Intelligent site selection method and system for unmanned aerial vehicle take-off and landing sites in urban building group

Abstract

The invention relates to the technical field of unmanned aerial vehicle application and urban airspace management, in particular to an intelligent site selection method and system for unmanned aerial vehicle take-off and landing points in an urban building group, and the intelligent site selection system for the unmanned aerial vehicle take-off and landing points in the urban building group specifically comprises the following steps: S1, carrying out the multi-source heterogeneous data fusion and high-precision three-dimensional environment modeling; s2, fluid dynamics simulation and airflow field pre-calculation are calculated; s3, real-time meteorological monitoring and dynamic risk correction; s4, establishing a quantitative safety evaluation model and multi-objective optimization site selection; and S5, visual decision support and feedback learning. According to the scheme, by deeply fusing multi-source heterogeneous geographic information data and constructing a high-precision three-dimensional digital twinborn city environment, accurate simulation and risk assessment of an airflow field around a building group are realized, geometric, material and semantic information is automatically extracted by using oblique photography, laser point cloud and a building information model, and the construction efficiency is improved. A unified three-dimensional grid with physical attributes is formed, and a reliable basis is provided for subsequent fluid calculation.

Description

Technical Field

[0001] This invention relates to the fields of unmanned aerial vehicle (UAV) applications and urban airspace management technology, and in particular to a method and system for intelligent site selection of UAV take-off and landing points in urban building complexes. Background Technology

[0002] In recent years, drones have been widely used and developed in many fields. In urban planning and construction, drones can be used for topographic mapping and building monitoring. By carrying high-precision mapping equipment, drones can quickly and accurately acquire data on urban topography and buildings, providing basic information for urban planning and architectural design. In addition, drones can also be used to monitor construction, ensuring project quality and progress. In the logistics field, drones are used for express delivery. With the continuous advancement of technology, drones have been able to achieve automatic flight and precise landing, which gives them a great advantage in express delivery. Drone delivery can not only shorten delivery time, but also reduce traffic congestion and environmental pollution.

[0003] With the large-scale application of drones in urban logistics, emergency inspection, and urban surveying, the safety and efficiency of take-off and landing points have become the core bottlenecks restricting their operation. Existing technologies have the following significant shortcomings:

[0004] 1. Lack of three-dimensional environment: For example, patent CN110298189A only relies on GIS two-dimensional planar data and does not consider the three-dimensional features such as building height and shape on the flow around and turbulence effect of airflow, resulting in a significant deviation between the site selection result and the actual airflow environment;

[0005] 2. Insufficient dynamic adaptability: Traditional methods are mostly based on static assessments of historical meteorological data, which cannot respond to microclimate changes such as wind speed and wind direction in real time, and their safety is difficult to guarantee in scenarios of gusts and sudden airflows;

[0006] 3. Lack of safety quantification: Quantitative indicators for key parameters such as turbulence intensity and airflow stability have not been established. Safety is judged only by simple rules such as "distance from obstacle threshold", which cannot meet the requirements of high-precision take-off and landing. Summary of the Invention

[0007] The purpose of this invention is to address the aforementioned shortcomings in the existing technology by proposing a method that integrates CFD three-dimensional airflow simulation, real-time data-driven machine learning evaluation, and adaptive optimization algorithms to solve the problem of synergistic optimization of UAV take-off and landing safety and efficiency in complex airflow environments of urban building clusters.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent site selection system for UAV take-off and landing points in urban building clusters, wherein the intelligent site selection system for UAV take-off and landing points in urban building clusters is specifically as follows:

[0009] S1. Multi-source heterogeneous data fusion and high-precision 3D environment modeling are used to construct a 3D digital twin model of the city containing accurate geometry, surface material and semantic information in order to accurately calculate the airflow field;

[0010] S1.1 Data Acquisition: Oblique photogrammetry real-world 3D model, airborne or vehicle-mounted LiDAR point cloud data, building information model or urban planning model;

[0011] S1.2 Data processing and fusion;

[0012] S2. Computational fluid dynamics simulation and airflow field pre-calculation: Based on a three-dimensional model, simulate the wind field environment around the building complex under different prevailing winds and predict potential airflow turbulence areas.

[0013] S2.1 Boundary condition settings;

[0014] S2.2, CFD simulation calculation;

[0015] S2.3, Generate an airflow risk map;

[0016] S3, real-time meteorological monitoring and dynamic risk correction, overcome the limitations of static simulation, enabling the system to respond to microclimate changes in real time;

[0017] S3.1 Deploy an IoT weather sensor network;

[0018] S3.2 Data assimilation and dynamic correction;

[0019] S4. Establish a quantitative security assessment model and multi-objective optimized site selection, replace simple distance rules with precise quantitative indicators, and make decisions by comprehensively considering efficiency factors;

[0020] S4.1 Define and quantify safety indicators;

[0021] S4.2 Multi-objective optimization algorithm;

[0022] S5 provides a visual decision support and feedback learning interface for human-computer interaction and enables the system to continuously optimize itself.

[0023] S5.1, Visualization Platform;

[0024] S5.2 Feedback learning mechanism.

[0025] Preferably, the statement according to S1 is as follows:

[0026] Oblique photogrammetry-based 3D model: acquired through drone aerial photography, providing precise data on the building's external geometry, height, roof structure, etc.

[0027] Airborne or vehicle-mounted LiDAR point cloud data: supplement the acquisition of high-precision terrain, vegetation and building facade data, overcoming the shortcomings of photogrammetry in areas such as glass curtain walls;

[0028] Building Information Model or Urban Planning Model: Obtain the official design dimensions, material properties, and semantic information of the building, where the material properties are roughness and the semantic information is purpose or height;

[0029] Based on the above data, registration and fusion are performed in a unified coordinate system to generate a high-precision three-dimensional voxel mesh or surface mesh model rich in semantic information. The model includes not only shape, but also material and type attributes, providing a foundation for subsequent fluid dynamics calculations.

[0030] Preferably, the following is stated in S2:

[0031] Boundary condition settings: Input the typical wind direction and wind speed provided by the local weather station as the entry conditions for the CFD simulation, and set different surface roughness parameters according to the building surface material;

[0032] For the formula of numerical wind speed profile:

[0033]

[0034] Where u(z) is the wind speed at height z in m / s, which is the wind speed that needs to be set for the CFD inlet boundary conditions;

[0035] u * The friction velocity (in m / s) is a scale velocity related to the surface shear stress. It can be calculated from the wind speed at a known reference height.

[0036] R is the Karman constant, R≈0.4;

[0037] z represents the height above the ground;

[0038] z0 is the surface roughness length, which is set according to the building surface material;

[0039] CFD simulation calculation: using high-performance computing clusters or cloud computing resources, steady-state CFD simulation is performed on large-scale building complex areas. The simulation output results include key fluid parameters such as wind speed vector field, wind pressure distribution, turbulent kinetic energy, and eddy viscosity. These parameters directly quantify the instability and turbulence of the airflow.

[0040] Turbulent kinetic energy transport equation k, steady RANS model:

[0041]

[0042] in: Turbulent kinetic energy, unit: m 2 / s2 , representing turbulence intensity;

[0043] ρ is the fluid density;

[0044] u is the time-averaged velocity vector;

[0045] μ is the dynamic viscosity, μ t Turbulent viscosity;

[0046] σ k The turbulent Prandtl number represents the turbulent kinetic energy.

[0047] For turbulence generation;

[0048] ∈ represents the turbulent dissipation rate, which indicates the rate at which kinetic energy is dissipated into heat energy;

[0049] Generate airflow risk maps: Visualize and overlay CFD simulation results, especially turbulent kinetic energy and wind speed gradients, to generate static airflow risk probability maps under different wind directions;

[0050] Wind speed includes the prevailing wind direction and maximum wind speed in all four seasons; building surface materials include glass, concrete, and vegetation.

[0051] The static airflow risk probability map also includes:

[0052] Strong turbulence areas: leeward side of buildings, narrow streets, and areas at building angles;

[0053] Areas with strong descending or ascending air currents: areas near the windward and leeward sides of tall buildings;

[0054] Stable airflow areas: open squares, parks, and rooftops of low-rise building complexes.

[0055] Preferably, the following is stated in S3:

[0056] Deploy an IoT-based weather sensor network: Deploy low-cost micro weather stations in key locations such as around candidate take-off and landing points and on the rooftops of high-rise buildings to monitor wind speed, wind direction, temperature, and air pressure in real time;

[0057] Data assimilation and dynamic correction: Real-time monitoring data is used as the real boundary conditions and assimilated with the pre-computed CFD model. A lightweight CFD fast correction algorithm or a machine learning-based surrogate model is used to quickly deduce the subtle changes in the airflow field of the entire area at the current moment based on real-time wind direction and speed. The airflow risk map is dynamically updated. In the event of sudden gusts or extreme weather, the system can trigger an alarm and even temporarily close high-risk take-off and landing points.

[0058] Preferably, the statement according to S4 is as follows:

[0059] Define quantitative safety indicators: A turbulence intensity value exceeding 15% is considered to pose a high risk to drone take-off and landing. The ratio of wind speed at a certain point to the incoming wind speed is also considered; a ratio that is too large or too small indicates an abnormal wind field. The stability of the wind direction at that point in both time and space is assessed. Based on the above indicators, a safety score between 0 and 1 is calculated.

[0060] Multi-objective optimization algorithm: Input: set of candidate take-off and landing points, flat areas extracted from the model in step 1, safety score for each point, business requirements: distance to delivery point, ground pedestrian density, airspace restrictions, infrastructure costs, etc.

[0061] Optimization: Employ multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization, to maximize safety scores and operational efficiency while minimizing costs and risks, and perform optimization calculations.

[0062] Output: A set of Pareto optimal takeoff and landing site selection schemes, which operators can use to make a final decision based on their current preferences, whether they prioritize safety or efficiency.

[0063] Preferably, the following is stated in S5:

[0064] Visualization platform: In the 3D digital twin city model, the airflow risk map, real-time sensor data, candidate point safety scores and optimization results are displayed intuitively, and managers can see the airflow conditions and risk levels of different locations at a glance;

[0065] Feedback learning mechanism: The system records the actual take-off and landing data of each drone, compares this real data with the system's predicted risk value, and continuously corrects and optimizes the CFD model parameters and safety assessment model through machine learning algorithms, making the predictions more and more accurate.

[0066] Preferably, the further optimization scheme based on the content in S1 is as follows:

[0067] A1. Intelligent fusion and semantic enhancement of multi-source data: Using AI technology, it automatically fuses geometric, material and semantic information from oblique photogrammetry models, LiDAR point clouds and BIM / CIM models. It identifies and fills in missing areas through deep learning, and extracts high-precision semantic labels to form a unified three-dimensional voxel mesh with physical properties and semantic information.

[0068] A2. Dynamic data embedding and real-time update mechanism: Introduces IoT sensor data and real-time remote sensing data, maps dynamic information to three-dimensional grid nodes through spatiotemporal coding, establishes an incremental update algorithm, realizes automatic correction of the model in a short period of time, and ensures that fluid dynamics calculations are based on the latest environmental conditions.

[0069] A3. Multi-scale adaptive mesh generation and optimization: To meet the needs of fluid computing, an adaptive mesh refinement algorithm is developed: Based on surface curvature, semantic importance and fluid sensitivity, the voxel resolution is dynamically adjusted, while the mesh topology is optimized to reduce computational redundancy and retain key attributes, and the output is directly in a CFD-compatible format.

[0070] A4. Digital twin-driven fluid simulation interface integration: Constructs a seamless interface with fluid simulation software, embeds the 3D model into the digital twin platform, realizes parameterized interaction through API and provides real-time feedback of simulation results, supports scene simulation, and improves the model's decision support capabilities in complex urban environments.

[0071] Preferably, the airflow risk map generated in S2 is further optimized into an annual comprehensive airflow risk index map, as follows:

[0072] B1. Data Input and Preprocessing:

[0073] Input: Multiple static airflow risk probability maps generated by S2, covering the prevailing wind direction and maximum wind speed conditions for all four seasons;

[0074] Processing: Assign a weighting coefficient to each wind direction, based on the annual frequency of that wind direction provided by the local weather station. Maximum wind speed can be considered as a special case, or weighted according to its probability of occurrence.

[0075] B2. Weighted superposition calculation:

[0076] Algorithm: Establish a gridded model to divide the map area into small grid units;

[0077] Calculation: For each grid cell, its risk value is read from all static wind direction maps, and then a weighted average is calculated based on the weight of the corresponding wind direction.

[0078] Output: Generate a new grid data layer, namely the annual airflow risk composite index. The higher the index value, the greater the probability and intensity of the location being exposed to high-risk airflow environments such as high turbulence and strong winds throughout the year.

[0079] Generate a comprehensive map: Visualize the calculated annual comprehensive index and use color gradient rendering to generate a brand new comprehensive risk map;

[0080] Develop interactive features: Provide users with an interactive interface. When a user clicks on any point on the map, the interface not only displays the comprehensive risk index of that point, but also traces and lists the top three wind directions that contribute the most to the risk of that point and their contribution.

[0081] Preferably, the further optimization scheme based on the content in S3 is as follows:

[0082] C1. Construct a swarm intelligence-digital twin collaborative computing framework to achieve ultra-high resolution and forward-looking prediction of airflow fields. Based on the existing real-time data assimilation CFD model, introduce a mechanism for deep collaboration between swarm intelligence computing and digital twin.

[0083] C2. Establish a dynamic airspace geofence adaptive management system to maximize risk avoidance and airspace resource utilization. Based on the ultra-high resolution and forward-looking dynamic airflow risk map generated by C1, an innovative dynamic airspace geofence management protocol is developed.

[0084] Preferably, the intelligent site selection method for UAV take-off and landing points in the urban building complex is as follows:

[0085] T1. Start the system, load multi-source data, including oblique photogrammetry model, LiDAR point cloud, BIM / CIM model, run AI fusion algorithm, automatically fill in missing areas and extract semantic information, generate high-precision three-dimensional voxel mesh model, confirm that coordinate system I and data registration are completed, and enter the fluid simulation preparation state.

[0086] T2. Set CFD boundary conditions: Input local meteorological data, start CFD simulation calculation, output parameters such as wind speed, turbulent kinetic energy, and eddy viscosity, generate and visualize static airflow risk probability map, and identify strong turbulence area, downdraft area and steady airflow area.

[0087] T3. Import static risk maps under different wind directions in all four seasons, calculate the annual airflow risk index of each grid cell based on the annual frequency of wind direction, generate an annual comprehensive risk map, and support interactive query of the risk sources of each point;

[0088] T4. Deploy IoT meteorological sensors in key locations, calibrate sensor data to ensure time synchronization and data format consistency with the central system, and activate edge AI prediction function to achieve short-term local microclimate forecasting.

[0089] T5. The system automatically collects sensor data and edge prediction results, runs data assimilation algorithms, dynamically corrects the CFD model, and updates the airflow risk map. If a sudden gust of wind or extreme weather is detected, an early warning mechanism is triggered.

[0090] T6. Input the candidate take-off and landing point locations into the system, run the safety assessment model, calculate the turbulence intensity, wind speed ratio, wind direction stability and other indicators of each point, use a multi-objective optimization algorithm, take into account factors such as safety, efficiency and cost, and output the Pareto optimal location scheme.

[0091] T7. Based on the real-time airflow risk map, dynamically divide the airspace risk units, set an acceptable risk threshold for each drone, generate a dynamic geofence, and automatically adjust the no-fly zone and replan the flight path in case of sudden wind conditions to ensure a safe flight corridor.

[0092] T8. Record the actual airflow data and system prediction values ​​for each takeoff and landing, compare the prediction and actual data through machine learning algorithms, optimize CFD parameters and safety models, and regularly update the three-dimensional model and meteorological weights to maintain the system's prediction accuracy and adaptability.

[0093] The design scheme proposed in this invention has the following beneficial effects during application:

[0094] 1. This solution achieves accurate simulation and risk assessment of the airflow field around building clusters by deeply integrating multi-source heterogeneous geographic information data and constructing a high-precision three-dimensional digital twin urban environment. Using oblique photography, laser point cloud and building information model, it automatically extracts geometric, material and semantic information to form a unified three-dimensional mesh with physical properties, providing a reliable foundation for subsequent fluid calculations. Combined with meteorological data and computational fluid dynamics simulation, the system can accurately identify high-risk areas such as strong turbulence areas and downdraft areas, and generate static airflow risk probability maps under different wind directions, significantly improving the accuracy and spatial integrity of airflow field expression in complex urban environments.

[0095] 2. As described in 1, the introduction of real-time meteorological sensing and dynamic data assimilation mechanism effectively overcomes the limitations of traditional static simulation, enabling rapid response and forward-looking prediction of microclimate changes. By deploying a smart IoT meteorological station network and embedding edge computing and swarm intelligence collaborative capabilities, the system can perform autonomous prediction of airflow at the very short-term microscale and perform bidirectional verification with the central digital twin platform. It also supports minute-level forward warnings for phenomena such as sudden gusts and local eddies, and dynamically updates the airflow risk map, greatly improving the system's adaptability and decision-making timeliness in real meteorological environments.

[0096] 3. As described in section 2, by defining quantitative safety indicators such as turbulence intensity, wind speed ratio, and wind direction stability, and combining these with business requirements such as delivery distance and pedestrian density, a multi-objective optimization algorithm is used to output a Pareto optimal solution set. Simultaneously, the system supports dynamic airspace management and risk unit division, generating safe flight corridors and geofences for each drone based on real-time wind conditions. This maximizes airspace operational efficiency while ensuring safety, and continuously optimizes model accuracy and overall system performance through a continuous learning mechanism. Attached Figure Description

[0097] Figure 1 This is the basic system flowchart of the present invention;

[0098] Figure 2This is a detailed schematic diagram of the multi-source heterogeneous data fusion and high-precision 3D environment modeling of the present invention;

[0099] Figure 3 This is a schematic diagram illustrating the steps of computational fluid dynamics simulation and airflow field pre-calculation in this invention;

[0100] Figure 4 This is a detailed schematic diagram illustrating the real-time meteorological monitoring and dynamic risk correction of the present invention. Detailed Implementation

[0101] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0102] Example 1

[0103] Reference Figures 1-4 The intelligent site selection system for drone take-off and landing points in urban building complexes is as follows:

[0104] S1. Multi-source heterogeneous data fusion and high-precision 3D environment modeling are used to construct a 3D digital twin model of the city containing accurate geometry, surface material and semantic information in order to accurately calculate the airflow field;

[0105] S1.1 Data Acquisition: Oblique photogrammetry real-world 3D model, airborne or vehicle-mounted LiDAR point cloud data, building information model or urban planning model;

[0106] S1.2 Data processing and fusion;

[0107] S2. Computational fluid dynamics simulation and airflow field pre-calculation: Based on a three-dimensional model, simulate the wind field environment around the building complex under different prevailing winds and predict potential airflow turbulence areas.

[0108] S2.1 Boundary condition settings;

[0109] S2.2, CFD simulation calculation;

[0110] S2.3, Generate an airflow risk map;

[0111] S3, real-time meteorological monitoring and dynamic risk correction, overcome the limitations of static simulation, enabling the system to respond to microclimate changes in real time;

[0112] S3.1 Deploy an IoT weather sensor network;

[0113] S3.2 Data assimilation and dynamic correction;

[0114] S4. Establish a quantitative security assessment model and multi-objective optimized site selection, replace simple distance rules with precise quantitative indicators, and make decisions by comprehensively considering efficiency factors;

[0115] S4.1 Define and quantify safety indicators;

[0116] S4.2 Multi-objective optimization algorithm;

[0117] S5 provides a visual decision support and feedback learning interface for human-computer interaction and enables the system to continuously optimize itself.

[0118] S5.1, Visualization Platform;

[0119] S5.2 Feedback learning mechanism.

[0120] According to S1:

[0121] Oblique photogrammetry-based 3D model: acquired through drone aerial photography, providing precise data on the building's external geometry, height, roof structure, etc.

[0122] Airborne or vehicle-mounted LiDAR point cloud data: supplement the acquisition of high-precision terrain, vegetation and building facade data, overcoming the shortcomings of photogrammetry in areas such as glass curtain walls;

[0123] Building Information Model or Urban Planning Model: Obtain the official design dimensions, material properties, and semantic information of the building, where the material properties are roughness and the semantic information is purpose or height;

[0124] Based on the above data, registration and fusion are performed in a unified coordinate system to generate a high-precision three-dimensional voxel mesh or surface mesh model rich in semantic information. The model includes not only shape, but also material and type attributes, providing a foundation for subsequent fluid dynamics calculations.

[0125] According to S2:

[0126] Boundary condition settings: Input the typical wind direction and wind speed provided by the local weather station as the entry conditions for the CFD simulation, and set different surface roughness parameters according to the building surface material;

[0127] For the formula of numerical wind speed profile:

[0128]

[0129] Where u(z) is the wind speed at height z in m / s, which is the wind speed that needs to be set for the CFD inlet boundary conditions;

[0130] u * The friction velocity (in m / s) is a scale velocity related to the surface shear stress. It can be calculated from the wind speed at a known reference height.

[0131] R is the Karman constant, R≈0.4;

[0132] z represents the height above the ground;

[0133] z0 is the surface roughness length, which is set according to the building surface material;

[0134] CFD simulation calculation: using high-performance computing clusters or cloud computing resources, steady-state CFD simulation is performed on large-scale building complex areas. The simulation output results include key fluid parameters such as wind speed vector field, wind pressure distribution, turbulent kinetic energy, and eddy viscosity. These parameters directly quantify the instability and turbulence of the airflow.

[0135] Turbulent kinetic energy transport equation k, steady RANS model:

[0136]

[0137] in: Turbulent kinetic energy, unit: m 2 / s 2 , representing turbulence intensity;

[0138] ρ is the fluid density;

[0139] u is the time-averaged velocity vector;

[0140] μ is the dynamic viscosity, μ t Turbulent viscosity;

[0141] σ k The turbulent Prandtl number represents the turbulent kinetic energy.

[0142] For turbulence generation;

[0143] ∈ represents the turbulent dissipation rate, which indicates the rate at which kinetic energy is dissipated into heat energy;

[0144] Generate airflow risk maps: Visualize and overlay CFD simulation results, especially turbulent kinetic energy and wind speed gradients, to generate static airflow risk probability maps under different wind directions;

[0145] Wind speed includes the prevailing wind direction and maximum wind speed in all four seasons; building surface materials include glass, concrete, and vegetation.

[0146] The static airflow risk probability map also includes:

[0147] Strong turbulence areas: leeward side of buildings, narrow streets, and areas at building angles;

[0148] Areas with strong descending or ascending air currents: areas near the windward and leeward sides of tall buildings;

[0149] Stable airflow areas: open squares, parks, and rooftops of low-rise building complexes.

[0150] According to S3:

[0151] Deploy an IoT-based weather sensor network: Deploy low-cost micro weather stations in key locations such as around candidate take-off and landing points and on the rooftops of high-rise buildings to monitor wind speed, wind direction, temperature, and air pressure in real time;

[0152] Data assimilation and dynamic correction: Real-time monitoring data is used as the real boundary conditions and assimilated with the pre-computed CFD model. A lightweight CFD fast correction algorithm or a machine learning-based surrogate model is used to quickly deduce the subtle changes in the airflow field of the entire area at the current moment based on real-time wind direction and speed. The airflow risk map is dynamically updated. In the event of sudden gusts or extreme weather, the system can trigger an alarm and even temporarily close high-risk take-off and landing points.

[0153] According to S4:

[0154] Define quantitative safety indicators: A turbulence intensity value exceeding 15% is considered to pose a high risk to drone take-off and landing. The ratio of wind speed at a certain point to the incoming wind speed is also considered; a ratio that is too large or too small indicates an abnormal wind field. The stability of the wind direction at that point in both time and space is assessed. Based on the above indicators, a safety score between 0 and 1 is calculated.

[0155] Multi-objective optimization algorithm: Input: set of candidate take-off and landing points, flat areas extracted from the model in step 1, safety score for each point, business requirements: distance to delivery point, ground pedestrian density, airspace restrictions, infrastructure costs, etc.

[0156] Optimization: Employ multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization, to maximize safety scores and operational efficiency while minimizing costs and risks, and perform optimization calculations.

[0157] Output: A set of Pareto optimal takeoff and landing site selection schemes, which operators can use to make a final decision based on their current preferences, whether they prioritize safety or efficiency.

[0158] According to S5:

[0159] Visualization platform: In the 3D digital twin city model, the airflow risk map, real-time sensor data, candidate point safety scores and optimization results are displayed intuitively, and managers can see the airflow conditions and risk levels of different locations at a glance;

[0160] Feedback learning mechanism: The system records the actual take-off and landing data of each drone, compares this real data with the system's predicted risk value, and continuously corrects and optimizes the CFD model parameters and safety assessment model through machine learning algorithms, making the predictions more and more accurate.

[0161] The following is a further optimization plan based on the content in S1:

[0162] A1. Intelligent fusion and semantic enhancement of multi-source data: Using AI technology, it automatically fuses geometric, material and semantic information from oblique photogrammetry models, LiDAR point clouds and BIM / CIM models. It identifies and fills in missing areas through deep learning, and extracts high-precision semantic labels to form a unified three-dimensional voxel mesh with physical properties and semantic information.

[0163] A2. Dynamic data embedding and real-time update mechanism: Introduces IoT sensor data and real-time remote sensing data, maps dynamic information to three-dimensional grid nodes through spatiotemporal coding, establishes an incremental update algorithm, realizes automatic correction of the model in a short period of time, and ensures that fluid dynamics calculations are based on the latest environmental conditions.

[0164] A3. Multi-scale adaptive mesh generation and optimization: To meet the needs of fluid computing, an adaptive mesh refinement algorithm is developed: Based on surface curvature, semantic importance and fluid sensitivity, the voxel resolution is dynamically adjusted, while the mesh topology is optimized to reduce computational redundancy and retain key attributes, and the output is directly in a CFD-compatible format.

[0165] For example:

[0166] The surface curvature is C i Based on geometric features, higher resolution is used to capture geometric details in areas with high surface curvature, such as the leading edge of the wing and near the vortex generator.

[0167] C i =‖k i ‖·A i

[0168] Where: k i The average principal curvature or Gaussian curvature of the surface region where element i is located;

[0169] A i For the surface area represented by element i (for boundary elements) or the volume of influence (for interior elements), this ensures that the curvature effect is weighted by the local area or volume.

[0170] Semantic importance is S i Based on user-defined or physically defined key areas, forced refinement is applied to areas of particular interest to the user, such as airfoil surfaces, combustion chamber inlets, sensor placement points, or known key flow areas.

[0171] S i =I i ·exp(-d i / σ)

[0172] Among them: I i The importance weights for semantic regions are set, such as 1.0 for the airfoil surface and 0.1 for the far-field region;

[0173] d i The distance from unit i to the nearest semantic key;

[0174] σ is the attenuation coefficient, which controls the range of the influence of importance;

[0175] A4. Digital twin-driven fluid simulation interface integration: Constructs a seamless interface with fluid simulation software, embeds the 3D model into the digital twin platform, realizes parameterized interaction through API and provides real-time feedback of simulation results, supports scene simulation, and improves the model's decision support capabilities in complex urban environments.

[0176] The generated airflow risk map in S2 is further optimized into an annual comprehensive airflow risk index map, as follows:

[0177] B1. Data Input and Preprocessing:

[0178] Input: Multiple static airflow risk probability maps generated by S2, covering the prevailing wind direction and maximum wind speed conditions for all four seasons;

[0179] Processing: Assign a weighting coefficient to each wind direction, based on the annual frequency of that wind direction provided by the local weather station. Maximum wind speed can be considered as a special case, or weighted according to its probability of occurrence.

[0180] B2. Weighted superposition calculation:

[0181] Algorithm: Establish a gridded model to divide the map area into small grid units;

[0182] Calculation: For each grid cell, its risk value is read from all static wind direction maps, and then a weighted average is calculated based on the weight of the corresponding wind direction.

[0183] Output: Generate a new grid data layer, namely the annual airflow risk composite index. The higher the index value, the greater the probability and intensity of the location being exposed to high-risk airflow environments such as high turbulence and strong winds throughout the year.

[0184] The annual comprehensive airflow risk index G is calculated for each grid cell (i, j) on the map as follows:

[0185]

[0186] Among them: G i，j The annual airflow risk index representing the grid cell located in the i-th row and j-th column is the final output, a value representing the overall risk for the whole year;

[0187] d is the wind direction index, for example, d=1 represents northerly wind, d=2 represents northeasterly wind, ..., d=8 represents northwesterly wind, and n is the total number of wind directions to be considered, usually 8 or 16 directions;

[0188] W d The weight representing the d-th wind direction is typically derived from the frequency of that wind direction throughout the year. The sum of all weights should be 1.

[0189] R d，i，j Let represent the risk value of grid cell (i,j) in the static risk map representing the d-th wind direction;

[0190] B3. Visualization and Interaction:

[0191] Generate a comprehensive map: Visualize the calculated annual comprehensive index and use color gradient rendering to generate a brand new comprehensive risk map;

[0192] Develop interactive features: Provide users with an interactive interface. When a user clicks on any point on the map, the interface not only displays the comprehensive risk index of that point, but also traces and lists the top three wind directions that contribute the most to the risk of that point and their contribution.

[0193] Based on the content in S3, the following further optimization scheme is proposed:

[0194] C1. Construct a swarm intelligence-digital twin collaborative computing framework to achieve ultra-high resolution and forward-looking prediction of airflow fields. Based on the existing real-time data assimilation CFD model, introduce a mechanism for deep collaboration between swarm intelligence computing and digital twin.

[0195] C2. Establish a dynamic airspace geofence adaptive management system to maximize risk avoidance and airspace resource utilization. Based on the ultra-high resolution and forward-looking dynamic airflow risk map generated by C1, an innovative dynamic airspace geofence management protocol is developed.

[0196] Example 2

[0197] Based on C1, the following is a detailed description: The deployed IoT meteorological sensor network is no longer just a discrete data collection point, but constitutes a swarm intelligent sensing array with edge computing capabilities. Each micro-weather station has a built-in lightweight AI chip, which can not only collect data, but also make very short-term autonomous predictions of microscale airflow changes in the next moment, such as 30-60 seconds, based on local historical data and simple models. These distributed prediction results, not just raw data, will be uploaded to the central digital twin platform as collaborative constraints. The core algorithm of the platform will be upgraded accordingly. It will no longer passively accept data and perform assimilation and correction, but actively integrate these swarm intelligent predictions from the edge and perform bidirectional verification and game learning with high-precision CFD simulation. This allows the digital twin to not only respond to changes more quickly, but also to make forward-looking predictions of the generation and evolution paths of microscale eddies and sudden gusts based on the multi-node trend clues provided by swarm intelligence, with a range of 1-5 minutes in the future. This greatly extends the lead time of the system's early warning and provides a more strategically valuable window of opportunity for safety decisions.

[0198] Example 3

[0199] Based on C2, the system is described in detail as follows: Instead of simply marking a take-off and landing point as open or closed, the entire airspace is divided into multi-dimensional dynamic risk units. The risk level of each unit, such as stable, mild turbulence, severe turbulence, no-fly zone, and its geographical boundaries, will be adjusted and changed in real time and automatically as the airflow risk map is updated. The system will assign an acceptable risk threshold that matches the flight performance and mission urgency of each UAV or eVTOL. When an aircraft requests to fly, the system will dynamically generate an optimal safe flight corridor composed of countless instantaneous safe units and apply corresponding dynamic geofencing. This means that when a sudden gust of wind arrives, the system does not need to close the entire take-off and landing point or airspace on a large scale, but will intelligently temporarily prohibit high-risk units from flying, while replanning the flight path and guiding the aircraft to detour through units that are still in a safe state. Thus, under the premise of ensuring absolute safety, the system maximizes the overall airspace capacity and achieves a balance between safety and efficiency.

[0200] The intelligent site selection method for drone take-off and landing points in urban building complexes is as follows:

[0201] T1. Start the system, load multi-source data, including oblique photogrammetry model, LiDAR point cloud, BIM / CIM model, run AI fusion algorithm, automatically fill in missing areas and extract semantic information, generate high-precision three-dimensional voxel mesh model, confirm that coordinate system I and data registration are completed, and enter the fluid simulation preparation state.

[0202] T2. Set CFD boundary conditions: Input local meteorological data, start CFD simulation calculation, output parameters such as wind speed, turbulent kinetic energy, and eddy viscosity, generate and visualize static airflow risk probability map, and identify strong turbulence area, downdraft area and steady airflow area.

[0203] T3. Import static risk maps under different wind directions in all four seasons, calculate the annual airflow risk index of each grid cell based on the annual frequency of wind direction, generate an annual comprehensive risk map, and support interactive query of the risk sources of each point;

[0204] T4. Deploy IoT meteorological sensors in key locations, calibrate sensor data to ensure time synchronization and data format consistency with the central system, and activate edge AI prediction function to achieve short-term local microclimate forecasting.

[0205] T5. The system automatically collects sensor data and edge prediction results, runs data assimilation algorithms, dynamically corrects the CFD model, and updates the airflow risk map. If a sudden gust of wind or extreme weather is detected, an early warning mechanism is triggered.

[0206] T6. Input the candidate take-off and landing point locations into the system, run the safety assessment model, calculate the turbulence intensity, wind speed ratio, wind direction stability and other indicators of each point, use a multi-objective optimization algorithm, take into account factors such as safety, efficiency and cost, and output the Pareto optimal location scheme.

[0207] T7. Based on the real-time airflow risk map, dynamically divide the airspace risk units, set an acceptable risk threshold for each drone, generate a dynamic geofence, and automatically adjust the no-fly zone and replan the flight path in case of sudden wind conditions to ensure a safe flight corridor.

[0208] T8. Record the actual airflow data and system prediction values ​​for each takeoff and landing, compare the prediction and actual data through machine learning algorithms, optimize CFD parameters and safety models, and regularly update the three-dimensional model and meteorological weights to maintain the system's prediction accuracy and adaptability.

[0209] In practice

[0210] Based on the multi-source data acquisition and fusion method described in the S1 phase, a unified three-dimensional voxel or surface mesh model with geometric accuracy, physical properties, and semantic labels is constructed by integrating oblique photogrammetry real-scene 3D models, LiDAR point clouds, and BIM / CIM models. With the enhancement of the A1 scheme, AI-driven semantic recognition and data completion technology is introduced to automatically extract and fuse key features from different data sources, solving problems such as reflection loss caused by glass curtain walls, forming a high-fidelity urban digital twin foundation. Furthermore, through the multi-scale adaptive mesh generation algorithm in the A3 scheme, the mesh resolution is dynamically optimized based on geometric curvature, semantic importance, and fluid sensitivity, and the mesh density is increased in key areas, thereby providing an efficient and accurate computational domain for CFD simulation, directly supporting the reliability and computational efficiency of airflow field simulation.

[0211] Based on the boundary conditions and CFD simulation process defined in phase S2, the inlet boundary conditions under different wind directions and speeds are calculated using the numerical wind speed profile formula combined with building surface roughness parameters. The turbulent kinetic energy (k) and dissipation rate (∈) are solved using a steady-state RANS model to quantify the intensity of airflow turbulence. Under scheme B optimization, multi-wind-direction static risk probability maps are further integrated, and an annual comprehensive airflow risk index map is generated by weighting and superimposing data according to annual occurrence frequency. This index comprehensively reflects the annual risk exposure level of various locations, and risk tracing is achieved through visualization and an interactive interface. Simultaneously, leveraging the swarm intelligence-digital twin collaborative computation framework proposed in scheme C1, the system integrates real-time sensor data and pre-calculated models. Through data assimilation and machine learning proxy models, dynamic correction and short-term prediction of the airflow field are achieved, improving the system's response capability under microclimate abrupt changes.

[0212] In the S4 phase, a quantitative safety assessment model is established, which calculates safety scores based on indicators such as turbulence intensity, wind speed ratio, and wind direction stability. Combined with business needs, a multi-objective optimization algorithm is used to solve for the Pareto optimal location scheme. Through the S5 visualization platform, airflow risk, real-time monitoring data, and optimization results are embedded into the three-dimensional environment to provide intuitive decision support. With the enhancement of the C2 scheme, a dynamic airspace geofencing adaptive management system is further constructed. Based on real-time and forward-looking airflow risk predictions, it automatically generates variable airspace usage strategies and temporarily adjusts or prohibits take-off and landing points under high-risk weather conditions to achieve a dynamic balance between airspace resources and safety requirements. At the same time, the system continuously collects actual take-off and landing data through a feedback learning mechanism to optimize the CFD model and safety assessment parameters, forming a closed-loop self-evolution capability.

[0213] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent site selection system for UAV take-off and landing points in urban building complexes, characterized by: Includes the following steps: S1. Multi-source heterogeneous data fusion and high-precision 3D environment modeling are used to construct a 3D digital twin model of the city containing accurate geometry, surface material and semantic information in order to accurately calculate the airflow field; S2. Computational fluid dynamics simulation and airflow field pre-calculation: Based on a three-dimensional model, simulate the wind field environment around the building complex under different prevailing winds and predict potential airflow turbulence areas. S3, real-time meteorological monitoring and dynamic risk correction, overcome the limitations of static simulation, enabling the system to respond to microclimate changes in real time; S4. Establish a quantitative security assessment model and multi-objective optimized site selection, replace simple distance rules with precise quantitative indicators, and make decisions by comprehensively considering efficiency factors; S5 provides a visual decision support and feedback learning interface for human-computer interaction and enables the system to continuously optimize itself.

2. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 1, characterized in that: S1 includes: S1.1 Data Acquisition: Oblique photogrammetry real-world 3D model, airborne or vehicle-mounted LiDAR point cloud data, building information model or urban planning model; S1.2 Data processing and fusion; Among them, the oblique photogrammetry real-scene 3D model: obtained through drone aerial photography, provides accurate data on the building's external geometry, height, and roof structure; Airborne or vehicle-mounted LiDAR point cloud data: supplement the acquisition of high-precision terrain, vegetation and building facade data, overcoming the shortcomings of photogrammetry in glass curtain wall areas; Building Information Model or Urban Planning Model: Obtain the official design dimensions, material properties, and semantic information of the building, where the material properties are roughness and the semantic information is purpose or height; Based on the above data, registration and fusion are performed in a unified coordinate system to generate a high-precision three-dimensional voxel mesh or surface mesh model rich in semantic information. The model includes not only shape, but also material and type attributes, providing a foundation for subsequent fluid dynamics calculations.

3. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 2, characterized in that: S2 includes: S2.1 Boundary condition settings; S2.2, CFD simulation calculation; S2.3, Generate an airflow risk map; Among them, the boundary conditions are set by inputting the typical wind direction and wind speed provided by the local weather station as the entry conditions for the CFD simulation, and setting different surface roughness parameters according to the building surface material. For the formula of numerical wind speed profile: Where u(z) is the wind speed at height z in m / s, which is the wind speed that needs to be set for the CFD inlet boundary conditions; u * The friction velocity (in m / s) is a scale velocity related to the surface shear stress. It can be calculated from the wind speed at a known reference height. R is the Karman constant, R≈0.4; z represents the height above the ground; z0 is the surface roughness length, which is set according to the building surface material; CFD simulation calculation: using high-performance computing clusters or cloud computing resources, steady-state CFD simulation is performed on large-scale building complex areas. The simulation output results include: wind speed vector field, wind pressure distribution, turbulent kinetic energy, eddy viscosity and other key fluid parameters. These parameters directly quantify the instability and turbulence of the airflow. Turbulent kinetic energy transport equation k, steady RANS model: in: Turbulent kinetic energy, unit: m 2 / s 2 , representing turbulence intensity; ρ is the fluid density; u is the time-averaged velocity vector; μ is the dynamic viscosity, μ t Turbulent viscosity; σ k The turbulent Prandtl number represents the turbulent kinetic energy. For turbulence generation; ∈ represents the turbulent dissipation rate, which indicates the rate at which kinetic energy is dissipated into heat energy; Generate airflow risk maps: Visualize and overlay CFD simulation results, especially turbulent kinetic energy and wind speed gradients, to generate static airflow risk probability maps under different wind directions; Wind speed includes the prevailing wind direction and maximum wind speed in all four seasons; building surface materials include glass, concrete, and vegetation. The static airflow risk probability map also includes: Strong turbulence areas: leeward side of buildings, narrow streets, and areas at building angles; Areas with strong descending or ascending air currents: areas near the windward and leeward sides of tall buildings; Stable airflow areas: open squares, parks, and rooftops of low-rise building complexes.

4. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 3, characterized in that: S3 includes: S3.1 Deploy an IoT weather sensor network; S3.2 Data assimilation and dynamic correction; Among them, the deployment of an Internet of Things (IoT) meteorological sensor network includes deploying low-cost micro weather stations in key locations around candidate take-off and landing points and on the rooftops of high-rise buildings to monitor wind speed, wind direction, temperature, and air pressure in real time. Data assimilation and dynamic correction: Real-time monitoring data is used as the real boundary conditions and assimilated with the pre-computed CFD model. A lightweight CFD fast correction algorithm or a machine learning-based surrogate model is used to quickly deduce the subtle changes in the airflow field of the entire area at the current moment based on real-time wind direction and speed. The airflow risk map is dynamically updated. In the event of sudden gusts or extreme weather, the system can trigger an alarm and even temporarily close high-risk take-off and landing points.

5. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 4, characterized in that: S4 includes: S4.1 Define and quantify safety indicators; S4.2 Multi-objective optimization algorithm; Among them, quantitative security indicators are defined as follows: A turbulence intensity value exceeding 15% is considered to pose a high risk to drone take-off and landing. The ratio of wind speed at a certain point to the incoming wind speed is also considered; a ratio that is too large or too small indicates an abnormal wind field. The stability of the wind direction at that point in both time and space is assessed. Based on the above indicators, a safety score between 0 and 1 is calculated. Multi-objective optimization algorithm: Input: set of candidate take-off and landing points, flat areas extracted from the model in step 1, safety score for each point, business requirements: distance to delivery point, ground pedestrian density, airspace restrictions, infrastructure costs; Optimization: Employ multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization, to maximize safety scores and operational efficiency while minimizing costs and risks, and perform optimization calculations. Output: A set of Pareto optimal takeoff and landing site selection schemes, which operators can use to make a final decision based on their current preferences, whether they prioritize safety or efficiency.

6. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 5, characterized in that: S5 includes: S5.1, Visualization Platform; S5.2 Feedback Learning Mechanism Among them, the visualization platform: In the three-dimensional digital twin city model, the airflow risk map, real-time sensor data, candidate point safety scores and optimization results are displayed intuitively, and managers can see the airflow conditions and risk levels of different locations at a glance; Feedback learning mechanism: The system records the actual take-off and landing data of each drone, compares this real data with the system's predicted risk value, and continuously corrects and optimizes the CFD model parameters and safety assessment model through machine learning algorithms, making the predictions more and more accurate.

7. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 6, characterized in that: The further optimization scheme based on the content in S1 is as follows: A1. Intelligent fusion and semantic enhancement of multi-source data: Using AI technology, it automatically fuses geometric, material and semantic information from oblique photogrammetry models, LiDAR point clouds and BIM / CIM models. It identifies and fills in missing areas through deep learning, and extracts high-precision semantic labels to form a unified three-dimensional voxel mesh with physical properties and semantic information. A2. Dynamic data embedding and real-time update mechanism: Introduces IoT sensor data and real-time remote sensing data, maps dynamic information to three-dimensional grid nodes through spatiotemporal coding, establishes an incremental update algorithm, realizes automatic correction of the model in a short period of time, and ensures that fluid dynamics calculations are based on the latest environmental conditions. A3. Multi-scale adaptive mesh generation and optimization: To meet the needs of fluid computing, an adaptive mesh refinement algorithm is developed: Based on surface curvature, semantic importance and fluid sensitivity, the voxel resolution is dynamically adjusted, while the mesh topology is optimized to reduce computational redundancy and retain key attributes, and the output is directly in a CFD-compatible format. A4. Digital twin-driven fluid simulation interface integration: Constructs a seamless interface with fluid simulation software, embeds the 3D model into the digital twin platform, realizes parameterized interaction through API and provides real-time feedback of simulation results, supports scene simulation, and improves the model's decision support capabilities in complex urban environments.

8. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 7, characterized in that: The airflow risk map generated in S2 is further optimized into an annual comprehensive airflow risk index map, and the scheme is as follows: B1. Data Input and Preprocessing: Input: Multiple static airflow risk probability maps generated by S2, covering the prevailing wind direction and maximum wind speed conditions for all four seasons; Processing: Assign a weighting coefficient to each wind direction, based on the annual frequency of that wind direction provided by the local weather station. Maximum wind speed can be considered as a special case, or weighted according to its probability of occurrence. B2. Weighted superposition calculation: Algorithm: Establish a gridded model to divide the map area into small grid units; Calculation: For each grid cell, its risk value is read from all static wind direction maps, and then a weighted average is calculated based on the weight of the corresponding wind direction. Output: Generate a new grid data layer, namely the annual airflow risk composite index. The higher the index value, the greater the probability and intensity of the location being exposed to a high-turbulence, high-wind, high-risk airflow environment throughout the year. Generate a comprehensive map: Visualize the calculated annual comprehensive index and use color gradient rendering to generate a brand new comprehensive risk map; Develop interactive features: Provide users with an interactive interface. When a user clicks on any point on the map, the interface not only displays the comprehensive risk index of that point, but also traces and lists the top three wind directions that contribute the most to the risk of that point and their contribution.

9. The intelligent site selection system for UAV take-off and landing points in urban building complexes according to claim 8, characterized in that: The further optimization scheme based on the content in S3 is as follows: C1. Construct a swarm intelligence-digital twin collaborative computing framework to achieve ultra-high resolution and forward-looking prediction of airflow fields. Based on the existing real-time data assimilation CFD model, introduce a mechanism for deep collaboration between swarm intelligence computing and digital twin. C2. Establish a dynamic airspace geofence adaptive management system to maximize risk avoidance and airspace resource utilization. Based on the ultra-high resolution and forward-looking dynamic airflow risk map generated by C1, an innovative dynamic airspace geofence management protocol is developed.

10. The intelligent site selection method for UAV take-off and landing points in urban building clusters according to any one of claims 1-9, characterized in that: The intelligent site selection method for UAV take-off and landing points in the urban building complex is as follows: T1. Start the system, load multi-source data, including oblique photogrammetry model, LiDAR point cloud, BIM / CIM model, run AI fusion algorithm, automatically fill in missing areas and extract semantic information, generate high-precision three-dimensional voxel mesh model, confirm that coordinate system I and data registration are completed, and enter the fluid simulation preparation state. T2. Set CFD boundary conditions: Input local meteorological data, start CFD simulation calculation, output wind speed, turbulent kinetic energy, eddy viscosity parameters, generate and visualize static airflow risk probability map, and identify strong turbulence area, downdraft area and steady airflow area. T3. Import static risk maps under different wind directions in all four seasons, calculate the annual airflow risk index of each grid cell based on the annual frequency of wind direction, generate an annual comprehensive risk map, and support interactive query of the risk sources of each point; T4. Deploy IoT meteorological sensors in key locations, calibrate sensor data to ensure time synchronization and data format consistency with the central system, and activate edge AI prediction function to achieve short-term local microclimate forecasting. T5. The system automatically collects sensor data and edge prediction results, runs data assimilation algorithms, dynamically corrects the CFD model, and updates the airflow risk map. If a sudden gust of wind or extreme weather is detected, an early warning mechanism is triggered. T6. Input the candidate take-off and landing point locations into the system, run the safety assessment model, calculate the turbulence intensity, wind speed ratio, and wind direction stability index of each point, use a multi-objective optimization algorithm, and output the Pareto optimal location scheme by comprehensively considering safety, efficiency, and cost factors. T7. Based on the real-time airflow risk map, dynamically divide the airspace risk units, set an acceptable risk threshold for each drone, generate a dynamic geofence, and automatically adjust the no-fly zone and replan the flight path in case of sudden wind conditions to ensure a safe flight corridor. T8. Record the actual airflow data and system prediction values ​​for each takeoff and landing, compare the prediction and actual data through machine learning algorithms, optimize CFD parameters and safety models, and regularly update the three-dimensional model and meteorological weights to maintain the system's prediction accuracy and adaptability.

Images