurately the super-resolution reconstruction model reproduces the energy distribution according to changes in turbulence intensity.

[0110] In addition, the processor can evaluate physical consistency using at least one of the Kolmogorov-Smirnov statistic, Integrated Quadratic Distance (IQD), Probability Density Function, and turbulent kinetic energy spectrum. The Kolmogorov-Smirnov statistic can represent the maximum difference between the generated turbulent field and the cumulative distribution function of high-resolution urban wind field data, the IQD can represent the overall difference between the two distributions in an integral form, and the probability density function can be used to compare statistical characteristics such as mean, standard deviation, skewness, and kurtosis for each wind speed component.

[0112] In one embodiment, the turbulent field generated by ESRGAN can reproduce micro-turbulent structures similar to high-resolution reference data and, in particular, improve the conservation of the turbulent kinetic energy spectrum in the mid-wavenumber and high-wavenumber regions. In another embodiment, the turbulent field generated by LDM can stably preserve large-scale flow structures and average wind speed distributions. Thus, according to one embodiment of the present invention, physical consistency evaluations can be performed for ultra-high-resolution micro-scale turbulent fields reconstructed using different generative models, according to turbulence intensity conditions and wavenumber regions.

[0114] In one embodiment, physical consistency evaluation can be performed on wind field data at an altitude of approximately 300m associated with the flight altitude of an urban air mobility vehicle. In particular, ultra-high resolution micro-scale turbulence fields can be generated with an inference time of approximately 1.5 seconds per sample, and can be utilized to support near-real-time urban air mobility operations.

[0116] The results of the physical consistency assessment can be utilized as high-resolution wind field information to support urban air mobility operations. For example, the generated ultra-high-resolution micro-scale turbulence fields can be used to assess the operational safety of urban air mobility vehicles, establish flight paths, analyze operational risks, or provide wind environment information for low-altitude urban airspace.

[0118] As such, according to one embodiment of the present invention, an ultra-high-resolution micro-scale turbulent field with a resolution of 5m can be generated from urban wind field data with a resolution of 100m without repeatedly performing high-resolution PALM simulations, which are computationally expensive. In addition, by comparing the turbulent kinetic energy spectra for the generated turbulent field according to turbulence intensity conditions, it is possible to evaluate whether the physical characteristics of the actual turbulent field are maintained, rather than merely visual similarity. Accordingly, high-resolution wind field information necessary for supporting urban air mobility operations can be efficiently provided.

[0120] Meanwhile, the present invention is characterized by providing a computer program stored on a computer-readable recording medium to perform operations implemented by various communication terminals. For example, a computer-readable medium may include magnetic media such as a hard disk, a floppy disk, and a magnetic tape; optical recording media such as a CD-ROM and a DVD; magneto-optical media such as a floptical disk; and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, and flash memory.

[0122] Such a computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. In this case, the program instructions recorded on the computer-readable medium may be those specifically designed and configured to implement the present invention, or they may be those known and available to those skilled in the art of computer software. For example, they may include not only machine code, such as that generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter, etc.

[0124] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0126] The scope of the present invention is defined by the claims set forth below rather than by the detailed description, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0127] S100: A step of generating low-resolution urban wind field data of 100m by applying an area interpolation method to high-resolution urban wind field data of 5m resolution generated using PALM. S200: A step of training ESRGAN based on a Generative Adversarial Network using low-resolution urban wind field data and high-resolution urban wind field data as training data. S300: A step of inputting urban wind field data with a resolution of 100m into a trained Generative Adversarial Network-based ESRGAN or a pre-trained LDM to generate an ultra-high-resolution micro-scale turbulent field with a resolution of 5m through a 20-fold improvement in spatial resolution. S400: A step of calculating turbulence intensity (TI) using u, v, and w components from an ultra-high resolution micro-scale turbulence field; S500: A step of classifying into one of low turbulence conditions (TI<0.1), medium turbulence conditions (0.1≤TI<0.3), and high turbulence conditions (TI≥0.3) based on turbulence intensity. S600: A step of evaluating physical consistency by comparing the turbulent kinetic energy spectrum of the generated ultra-high-resolution micro-scale turbulent field with the turbulent kinetic energy spectrum of PALM-based high-resolution urban wind field data for each of the multiple classified turbulent conditions.

Claims

Claim 1 A method for reconstructing a super-high-resolution micro-scale turbulence field performed on a computer to support Urban Air Mobility (UAM) operations, comprising: a step of generating low-resolution urban wind field data with a horizontal resolution of 100m by applying area interpolation to high-resolution urban wind field data with a horizontal resolution of 5m generated using PALM (Parallelized Large-Eddy Simulation Model); a step of training an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) based on a Generative Adversarial Network using the low-resolution urban wind field data and the high-resolution urban wind field data as training data; a step of inputting the urban wind field data with a horizontal resolution of 100m into the trained ESRGAN based on the Generative Adversarial Network or a pre-trained Latent Diffusion Model (LDM) to generate a super-high-resolution micro-scale turbulence field with a horizontal resolution of 5m through a 20-fold improvement in spatial resolution; and using the u component, v component, and w component from the super-high-resolution micro-scale turbulence field A method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations, characterized by comprising: a step of calculating a turbulence intensity (TI); a step of classifying into one of low turbulence conditions (TI<0.1), medium turbulence conditions (0.1≤TI<0.3), and high turbulence conditions (TI≥0.3) based on the turbulence intensity; and a step of evaluating physical consistency by comparing the turbulent kinetic energy spectrum of the generated ultra-high-resolution micro-scale turbulence field with the turbulent kinetic energy spectrum of PALM-based high-resolution urban wind field data for each of the classified plurality of turbulence conditions. Claim 2 A method for reconstructing an ultra-high-resolution micro-scale turbulence field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations, characterized in that, in claim 1, the input data of the ESRGAN is normalized using the mean value and standard deviation for each of the u, v, and w components of wind speed, and the generated turbulence field is denormalized using the same mean value and standard deviation, and the input data of the LDM is normalized using the minimum and maximum values ​​for each wind speed component, and the generated turbulence field is denormalized using the same minimum and maximum values. Claim 3 A method for reconstructing a super-high-resolution micro-scale turbulent field based on a generative adversarial network and a latent diffusion model for supporting urban air mobility operations, wherein, in the step of evaluating physical consistency, the energy distribution of the turbulent kinetic energy spectrum in the low-wavenumber region, the mid-wavenumber region, and the high-wavenumber region is analyzed to evaluate whether the energy spectrum of the generated turbulent field is preserved.

Images