Urban multi-scale wind environment numerical simulation method

Abstract

The invention discloses an urban multi-scale wind environment numerical simulation method, which comprises the following steps of: obtaining an urban multi-source data set, and performing data preprocessing on the urban multi-source data set to obtain an initial urban multi-source data set; building types, block functions and geographical partitions are identified according to the initial city multi-source data set through modeling software, and an initial multi-scale model is generated based on a preset template; calling meteorological big data to train an initial condition model, and injecting each scale into a wind profile and a temperature and humidity background field to obtain an adaptive simulation process state; and activating a digital twinning optimization engine to carry out iterative correction model re-simulation, and when the comfort level of the urban wind environment and the ecological functionality reach optimal balance, outputting an urban wind environment planning blueprint. Multi-source data are fused in an all-around manner, accurate modeling from microcosmic to macroscopic manner is realized, urban styles and features are carefully restored, and wind environment simulation fits reality; and rapid adjustment is realized in case of abnormity, stable and accurate calculation is guaranteed, and simulation time is greatly saved.

Description

Technical Field

[0001] The invention relates to the field of urban environmental technology, in particular to a method for numerical simulation of urban multi-scale wind environment. Background Art

[0002] With the rapid development of urbanization, many difficult problems have arisen in the urban wind environment, such as local strong winds caused by high-rise buildings interfering with pedestrian safety, and blocked ventilation corridors causing the heat island effect to intensify, which seriously affect residents' living comfort, energy consumption pattern and air quality level. Traditional wind environment simulation methods are mostly trapped in the narrow perspective of a single scale, and cannot take into account the fine airflow details of micro-building communities, the ventilation layout characteristics of meso-blocks, and the interaction between macro-urban topography and large-scale climate systems. At the micro level, the wind environment at pedestrian height is affected by the micro-structure of the building, and traditional methods are difficult to accurately capture; at the meso level, the efficiency of the block ventilation network and the microcirculation of courtyard airflow are difficult to effectively quantify; from a macro perspective, the dynamic influence mechanism of the overall wind field of the city by geographical elements such as mountains and water systems needs to be deeply explored. The existing technology lacks an efficient means to integrate multiple scales and realize linkage simulation, which is difficult to meet the needs of comprehensive urban wind environment optimization. Summary of the invention

[0003] The purpose of the present invention is to solve the above problems and to design a numerical simulation method for urban multi-scale wind environment.

[0004] To achieve the above object, the technical solution of the present invention is that, further, in the above-mentioned urban multi-scale wind environment numerical simulation method, the numerical simulation method comprises the following steps:

[0005] Acquire a city multi-source data set, and perform data preprocessing on the city multi-source data set to obtain an initial city multi-source data set;

[0006] Using modeling software to identify building types, block functions, and geographic divisions based on the initial urban multi-source dataset, and generate an initial multi-scale model based on a preset template;

[0007] The initial condition model is trained by calling meteorological big data, and wind profiles and temperature and humidity background fields of various scales are injected to obtain an adaptive simulation process state;

[0008] Perform simulation according to the state of the adaptive simulation process, use AI to intelligently monitor the calculation stability and rationality of physical quantities, dynamically adjust the grid density, solution accuracy and time step according to the deep learning model, and obtain the initial simulation results;

[0009] The initial simulation results are input into the VR system, and based on the user's physical feedback and the wind environment index threshold, the digital twin optimization engine is activated to iteratively correct the model and re-simulate it. When the urban wind environment comfort and ecological functionality reach an optimal balance, the urban wind environment planning blueprint is output.

[0010] Furthermore, in the above-mentioned urban multi-scale wind environment numerical simulation method, the obtaining of the urban multi-source data set, and performing data preprocessing on the urban multi-source data set to obtain the initial urban multi-source data set include:

[0011] Use drone clusters equipped with optical cameras, lidar, and multispectral scanners to scan the entire target city according to the planned route to obtain image data and point cloud data;

[0012] Extracting spatiotemporal layer data from the GIS geographic information system platform, crawling social network heat data through web crawler technology, integrating the image data and point cloud data, spatiotemporal layer data and social network heat data to obtain a city multi-source data set;

[0013] Using filtering to remove outliers from the city multi-source dataset to obtain a first city multi-source dataset;

[0014] The missing values ​​in the first city multi-source dataset are filled using a spatial interpolation algorithm to obtain an initial city multi-source dataset.

[0015] Furthermore, in the above-mentioned urban multi-scale wind environment numerical simulation method, the modeling software is used to identify building types, block functions and geographical divisions according to the initial urban multi-source data set, and an initial multi-scale model is generated based on a preset template, including:

[0016] Based on the CNN convolutional neural network in the modeling software, the initial urban multi-source data set is identified, and the texture features, geometric shape features, and height features of the building facades are learned to obtain urban feature data;

[0017] The GCN graph convolutional network in the modeling software is used to mine the functional association pattern of the blocks in the multi-source data of the city, divide the commercial area, residential area, cultural and educational area and industrial area, and obtain the building distribution data;

[0018] The K-means cluster analysis algorithm was used to delineate the waterfront ecological area, mountain conservation area, urban core built-up area and urban-rural fringe area to obtain geographical classification data;

[0019] Generate an initial multi-scale model based on urban characteristic data, building distribution data and geographic classification data.

[0020] Furthermore, in the above-mentioned urban multi-scale wind environment numerical simulation method, the initial condition model is trained by calling meteorological big data, and each scale is injected into the wind profile and the temperature and humidity background field to obtain the adaptive simulation process state, including:

[0021] The initial multi-scale model is computationally solved based on the CFD computational fluid dynamics algorithm. At the micro scale, LES large eddy simulation is used to capture the small-scale turbulent vortex structure; at the meso scale, the Reynolds average equation is used to simulate the overall ventilation situation of the block; at the macro scale, the meteorological model solver is used to couple the urban and surrounding atmospheric circulations.

[0022] Furthermore, in the above-mentioned urban multi-scale wind environment numerical simulation method, the simulation is performed according to the state of the adaptive simulation process, the calculation stability and the rationality of the physical quantity are monitored by AI intelligently, and the grid density, solution accuracy and time step are dynamically adjusted according to the deep learning model to obtain the initial simulation results, including:

[0023] Use LSTM long short-term memory network and VAE variational autoencoder to build AI intelligent monitoring module;

[0024] Use LSTM long short-term memory network to track the time series changes of simulated physical quantities in real time and learn the fluctuation patterns of wind speed, temperature and pressure;

[0025] Based on the VAE variational autoencoder, the simulated data is reconstructed and compared to reduce the dimension and monitor whether the data distribution is consistent;

[0026] If airflow is blocked due to uneven grid transition between building clusters, the key channel grid will be automatically encrypted by 0.5-2 times based on the deep learning algorithm;

[0027] If the solution format dissipates too much and affects the accuracy, switch to a low numerical viscosity high-order format; in complex flow areas, shorten the time step by 40%-90%, and allocate resources in combination with the elastic computing results on the cloud.

[0028] Furthermore, in the above-mentioned urban multi-scale wind environment numerical simulation method, the initial simulation results are input into the VR system, and based on the user's somatosensory feedback and the wind environment index threshold, the digital twin optimization engine is activated to iteratively correct the model and re-simulate it. When the urban wind environment comfort and ecological functionality reach an optimal balance, the urban wind environment planning blueprint is output, including:

[0029] The initial simulation results are input into the VR system to evaluate the pedestrian wind comfort at the bottom of the micro-building and the wind load risk of the building at the high-rise level; the ventilation efficiency of the ventilation corridor and the microcirculation of the courtyard wind are perceived at the meso-block level; and the distribution of urban heat islands and the influence of the prevailing wind direction are overlooked at the macro-level;

[0030] Activate the digital twin optimization engine to iteratively modify the model and re-simulate it. Based on the multi-objective genetic algorithm in the optimization engine and combined with the physical mechanism constraints of the wind environment, generate multiple sets of building layout adjustments within minutes.

[0031] The model is revised and simulated repeatedly, each time focusing on key indicators that were not met in the previous stage to strengthen the optimization. This process goes through multiple cycles until the urban wind environment comfort and ecological functionality reach a global optimal balance, and the urban wind environment planning blueprint is output.

[0032] Furthermore, in a system for implementing the above-mentioned urban multi-scale wind environment numerical simulation method, the system includes the following modules:

[0033] A data acquisition module is used to acquire a city multi-source data set, perform data preprocessing on the city multi-source data set, and obtain an initial city multi-source data set;

[0034] A model building module, for using modeling software to identify building types, block functions and geographical divisions according to the initial urban multi-source data set, and generating an initial multi-scale model based on a preset template;

[0035] A numerical simulation module is used to call meteorological big data to train the initial condition model, inject wind profiles and temperature and humidity background fields at various scales, and obtain an adaptive simulation process state;

[0036] A state adjustment module is used to simulate according to the state of the adaptive simulation process, use AI intelligent monitoring to calculate stability and rationality of physical quantities, dynamically adjust the grid density, solution accuracy and time step according to the deep learning model, and obtain the initial simulation results;

[0037] The result output module is used to input the initial simulation results into the VR system, activate the digital twin optimization engine to iteratively correct the model and re-simulate it based on user physical feedback and wind environment index thresholds, and output the urban wind environment planning blueprint when the urban wind environment comfort and ecological functionality reach an optimal balance.

[0038] Furthermore, in a system for implementing the above-mentioned urban multi-scale wind environment numerical simulation method, the system includes the following submodules:

[0039] The scanning submodule is used to use the drone cluster equipped with optical cameras, laser radars, and multispectral scanners to scan the entire target city according to the planned route to obtain image data and point cloud data;

[0040] The integration submodule is used to extract spatiotemporal layer data from the GIS geographic information system platform, capture social network heat data through web crawler technology, integrate the image data and point cloud data, spatiotemporal layer data and social network heat data, and obtain a city multi-source data set;

[0041] A deletion submodule, used for removing outliers from the city multi-source data set by filtering to obtain a first city multi-source data set;

[0042] The submodule is used to fill the missing values ​​in the first city multi-source dataset by using a spatial interpolation algorithm to obtain an initial city multi-source dataset.

[0043] Furthermore, in a system for implementing the above-mentioned urban multi-scale wind environment numerical simulation method, the system includes the following submodules:

[0044] The recognition submodule is used to recognize the initial urban multi-source data set based on the CNN convolutional neural network in the modeling software, learn the texture features, geometric shape features, and height features of the building facades, and obtain urban feature data;

[0045] A mining submodule is used to mine the functional association pattern of blocks in the multi-source data of the city by using the GCN graph convolutional network in the modeling software, divide the commercial area, residential area, cultural and educational area and industrial area, and obtain the building distribution data;

[0046] The classification submodule is used to use the K-means cluster analysis algorithm to delineate the waterfront ecological area, mountain conservation area, urban core built-up area and urban-rural fringe area to obtain geographical classification data;

[0047] The generation submodule is used to generate an initial multi-scale model based on urban feature data, building distribution data and geographic classification data.

[0048] Its beneficial effects lie in that, by acquiring a city multi-source data set, data preprocessing is performed on the city multi-source data set to obtain an initial city multi-source data set; using modeling software to identify building types, block functions and geographical divisions according to the initial city multi-source data set, an initial multi-scale model is generated based on a preset template; calling meteorological big data to train the initial condition model, injecting each scale into the wind profile and temperature and humidity background field, and obtaining an adaptive simulation process state; simulating according to the adaptive simulation process state, using AI intelligent monitoring to calculate stability and physical quantity rationality, dynamically adjusting the grid density, solution accuracy and time step according to the deep learning model, and obtaining an initial simulation result; inputting the initial simulation result into the VR system, activating the digital twin optimization engine to iteratively correct the model and re-simulate based on user somatosensory feedback and wind environment indicator thresholds, and outputting an urban wind environment planning blueprint when the urban wind environment comfort and ecological functionality reach an optimal balance. 1. Comprehensive integration of multi-source data, accurate modeling from micro to macro, detailed restoration of urban landscape, and making wind environment simulation fit the reality; 2. Rapid adjustment in case of abnormalities to ensure stable and accurate calculations, greatly saving simulation time; 3. Greatly improve the accuracy of wind environment simulation, and quickly promote the optimization of urban planning and architectural design, reduce costs, and lay a solid foundation for creating a comfortable, ecological, and energy-saving urban space. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0050] Figure 1 It is a schematic diagram of a first embodiment of a method for numerical simulation of urban multi-scale wind environment in an embodiment of the present invention;

[0051] Figure 2 It is a schematic diagram of a second embodiment of the method for numerical simulation of urban multi-scale wind environment in an embodiment of the present invention;

[0052] Figure 3 Schematic diagram of a third embodiment of the method for numerical simulation of urban multi-scale wind environment in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0054] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0055] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a numerical simulation method for urban multi-scale wind environment includes the following steps:

[0056] Step 101: Acquire a city multi-source data set, perform data preprocessing on the city multi-source data set, and obtain an initial city multi-source data set;

[0057] Specifically, in this embodiment, a drone cluster equipped with optical cameras, laser radars, and multispectral scanners is used to scan the entire target city according to a planned route to obtain image data and point cloud data;

[0058] Extract spatiotemporal layer data from the GIS geographic information system platform, capture social network heat data through web crawler technology, integrate image data and point cloud data, spatiotemporal layer data and social network heat data to obtain a multi-source urban data set;

[0059] Filtering is used to remove outliers from the urban multi-source dataset to obtain the first urban multi-source dataset;

[0060] The spatial interpolation algorithm is used to fill the missing values ​​in the first city multi-source dataset to obtain the initial city multi-source dataset.

[0061] Step 102: using modeling software to identify building types, block functions, and geographic divisions based on the initial urban multi-source data set, and generating an initial multi-scale model based on a preset template;

[0062] Specifically, in this embodiment, based on the CNN convolutional neural network in the modeling software, the initial urban multi-source data set is identified, and the texture features, geometric shape features, and height features of the building facades are learned to obtain urban feature data;

[0063] The GCN graph convolutional network in the modeling software is used to mine the functional association pattern of blocks in the urban multi-source data set, divide the commercial area, residential area, cultural and educational area and industrial area, and obtain the building distribution data;

[0064] The K-means cluster analysis algorithm was used to delineate the waterfront ecological area, mountain conservation area, urban core built-up area and urban-rural fringe area to obtain geographical classification data;

[0065] Generate an initial multi-scale model based on urban characteristic data, building distribution data and geographic classification data.

[0066] Step 103: Call the meteorological big data to train the initial condition model, inject each scale into the wind profile and the temperature and humidity background field, and obtain the adaptive simulation process state;

[0067] Specifically, in this embodiment, the initial multi-scale model is calculated and solved based on the CFD computational fluid dynamics algorithm. The LES large eddy simulation is used at the micro scale to capture the small-scale turbulent vortex structure; the Reynolds average equation is used to simulate the overall ventilation situation of the block; and the meteorological model solver is used at the macro scale to couple the city and the surrounding atmospheric circulation.

[0068] Step 104: simulate according to the state of the adaptive simulation process, use AI to intelligently monitor the calculation stability and rationality of physical quantities, dynamically adjust the grid density, solution accuracy and time step according to the deep learning model, and obtain the initial simulation results;

[0069] Specifically, in this embodiment, an AI intelligent monitoring module is established by using an LSTM long short-term memory network and a VAE variational autoencoder;

[0070] Use LSTM long short-term memory network to track the time series changes of simulated physical quantities in real time and learn the fluctuation patterns of wind speed, temperature and pressure;

[0071] Based on the VAE variational autoencoder, the simulated data is reconstructed and compared to reduce the dimension and monitor whether the data distribution is consistent;

[0072] If airflow is blocked due to uneven grid transition between building clusters, the key channel grid will be automatically encrypted by 0.5-2 times based on the deep learning algorithm;

[0073] If the solution format dissipates too much and affects the accuracy, switch to a low numerical viscosity high-order format; in complex flow areas, shorten the time step by 40%-90%, and allocate resources in combination with the elastic computing results on the cloud.

[0074] Step 105: Input the initial simulation results into the VR system. Based on the user's sensory feedback and the wind environment index threshold, activate the digital twin optimization engine to iteratively correct the model and re-simulate it. When the urban wind environment comfort and ecological functionality reach an optimal balance, output the urban wind environment planning blueprint.

[0075] Specifically, in this embodiment, the initial simulation results are input into the VR system to evaluate the pedestrian wind comfort at the bottom of the micro-building and the wind load risk of the building at the high-rise level; the ventilation efficiency of the ventilation corridor and the microcirculation of the courtyard wind are perceived at the meso-block level; the distribution of urban heat islands and the influence of the prevailing wind direction are overlooked at the macro-level;

[0076] Activate the digital twin optimization engine to iteratively modify the model and re-simulate it. Based on the multi-objective genetic algorithm in the optimization engine and combined with the physical mechanism constraints of the wind environment, generate multiple sets of building layout adjustments within minutes.

[0077] The model is revised and simulated repeatedly, each time focusing on key indicators that were not met in the previous stage to strengthen the optimization. This process goes through multiple cycles until the urban wind environment comfort and ecological functionality reach a global optimal balance, and the urban wind environment planning blueprint is output.

[0078] Its beneficial effect lies in that, by acquiring the urban multi-source data set, the urban multi-source data set is preprocessed to obtain the initial urban multi-source data set; the modeling software is used to identify the building type, block function and geographical division according to the initial urban multi-source data set, and the initial multi-scale model is generated based on the preset template; the meteorological big data is called to train the initial condition model, and the wind profile and temperature and humidity background field of each scale are injected to obtain the adaptive simulation process state; the simulation is performed according to the adaptive simulation process state, and the calculation stability and physical quantity rationality are monitored by AI intelligently, and the grid density, solution accuracy and time step are dynamically adjusted according to the deep learning model to obtain the initial simulation results; the initial simulation results are input into the VR system, and based on the user's physical feedback and the wind environment index threshold, the digital twin optimization engine is activated to iteratively correct the model and re-simulate, and when the urban wind environment comfort and ecological functionality reach the optimal balance, the urban wind environment planning blueprint is output. 1. Comprehensive integration of multi-source data, accurate modeling from micro to macro, detailed restoration of urban landscape, and making wind environment simulation fit the reality; 2. Rapid adjustment in case of abnormalities to ensure stable and accurate calculations, greatly saving simulation time; 3. Greatly improve the accuracy of wind environment simulation, and quickly promote the optimization of urban planning and architectural design, reduce costs, and lay a solid foundation for creating a comfortable, ecological, and energy-saving urban space.

[0079] See also Figure 2 In the urban multi-scale wind environment numerical simulation method, obtaining an urban multi-source data set, performing data preprocessing on the urban multi-source data set, and obtaining an initial urban multi-source data set include the following steps:

[0080] Step 201: Use a drone cluster equipped with optical cameras, laser radars, and multispectral scanners to scan the entire target city according to a planned route to obtain image data and point cloud data;

[0081] Step 202: extracting spatiotemporal layer data from the GIS geographic information system platform, crawling social network heat data through web crawler technology, integrating image data and point cloud data, spatiotemporal layer data and social network heat data, and obtaining a city multi-source data set;

[0082] Step 203, using filtering to remove outliers from the city multi-source dataset to obtain a first city multi-source dataset;

[0083] Step 204: fill in the missing values ​​in the first city multi-source dataset using a spatial interpolation algorithm to obtain an initial city multi-source dataset.

[0084] See also Figure 3 ,In the urban multi-scale wind environment numerical simulation method, modeling software is used to ,identify building types, block functions and geographical divisions according to the initial urban ,multi-source data set, and generate the initial multi-scale model based on ,preset templates, including the following steps:

[0085] Step 301: Based on the CNN convolutional neural network in the modeling software, the initial urban multi-source data set is identified, and the texture features, geometric shape features, and height features of the building facades are learned to obtain urban feature data;

[0086] Step 302: Use the GCN graph convolutional network in the modeling software to mine the functional association pattern of blocks in the urban multi-source data set, divide the commercial area, residential area, cultural and educational area, and industrial area, and obtain the building distribution data;

[0087] Step 303: Use K-means cluster analysis algorithm to delineate waterfront ecological areas, mountain conservation areas, urban core built-up areas, and urban-rural fringe areas to obtain geographic classification data;

[0088] Step 304: Generate an initial multi-scale model based on the urban feature data, building distribution data and geographic classification data.

[0089] Specifically, this embodiment also includes

[0090] 1. Data acquisition and preprocessing stage

[0091] Urban multi-source data collection strategy

[0092] Using drone clusters equipped with optical cameras, lidar, multi-spectral scanners and other advanced equipment, the entire target city is scanned in a carpet-like manner according to the planned route. The route planning fully considers the topographical features of the city, and a regular grid route is used in the plain area to ensure uniform data coverage; in mountainous and hilly areas, the route is planned in layers according to contour lines, focusing on capturing the details of buildings and vegetation at different altitudes. The flight altitude is dynamically adjusted from 50 meters to 500 meters to obtain ultra-high-resolution images and point cloud data. The image resolution is better than 0.1 meters / pixel, and the point cloud density can reach thousands of points per square meter.

[0093] Obtain an authoritative Building Information Model (BIM) database from the city planning department, covering detailed information such as building structure, material, and construction year; extract spatiotemporal layers from the Geographic Information System (GIS) platform, including terrain elevation accurate to centimeters, land use change data over many years, and dynamic distribution and flow vector information of water systems; use web crawler technology to legally capture social network popularity data, locate densely populated areas and time periods, and integrate to form an initial urban multi-source data set with a data volume of several terabytes, which is stored in a high-performance distributed file system to ensure efficient data reading and writing.

[0094] Refined data preprocessing process

[0095] For the collected image data, we carry out image enhancement and denoising based on deep learning. We use the Generative Adversarial Network (GAN) model to learn the distribution of clear image features, repair blur and noise areas caused by poor lighting and equipment shaking, and improve image clarity and recognition. We use the semantic segmentation model to identify different objects in the image, such as the sky, buildings, vegetation, roads, etc., with an accuracy rate of over 95%, which is convenient for subsequent targeted processing.

[0096] For the lidar point cloud, statistical filtering is first performed to remove outliers. Based on the statistical characteristics of the local density of the point cloud, the threshold is set to 3 times the standard deviation of the distance mean to screen out isolated abnormal points. Then, downsampling is performed through voxel filtering to balance the data volume and feature retention. The voxel side length is dynamically adjusted between 0.2 and 1 meters depending on the complexity of the area, and the side length is appropriately increased in flat areas to reduce redundancy.

[0097] The coordinates of BIM, GIS, and social network data are uniformly converted, and precise alignment and fusion are achieved based on the geodetic coordinate system (such as WGS84). The spatial interpolation algorithm is used to fill the data gaps. For example, for locally missing terrain elevation data, the Kriging interpolation method is used to generate a continuous surface based on the surrounding known elevation points. Ultimately, a complete, clean, and high-precision initial urban multi-source dataset is obtained, laying a solid foundation for subsequent modeling.

[0098] 2. Intelligent Modeling and Environment Initialization Phase

[0099] Intelligent recognition mechanism of modeling software

[0100] The self-developed modeling software with powerful computing support is deployed in the cloud, integrating the cutting-edge deep learning algorithm framework. After importing the pre-processed data set with one click, the software immediately starts the building type recognition module based on the convolutional neural network (CNN). By learning the multi-dimensional features of the building facade texture, geometric shape, height, etc., it can accurately distinguish between residential buildings (subdivided into high-rise towers, multi-story slab buildings, etc.), commercial complexes, industrial plants, public cultural buildings, etc., with a recognition recall rate of over 90%.

[0101] In order to determine the functions of blocks, we combined the road network density, building type distribution, public facilities layout and social network popularity data, and used the graph convolutional network (GCN) to explore the functional association patterns of blocks, and accurately divide them into commercial areas (locating core business districts based on store density and traffic flow), residential areas (identified by the proportion of residential buildings and the completeness of supporting facilities), cultural and educational areas (defined around cultural facilities such as schools and libraries), industrial areas, and other different functional areas to help with refined modeling.

[0102] Based on GIS terrain data and land use types, with the help of cluster analysis algorithms, geographical divisions are quickly delineated, such as clearly dividing key areas such as waterfront ecological areas, mountain conservation areas, urban core built-up areas, and urban-rural fringe areas, in preparation for modeling adapted to different geographical characteristics.

[0103] Efficient model generation and initialization operations

[0104] Based on the precise identification results, the software automatically matches the preset template library generated by the built-in global massive case optimization. The templates cover successful wind environment planning examples under different climate zones, city sizes, and geographical features, such as coastal typhoon-prone cities and inland drought and sandstorm city templates. Combined with intelligent optimization rules, the micro-scale building surface grid is encrypted to the millimeter level according to complex decorations and opening features to ensure that the details of the building wind load are captured; the meso-blocks are refined to 0.3-1 meters in key areas such as ventilation corridors and street corner squares according to functional importance and ventilation needs; the macro-grid is coarsened from mountainous areas to plains according to the terrain gradient, with a span of 10-100 meters, and the initial multi-scale model with tens of millions of grids is generated very quickly.

[0105] The initial condition model trained by deep learning of meteorological big data is called. The model integrates decades of historical meteorological data of the target city, real-time meteorological monitoring data and climate model data of the surrounding area. The wind profile is accurately generated according to the season, time and geographical location. For example, considering the influence of urban heat island circulation, the wind speed profile of the low-altitude jet stream over the city center in summer nights shows a special change of power law exponent; the temperature and humidity background field is accurately set, and the temperature of the center of the heat island is reasonably increased by 2-6℃ according to the surrounding environment and building density. The relative humidity is dynamically adjusted according to vegetation coverage and water distribution. The initialization is completed and the adaptive simulation process is successfully started. The boundary conditions of each scale model are seamlessly connected.

[0106] 3. Simulation operation and intelligent monitoring stage

[0107] Adaptive simulation process advancement

[0108] After the simulation is started, the multi-scale model is solved in parallel based on the core algorithm of computational fluid dynamics (CFD). At the micro scale, the large eddy simulation (LES) is used to capture the small-scale turbulent vortex structure; at the meso scale, the Reynolds-averaged Navigator-Stokes (RANS) equations are combined to efficiently simulate the overall ventilation situation of the block; at the macro scale, the meteorological model solver is used to couple the urban and surrounding atmospheric circulations. The solvers at each scale dynamically and adaptively switch the calculation parameters according to the flow field characteristics to ensure the balance between simulation stability and efficiency. The simulation process displays key information such as calculation progress and resource utilization in real time.

[0109] AI intelligent monitoring for all-round protection

[0110] The AI ​​intelligent monitoring module embedded in the simulation core is constructed by the long short-term memory network (LSTM) and the variational autoencoder (VAE). LSTM tracks the changes in the time series of key simulated physical quantities in real time, learns the fluctuation patterns of wind speed, temperature, pressure, etc., and predicts abnormal trends in advance; VAE reduces the dimension of the simulated data, reconstructs and compares it, and monitors the consistency of data distribution. Once the overall data reconstruction error exceeds the 10% threshold or the physical quantity in the key area changes suddenly, an early warning is triggered immediately.

[0111] When an abnormality occurs, intelligent monitoring will trace back detailed data of the previous 50-200 time steps and conduct in-depth analysis of the cause. If the airflow is blocked due to uneven grid transition between building clusters, the key channel grid will be automatically encrypted by 0.5-2 times according to the deep learning recommendation algorithm; if the solution format dissipation is too large and affects the accuracy, it will be seamlessly switched to a low-value viscosity high-order format; the time step in complex flow areas will be shortened by 40%-90%, combined with the cloud elastic computing resource allocation (such as dynamically expanding container instances based on Kubernetes) to ensure that the simulation proceeds steadily and obtains reliable initial simulation results. The result data is stored in blockchain encrypted storage to ensure integrity and traceability.

[0112] 4. VR Immersive Optimization Iteration Phase

[0113] VR immersive interactive experience construction

[0114] We carefully built an immersive VR system, equipped with a high refresh rate (≥90Hz), high-resolution (monocular 4K) head-mounted display device, a high-precision spatial positioning tracking system with error control at the millimeter level, and a tactile feedback handle to enable users to experience realistic roaming in the virtual urban wind environment. We developed an efficient data transmission plug-in to map the key data of the simulation results (wind vector, wind pressure, temperature color field, etc.) to the VR scene in real time every second. The wind field visualization accuracy reaches 0.05 m / s resolution, and the dynamics of the wind environment are delicately presented.

[0115] Iterative optimization of closed-loop operation

[0116] Design team members wore VR equipment to immerse themselves in the wind field, assessing pedestrian wind comfort at the bottom of the micro-building, and inspecting the wind load risk of the building at the high level; sensing the ventilation efficiency of the ventilation corridor and the microcirculation of the courtyard wind at the meso-block level; and overlooking the distribution of urban heat islands and the influence of the prevailing wind direction at the macro level. Based on the user's body perception score (0-10 points for comfort evaluation) and strict wind environment index thresholds (pedestrian safety wind speed 5-7 meters / second, winter windproof and warm temperature range, etc.), the digital twin optimization engine is activated.

[0117] The optimization engine has built-in intelligent optimization strategies such as multi-objective genetic algorithms and particle swarm optimization algorithms. Combined with the constraints of the physical mechanism of the wind environment, it can generate multiple improvement plans within minutes, covering building layout adjustments (such as staggered layouts to optimize ventilation), green landscape optimization (windbreaks, green space layout improvements), and block planning reshaping (street widening, direction optimization). VR is used to intuitively compare the improvement effects of each plan indicator (such as visual display of the percentage increase in ventilation volume and the degree of improvement in thermal comfort), and the optimal solution is selected and fed back to the modeling software.

[0118] The model is revised and re-simulated repeatedly, each time focusing on key indicators that were not met in the early stage for enhanced optimization, and going through multiple cycles until the urban wind environment comfort (comprehensive score exceeds 85 points) and ecological functionality (ventilation corridor connectivity and pollutant dilution efficiency increased by more than 35%) reach a global optimal balance. Finally, a customized urban wind environment planning blueprint is accurately output, providing detailed guidance for the layout of new urban areas and the wind environment improvement project for old district renovation, helping the city move towards a new level of green and livable.

[0119] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for numerical simulation of urban multi-scale wind environment, characterized in that: The numerical simulation method comprises the following steps: Acquire a city multi-source data set, and perform data preprocessing on the city multi-source data set to obtain an initial city multi-source data set; Using modeling software to identify building types, block functions, and geographic divisions based on the initial urban multi-source dataset, and generate an initial multi-scale model based on a preset template; The initial condition model is trained by calling meteorological big data, and wind profiles and temperature and humidity background fields of various scales are injected to obtain an adaptive simulation process state; Perform simulation according to the state of the adaptive simulation process, use AI to intelligently monitor the calculation stability and rationality of physical quantities, dynamically adjust the grid density, solution accuracy and time step according to the deep learning model, and obtain the initial simulation results; The initial simulation results are input into the VR system, and based on the user's physical feedback and the wind environment index threshold, the digital twin optimization engine is activated to iteratively correct the model and re-simulate it. When the urban wind environment comfort and ecological functionality reach an optimal balance, the urban wind environment planning blueprint is output.

2. A method for numerical simulation of urban multi-scale wind environment according to claim 1, characterized in that: The step of obtaining a city multi-source data set and performing data preprocessing on the city multi-source data set to obtain an initial city multi-source data set includes: Use drone clusters equipped with optical cameras, lidar, and multispectral scanners to scan the entire target city according to the planned route to obtain image data and point cloud data; Extracting spatiotemporal layer data from the GIS geographic information system platform, crawling social network heat data through web crawler technology, integrating the image data and point cloud data, spatiotemporal layer data and social network heat data to obtain a city multi-source data set; Using filtering to remove outliers from the city multi-source dataset to obtain a first city multi-source dataset; The missing values ​​in the first city multi-source dataset are filled using a spatial interpolation algorithm to obtain an initial city multi-source dataset.

3. The method for numerical simulation of urban multi-scale wind environment according to claim 1, characterized in that: The method of using the modeling software to identify building types, block functions and geographical divisions according to the initial urban multi-source data set and generating an initial multi-scale model based on a preset template includes: Based on the CNN convolutional neural network in the modeling software, the initial urban multi-source data set is identified, and the texture features, geometric shape features, and height features of the building facades are learned to obtain urban feature data; The GCN graph convolutional network in the modeling software is used to mine the functional association pattern of the blocks in the multi-source data of the city, divide the commercial area, residential area, cultural and educational area and industrial area, and obtain the building distribution data; The K-means cluster analysis algorithm was used to delineate the waterfront ecological area, mountain conservation area, urban core built-up area and urban-rural fringe area to obtain geographical classification data; Generate an initial multi-scale model based on urban characteristic data, building distribution data and geographic classification data.

4. A method for numerical simulation of urban multi-scale wind environment according to claim 1, characterized in that: The calling of meteorological big data to train the initial condition model injects wind profiles and temperature and humidity background fields at various scales to obtain an adaptive simulation process state, including: The initial multi-scale model is computationally solved based on the CFD computational fluid dynamics algorithm. At the micro scale, LES large eddy simulation is used to capture the small-scale turbulent vortex structure; at the meso scale, the Reynolds average equation is used to simulate the overall ventilation situation of the block; at the macro scale, the meteorological model solver is used to couple the urban and surrounding atmospheric circulations.

5. The method for numerical simulation of urban multi-scale wind environment according to claim 1, characterized in that: The simulation is performed according to the state of the adaptive simulation process, the calculation stability and the rationality of the physical quantity are monitored by AI, and the grid density, solution accuracy and time step are dynamically adjusted according to the deep learning model to obtain the initial simulation results, including: Use LSTM long short-term memory network and VAE variational autoencoder to build AI intelligent monitoring module; Use LSTM long short-term memory network to track the time series changes of simulated physical quantities in real time and learn the fluctuation patterns of wind speed, temperature and pressure; Based on the VAE variational autoencoder, the simulated data is reconstructed and compared to reduce the dimension and monitor whether the data distribution is consistent; If airflow is blocked due to uneven grid transition between building clusters, the key channel grid will be automatically encrypted by 0.5-2 times based on the deep learning algorithm; If the solution format dissipates too much and affects the accuracy, switch to a low numerical viscosity high-order format; in complex flow areas, shorten the time step by 40%-90%, and allocate resources in combination with the elastic computing results on the cloud.

6. A method for numerical simulation of urban multi-scale wind environment according to claim 1, characterized in that: The initial simulation results are input into the VR system, and based on the user's somatosensory feedback and the wind environment index threshold, the digital twin optimization engine is activated to iteratively correct the model and re-simulate it. When the urban wind environment comfort and ecological functionality reach an optimal balance, the urban wind environment planning blueprint is output, including: The initial simulation results are input into the VR system to evaluate the pedestrian wind comfort at the bottom of the micro-building and the wind load risk of the building at the high-rise level; the ventilation efficiency of the ventilation corridor and the microcirculation of the courtyard wind are perceived at the meso-block level; and the distribution of urban heat islands and the influence of the prevailing wind direction are overlooked at the macro-level; Activate the digital twin optimization engine to iteratively modify the model and re-simulate it. Based on the multi-objective genetic algorithm in the optimization engine and combined with the physical mechanism constraints of the wind environment, generate multiple sets of building layout adjustments within minutes. The model is revised and simulated repeatedly, each time focusing on key indicators that have not been met in the early stage to strengthen the optimization. This process goes through multiple cycles until the urban wind environment comfort and ecological functionality reach a global optimal balance, and the urban wind environment planning blueprint is output.

7. A system for implementing a method for numerically simulating urban multi-scale wind environment as claimed in claim 1, characterized in that: The system includes the following modules: A data acquisition module is used to acquire a city multi-source data set, perform data preprocessing on the city multi-source data set, and obtain an initial city multi-source data set; A model building module, for using modeling software to identify building types, block functions and geographical divisions according to the initial urban multi-source data set, and generating an initial multi-scale model based on a preset template; A numerical simulation module is used to call meteorological big data to train the initial condition model, inject wind profiles and temperature and humidity background fields at various scales, and obtain an adaptive simulation process state; A state adjustment module is used to simulate according to the state of the adaptive simulation process, use AI intelligent monitoring to calculate stability and rationality of physical quantities, dynamically adjust the grid density, solution accuracy and time step according to the deep learning model, and obtain the initial simulation results; The result output module is used to input the initial simulation results into the VR system, activate the digital twin optimization engine to iteratively correct the model and re-simulate it based on user physical feedback and wind environment index thresholds, and output the urban wind environment planning blueprint when the urban wind environment comfort and ecological functionality reach an optimal balance.

8. A system for implementing a method for numerically simulating a multi-scale urban wind environment as claimed in claim 1, characterized in that: The system includes the following submodules: The scanning submodule is used to use the drone cluster equipped with optical cameras, laser radars, and multispectral scanners to scan the entire target city according to the planned route to obtain image data and point cloud data; The integration submodule is used to extract spatiotemporal layer data from the GIS geographic information system platform, capture social network heat data through web crawler technology, integrate the image data and point cloud data, spatiotemporal layer data and social network heat data, and obtain a city multi-source data set; A deletion submodule, used for removing outliers from the city multi-source data set by filtering to obtain a first city multi-source data set; The submodule is used to fill the missing values ​​in the first city multi-source dataset by using a spatial interpolation algorithm to obtain an initial city multi-source dataset.

9. A system for implementing a method for numerically simulating urban multi-scale wind environment as claimed in claim 1, characterized in that: The system includes the following submodules: The recognition submodule is used to recognize the initial urban multi-source data set based on the CNN convolutional neural network in the modeling software, learn the texture features, geometric shape features, and height features of the building facades, and obtain urban feature data; A mining submodule is used to mine the functional association pattern of blocks in the multi-source data of the city by using the GCN graph convolutional network in the modeling software, divide the commercial area, residential area, cultural and educational area and industrial area, and obtain the building distribution data; The classification submodule is used to use the K-means cluster analysis algorithm to delineate the waterfront ecological area, mountain conservation area, urban core built-up area and urban-rural fringe area to obtain geographical classification data; The generation submodule is used to generate an initial multi-scale model based on urban feature data, building distribution data and geographic classification data.

Images