A virtual-real fusion complex mountainous unmanned aerial vehicle low-altitude logistics simulation verification platform

Abstract

The application provides a virtual-real fusion complex mountain unmanned aerial vehicle low-altitude logistics simulation verification platform, and belongs to the technical field of simulation verification. A target mountain three-dimensional digital twin scene and a low-altitude wind field numerical model containing micro-terrain turbulence are constructed through a twin scene construction module. A wind field dynamic assimilation module generates a virtual-real mapping wind field by taking real unmanned aerial vehicle test flight measured wind field data as a constraint. A consistency deviation analysis module loads the mapping wind field in the scene, drives virtual unmanned aerial vehicle simulation flight, compares the attitude response with the real flight state, and performs consistency deviation analysis. A reverse calibration module reversely adjusts the virtual unmanned aerial vehicle aerodynamic parameters and flight control strategy according to the deviation until the dynamic response deviation is less than a threshold. A working condition verification module simulates and verifies the safety margin of the logistics route under combined working conditions based on the calibrated model. The technical scheme provided by the application can improve the accuracy of unmanned aerial vehicle low-altitude logistics flight safety evaluation in complex mountain environments.

Description

Technical Field

[0001] This application relates to the field of simulation verification technology, and more specifically, to a simulation verification platform for low-altitude logistics of complex mountain drones that integrates virtual and real elements. Background Technology

[0002] With the rapid application of drone technology in logistics and transportation, low-altitude logistics transportation in complex mountainous environments is becoming an important scenario for the commercial operation of drones. Mountainous terrain is characterized by dramatic undulations and variable weather conditions, especially the complex wind field features prevalent in low-altitude areas, such as micro-topographic disturbances, canyon winds, gusts, and terrain-induced eddies. These pose significant challenges to the flight stability, flight path maintenance capabilities, and energy consumption control of drones. Traditional simulation verification methods based on ideal wind fields or simplified environmental models cannot accurately reflect the dynamic response characteristics of drones in complex mountainous low-altitude environments, leading to significant discrepancies between simulation results and actual flight performance. This hinders the safety assessment and airworthiness certification of drone logistics systems in mountainous scenarios.

[0003] Currently, UAV flight simulation mainly relies on numerical wind field simulation or historical meteorological data to construct environmental conditions, lacking the ability to map high-precision wind fields synchronously with the actual flight process. For example, Chinese invention patent CN119493381A discloses a method and system for fusion simulation of virtual UAVs and real UAVs, involving the field of UAV modeling and simulation technology. This includes: acquiring the dynamic parameters of the real UAV and the kinematic parameters detected by sensor modules on the real UAV; sending the dynamic and kinematic parameters of the real UAV to the simulation platform via a communication module; establishing a virtual UAV model based on the dynamic parameters of the real UAV; configuring sensors in the virtual UAV model based on the kinematic parameters of the real UAV; and selecting a simulation environment based on the application scenario to achieve fusion simulation of the virtual UAV. This solution uses the dynamic and kinematic parameters of the real UAV to drive the virtual model, lacking the ability to acquire and assimilate complex environmental factors such as actual low-altitude wind fields and micro-topographical disturbances in real time. This results in a disconnect between the simulation environment of the virtual UAV and the actual flight conditions, failing to reflect the impact of real wind fields on the dynamic response of the UAV in complex mountainous scenarios. Meanwhile, the scheme does not introduce consistency deviation analysis and model back-correction steps between real flight data and virtual simulation results, making it difficult to continuously optimize UAV aerodynamic parameters and flight control strategy models through virtual-real comparison, thus limiting the credibility and airworthiness verification value of simulation results in real complex environments.

[0004] Therefore, there is an urgent need for a virtual-real fusion simulation verification platform that can achieve dynamic synchronization between real wind fields and virtual simulation environments and support the injection of high-risk complex working conditions, so as to improve the accuracy and reliability of safety assessment of UAV low-altitude logistics flights in complex mountainous environments. Summary of the Invention

[0005] This application provides a virtual-real integrated simulation and verification platform for low-altitude logistics of UAVs in complex mountainous environments, which can improve the accuracy of safety assessment of UAV low-altitude logistics flights in complex mountainous environments.

[0006] This application provides a simulation and verification platform for low-altitude logistics of complex mountain drones that integrates virtual and real elements. The platform includes: The twin scene construction module is used to construct a three-dimensional digital twin scene of the target mountainous area and to establish a low-altitude wind field numerical model containing micro-topographic disturbance features in the three-dimensional digital twin scene. The wind field dynamic assimilation module is used to dynamically assimilate and correct the low-altitude wind field numerical model by using the measured low-altitude wind field data of the actual UAV during the test flight of the preset logistics route in the target mountainous area as a constraint, so as to obtain a virtual-real mapped wind field that is synchronized with the real wind field. The consistency deviation analysis module is used to load the virtual-real mapped wind field into the three-dimensional digital twin scene, drive the virtual drone to perform full-process flight simulation according to the preset logistics route, and perform consistency deviation analysis between the flight attitude response sequence of the virtual drone and the flight status data of the real drone during the simulation process. The reverse calibration module is used to adjust the aerodynamic parameters and flight control strategy model of the virtual drone in reverse according to the consistency deviation analysis results, until the dynamic response deviation between the virtual drone and the real drone under the same wind field conditions is less than a preset threshold. The working condition verification module is used to inject artificially set extreme gust disturbances and mountain terrain-induced eddy currents into the virtual-real mapped wind field as a baseline using the adjusted aerodynamic parameters and flight control strategy model of the virtual UAV. This drives the virtual UAV to perform risk logistics flight simulation and verifies the safety margin of the preset logistics route under the combined working conditions during the simulation.

[0007] Preferably, in the twin scene construction module, constructing a three-dimensional digital twin scene of the target mountainous region specifically includes: Acquire multi-source geospatial data of the target mountainous region, register and fuse the multi-source geospatial data, and generate a basic terrain grid of the target mountainous region; The basic terrain grid is subjected to semantic segmentation and classification of land features, and the surface cover type of the mountain and the boundary of artificial structures are extracted to generate a terrain grid with semantic labels. Based on the terrain mesh with semantic tags, a 3D static scene model that is consistent with the geometric appearance of real terrain features is constructed. The surface reflectance and meteorological illumination data of the target mountainous area are obtained, and the material illumination mapping is performed on the three-dimensional static scene model to generate a three-dimensional digital twin scene with realistic textures.

[0008] Preferably, in the twin scene construction module, establishing a low-altitude wind field numerical model containing micro-topographic disturbance features in the three-dimensional digital twin scene specifically includes: Extract the digital elevation model of the target mountainous area in the three-dimensional digital twin scene, and calculate the terrain slope, aspect and undulation of the digital elevation model to obtain the micro-topographic feature parameter field; Using the micro-topographic feature parameter field as the lower boundary condition, a computational fluid dynamics grid is constructed for the low-altitude logistics flight altitude layer. The computational fluid dynamics grid is locally refined in the near-wall and ridge regions of the mountain. Meteorological background wind field data of the target mountainous area is obtained, and the meteorological background wind field data is loaded into the computational fluid dynamics grid as the incoming flow boundary condition. The Reynolds-averaged Navier-Stokes equations are solved to obtain a low-altitude wind field numerical model that includes the characteristics of mountain flow around, canyon acceleration and leeward vortex.

[0009] Preferably, in the wind field dynamic assimilation module, the measured low-altitude wind field data during the test flight of the real UAV on the target mountainous area executing a preset logistics route specifically includes: An atmospheric data sensor is mounted on a real drone, and the installation error between the measurement reference of the atmospheric data sensor and the coordinate system of the real drone is calibrated to obtain the sensor calibration parameters. Control the real drone to perform test flights along the preset logistics route, and use the atmospheric data sensor to collect raw airflow parameters during the flight at a preset sampling frequency; The original airflow parameters were processed by body motion compensation and outlier removal to obtain measured low-altitude wind field data after removing interference from the UAV's own motion.

[0010] Preferably, in the wind field dynamic assimilation module, the low-altitude wind field measurement data during the test flight of a real UAV executing a preset logistics route in the target mountainous area is used as a constraint to dynamically assimilate and correct the low-altitude wind field numerical model, resulting in a virtual-real mapped wind field that is synchronized with the real wind field. Specifically, this includes: Obtain the simulated wind field values ​​of the low-altitude wind field numerical model at the corresponding spatiotemporal points of the test flight route, and align the measured low-altitude wind field data with the simulated wind field values ​​in time and space to construct an assimilated observation increment field. Using the assimilated observation increment field as a constraint, a sequential assimilation correction based on ensemble Kalman filtering is applied to the state variables of the low-altitude wind field numerical model. In the assimilation process, a micro-topographic feature parameter field is introduced as a localized weighting factor for the error covariance to obtain the assimilated low-altitude wind field state variables. The assimilated low-altitude wind field state variables are subjected to time advancement and spatial interpolation to generate a virtual-real mapped wind field covering the entire area of ​​the preset logistics route.

[0011] Preferably, in the consistency deviation analysis module, loading the virtual-real mapped wind field into the three-dimensional digital twin scene and driving the virtual drone to perform full-process flight simulation according to the preset logistics route specifically includes: The virtual-real mapped wind field is discretized and mapped according to the spatial grid nodes of the three-dimensional digital twin scene, and a wind field data query structure with the coordinate system of the three-dimensional digital twin scene is established. During the virtual drone flight simulation, the wind vector at the corresponding position is obtained by interpolation from the wind field data query structure based on the real-time position of the virtual drone on the preset logistics route. The wind vector is loaded as an aerodynamic external force input into the six-degree-of-freedom dynamic model of the virtual UAV, driving the virtual UAV to complete the entire flight simulation process from takeoff, cruise, obstacle avoidance descent to landing.

[0012] Preferably, in the consistency deviation analysis module, the consistency deviation analysis of the flight attitude response sequence of the virtual UAV and the flight state data of the real UAV during the simulation process specifically includes: By aligning the flight attitude response sequence of the virtual drone with the flight status data recorded by the real drone during test flights on the same preset logistics route according to the timestamp, a spatiotemporally aligned sample set is constructed. The time-frequency features of attitude angle, angular velocity, linear acceleration and airspeed are extracted from the virtual attitude response sequence and real flight state data in the same spatiotemporal aligned sample set, respectively, to obtain the virtual response feature set and the real flight feature set; Calculate the multi-scale dynamic time warping distance between the virtual response feature set and the real flight feature set under the same flight phase label, and map the multi-scale dynamic time warping distance as a consistency deviation index.

[0013] Preferably, in the reverse calibration module, the aerodynamic parameters and flight control strategy model of the virtual UAV are adjusted in reverse according to the consistency deviation analysis results until the dynamic response deviation between the virtual UAV and the real UAV under the same wind field conditions is less than a preset threshold. Specifically, this includes: Using the consistency deviation index as the fitness function, a global sensitivity analysis is performed on the aerodynamic parameters and adjustable parameters in the flight control strategy model of the virtual UAV, and a subset of key parameters whose contribution rate to the deviation exceeds a preset proportion is selected. Using the subset of key parameters as optimization variables and minimizing the consistency deviation index as the optimization objective, a Bayesian optimization strategy based on a surrogate model is used to iteratively correct the aerodynamic parameters and flight control strategy model. After each iteration and correction, the virtual drone is re-driven to perform a full-process flight simulation in the virtual-real mapped wind field and the consistency deviation index is recalculated until the consistency deviation index is less than the preset threshold.

[0014] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The virtual-real fusion simulation and verification platform for low-altitude logistics of UAVs in complex mountainous environments provided in this application can improve the accuracy of safety assessment for low-altitude logistics flights of UAVs in complex mountainous environments. First, the twin scene construction module constructs a 3D digital twin scene consistent with the geometry, semantics, and lighting characteristics of real mountainous areas by fusing multi-source geospatial data. Within this scene, a low-altitude wind field numerical model incorporating micro-topographic disturbance features is established, enabling the simulation environment to physically reproduce the effects of real terrain on low-altitude wind field disturbances. This provides a unified spatial benchmark and wind field data source for subsequent virtual-real consistency analysis and safety verification. Second, the wind field dynamic assimilation module uses low-altitude wind field data measured during real UAV test flights to dynamically assimilate and correct the numerical model, generating a virtual-real mapped wind field that is time-synchronized and spatially continuous with the real wind field. This effectively eliminates the systematic bias of pure numerical simulation in near-surface wind fields in mountainous areas, providing highly reliable wind field excitation conditions for virtual UAV simulation. Furthermore, the consistency deviation analysis module loads a virtual-to-real wind field mapping into a 3D digital twin scene and drives the virtual UAV to perform a full-process flight simulation. It then performs multi-scale time-frequency consistency deviation analysis between the virtual attitude response sequence and the real flight state data, quantitatively revealing the differences in dynamic response between the virtual and real UAVs under the same wind field conditions. This provides a precise quantitative basis for the reverse calibration of model parameters. Next, the reverse calibration module uses the consistency deviation index as the fitness function. Through global sensitivity analysis, it filters a subset of key parameters and employs a Bayesian optimization strategy to reverse-adjust the aerodynamic parameters and flight control strategy model of the virtual UAV. This ensures that the dynamic response of the virtual UAV approximates that of the real UAV within a preset threshold, guaranteeing that the dynamic model used in high-risk simulations has response characteristics consistent with the real aircraft. This accurately reflects the dynamic response characteristics of the UAV in complex mountainous low-altitude environments, effectively avoiding significant deviations between simulation results and actual flight performance. Finally, the operational condition verification module injects a combination of extreme gusts and terrain-induced eddies into the model baseline after virtual-real calibration and performs a high-risk logistics flight simulation. It verifies the feasibility of the preset logistics route under extreme wind conditions using comprehensive safety margin indicators such as attitude stability, track maintenance, power reserve, and near-ground separation. This provides a quantitative evaluation method for the safety assessment and risk management of low-altitude logistics routes for UAVs in complex mountainous environments, improving the accuracy of safety assessments for low-altitude logistics flights of UAVs in complex mountainous environments. In summary, the technical solution provided in this application can improve the accuracy of safety assessments for low-altitude logistics flights of UAVs in complex mountainous environments. Attached Figure Description

[0015] Figure 1 This is a modular structure diagram of a complex mountain UAV low-altitude logistics simulation and verification platform that integrates virtual and real elements, as shown in this application. Figure 2 This is an exemplary flowchart for determining a virtual-to-real mapping wind field as shown in this application. Detailed Implementation

[0016] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] refer to Figure 1 As shown in the figure, this is a modular structure diagram of a complex mountain UAV low-altitude logistics simulation and verification platform that integrates virtual and real elements, according to this application. The system includes: a twin scene construction module 100, a wind field dynamic assimilation module 200, a consistency deviation analysis module 300, a reverse calibration module 400, and a working condition verification module 500, which are described below: The twin scene construction module 100 is used to construct a three-dimensional digital twin scene of the target mountainous area and to establish a low-altitude wind field numerical model containing micro-topographic disturbance features in the three-dimensional digital twin scene.

[0018] It should be noted that the target mountainous area in this application refers to a pre-defined mountainous geographical area with significant topographic relief features, used for conducting simulation verification of low-altitude logistics flights of unmanned aerial vehicles. This area includes typical micro-topographic units such as peaks, ridges, canyons, and steep slopes, and its topographic complexity is sufficient to produce a identifiable disturbance effect on the low-altitude wind field. It serves as a spatial carrier for the construction of three-dimensional digital twin scenes, numerical modeling of low-altitude wind fields, and virtual-real fusion simulation verification.

[0019] In the twin scene construction module, the construction of a three-dimensional digital twin scene of the target mountainous region is achieved through the following steps: Acquire multi-source geospatial data of the target mountainous region, register and fuse the multi-source geospatial data, and generate a basic terrain grid of the target mountainous region; The basic terrain grid is subjected to semantic segmentation and classification of land features, and the surface cover type of the mountain and the boundary of artificial structures are extracted to generate a terrain grid with semantic labels. Based on the terrain mesh with semantic tags, a 3D static scene model that is consistent with the geometric appearance of real terrain features is constructed. The surface reflectance and meteorological illumination data of the target mountainous area are obtained, and the material illumination mapping is performed on the three-dimensional static scene model to generate a three-dimensional digital twin scene with realistic textures.

[0020] In practice, firstly, digital surface models, high-resolution orthophotos, and digital elevation models of the target mountainous area are acquired through satellite remote sensing stereo image pairs or airborne lidar scanning to obtain multi-source geospatial data of the target mountainous area. Under the premise of uniformly adopting the CGCS2000 geographic coordinate system and Gauss-Kruger projection, the digital surface models are registered using the digital elevation model as the vertical reference. Then, the overlapping areas of the multi-source point cloud data are rigidly stitched and fused using the iterative nearest-point algorithm to obtain a basic topographic grid of the target mountainous area with a spatial resolution better than 0.5 meters and an elevation error not exceeding 1 meter. The basic topographic grid refers to a grid composed of irregular triangular meshes or regular grids. A three-dimensional spatial data structure representing the continuous landform of a mountainous region is constructed. Then, a high-resolution orthophoto of the target mountainous region is input into a land cover classification model pre-trained based on a DeepLabV3+ semantic segmentation network. The surface cover type and boundaries of artificial structures are extracted pixel-by-pixel. The extracted results are projected onto the base terrain mesh as a texture mask. Each terrain mesh patch is assigned semantic labels such as woodland, bare rock, shrubs, roads, and building roofs, generating a terrain mesh with semantic labels. Semantic labels are attribute markers used to identify the real land cover category corresponding to the terrain mesh patch. The land cover classification model uses historical orthophotos of the target mountainous region and human... The labeled samples serve as the training set, with labeled categories including at least woodland, bare rock, shrubs, roads, and building rooftops. The average intersection-union ratio (IUU) of the training and validation sets is no less than 0.75. Next, using the semantically labeled terrain mesh as the geometric framework, corresponding components are retrieved from a pre-defined 3D terrain model library based on the terrain feature categories corresponding to different semantic labels. Individual geometric reconstructions are performed on man-made structures such as buildings, power poles, communication base stations, fall arrest nets, and landing platforms. For natural cover types such as woodland and shrubs, a vegetation instantiation scattering algorithm is used to generate vegetation models with realistic plant height and canopy morphology on the corresponding semantic patches. This constructs a 3D static scene model consistent with the geometric appearance of real terrain features. The three-dimensional static scene model refers to a virtual scene of a mountainous area that includes the geometric structure of terrain, land features, and vegetation but does not include dynamic processes. Finally, multispectral remote sensing images of the target mountainous area are acquired to invert the surface reflectance, and solar altitude angle, solar azimuth angle, and sky diffuse illuminance data at the corresponding time on the test flight day are acquired. The surface reflectance is mapped to bidirectional reflectance distribution function parameters and assigned to the surface materials of various features in the three-dimensional static scene model. At the same time, global illumination rendering is performed on the three-dimensional static scene model using solar altitude angle and solar azimuth angle as directional light source parameters and sky diffuse illuminance data as ambient lighting parameters, thereby generating a three-dimensional digital twin scene with realistic textures.

[0021] It should be noted that the three-dimensional digital twin scene in this application refers to a virtual three-dimensional environment that is consistent with the target mountainous area in terms of geometric shape, land feature semantics, surface material, and lighting characteristics. The three-dimensional digital twin scene provides the boundary conditions and mesh generation basis for the low-altitude wind field numerical model with high-precision geometric and semantic information, so that the wind field calculation can accurately reflect the micro-topographic effects such as mountain bypass, canyon acceleration, and leeward vortex. At the same time, the scene serves as the operating environment for virtual UAV flight simulation, providing land feature collision detection, flight path space constraints, and sensor visual data support for UAV dynamics calculation.

[0022] In the twin scene construction module, the following steps are used to establish a low-altitude wind field numerical model that includes micro-topographic disturbance features in the three-dimensional digital twin scene: Extract the digital elevation model of the target mountainous area in the three-dimensional digital twin scene, and calculate the terrain slope, aspect and undulation of the digital elevation model to obtain the micro-topographic feature parameter field; Using the micro-topographic feature parameter field as the lower boundary condition, a computational fluid dynamics grid is constructed for the low-altitude logistics flight altitude layer. The computational fluid dynamics grid is locally refined in the near-wall and ridge regions of the mountain. Meteorological background wind field data of the target mountainous area is obtained, and the meteorological background wind field data is loaded into the computational fluid dynamics grid as the incoming flow boundary condition. The Reynolds-averaged Navier-Stokes equations are solved to obtain a low-altitude wind field numerical model that includes the characteristics of mountain flow around, canyon acceleration and leeward vortex.

[0023] In specific implementation, firstly, a digital elevation model of the target mountainous area is extracted from the basic terrain grid of the 3D digital twin scene. Using raster terrain analysis tools in ArcGIS or SAGA GIS, the slope, aspect, and topographic relief are calculated pixel-by-pixel in the digital elevation model using a 3×3 moving window. Slope refers to the angle between the tangent plane of the ground and the horizontal plane; aspect refers to the azimuth angle of the projection of the normal to the tangent plane onto the horizontal plane; and topographic relief refers to the difference between the maximum and minimum elevations within a defined neighborhood centered on a given pixel. These three terrain parameters are combined pixel-by-pixel to form a micro-topographic feature parameter field. This field is a set of terrain feature data describing the slope, aspect, and relief values ​​and their spatial distribution at each spatial location within the target mountainous area in raster form. Then, using the micro-topographic feature parameter field as the lower boundary condition, an unstructured tetrahedral computational fluid dynamics (CFD) mesh oriented towards the low-altitude logistics flight altitude layer is constructed using a computational fluid dynamics preprocessing tool. During mesh generation, slopes greater than a certain value are considered as low as a certain value. Steep slopes, 50-meter buffer zones on both sides of the ridgeline, and the narrowest section of the canyon are designated as localized densification zones. Within these zones, the height of the first-layer grid near the wall is designed to satisfy the requirement that the dimensionless wall distance y+ value is within the range of 30-300, meeting the grid requirements of the wall function method for the logarithmic law region. The y+ value refers to a parameter in computational fluid dynamics used to characterize the dimensionless distance from the first-layer grid node near the wall to the wall. Simultaneously, the vertical resolution of the grid is maintained at no less than 5 meters within the flight altitude layer, resulting in a computational fluid dynamics grid capable of analyzing near-surface turbulence characteristics. This computational fluid dynamics grid refers to a spatial grid system used to discretize the solution domain of the fluid control equations. Finally, hourly background wind field data for the target mountain area, provided by meteorological stations or numerical weather prediction models, is obtained during the flight test period. This background wind field data is then interpolated at different altitude levels based on wind speed, wind direction, and turbulence intensity, and applied as the inflow boundary condition to the inflow boundary of the computational fluid dynamics grid. The outflow boundary is set as a pressure outlet, the top boundary as a symmetry plane, and the mountain surface as a non-slip wall. SST (Self-Stress Transmission) is used. The k-ω turbulence model closes the Reynolds-averaged Navier-Stokes equations and uses the SIMPLE algorithm for pressure-velocity coupled iterative solutions until the residuals decrease to a certain value. The model converges when the wind speed at the monitoring point no longer changes significantly, resulting in a low-altitude wind field numerical model that includes characteristics of mountain bypass, canyon acceleration, and leeward vortex.

[0024] It should be noted that the low-altitude wind field numerical model in this application refers to a numerical solution model that expresses the spatial distribution and temporal evolution of airflow within the flight altitude layer of low-altitude logistics in the target mountainous area in the form of discrete wind vectors at grid nodes. The low-altitude wind field numerical model in this scheme is determined by: using the micro-topographic feature parameter field as the lower boundary condition and solving the Reynolds-averaged Navier-Stokes equations with locally refined computational fluid dynamics grids, which can quantitatively reproduce the micro-topographic disturbance characteristics such as flow around mountains, acceleration in canyons, and leeward vortices within the flight altitude layer of low-altitude logistics. This overcomes the shortcomings of traditional uniform wind field or simple wind profile models that cannot reflect the spatial heterogeneity of near-surface airflow in complex mountainous areas, and provides a priori wind field state with physical consistency for the subsequent dynamic assimilation correction of the virtual-real mapped wind field. At the same time, it provides a high-resolution, queryable three-dimensional wind vector data source for virtual UAV flight simulation.

[0025] The wind field dynamic assimilation module 200 is used to dynamically assimilate and correct the low-altitude wind field numerical model by using the measured low-altitude wind field data of the actual UAV during the test flight of the preset logistics route in the target mountainous area as a constraint, so as to obtain a virtual-real mapped wind field that is synchronized with the real wind field.

[0026] In the wind field dynamic assimilation module, the measured low-altitude wind field data of the real UAV during its test flight of a preset logistics route in the target mountainous area is obtained through the following steps: An atmospheric data sensor is mounted on a real drone, and the installation error between the measurement reference of the atmospheric data sensor and the coordinate system of the real drone is calibrated to obtain the sensor calibration parameters. Control the real drone to perform test flights along the preset logistics route, and use the atmospheric data sensor to collect raw airflow parameters during the flight at a preset sampling frequency; The original airflow parameters were processed by body motion compensation and outlier removal to obtain measured low-altitude wind field data after removing interference from the UAV's own motion.

[0027] In practice, firstly, an atmospheric data sensor, such as a five-hole probe or an ultrasonic anemometer, is mounted on the arm or body of a real drone using a vibration-damping gimbal. Using the drone's body coordinate system as a reference, a three-axis turntable is used to perform multi-angle static calibration of the atmospheric data sensor. The installation offset angle and scale factor error between the sensor's measurement axis and the drone's coordinate axis are calculated to obtain sensor calibration parameters. These calibration parameters refer to the compensation coefficients and rotation matrix used to convert the sensor's raw output into standard airflow parameters in the drone's coordinate system. Then, the real drone is controlled to perform a test flight along a preset logistics route. The atmospheric data sensor is used to synchronously collect raw airflow parameters during the flight at a preset sampling frequency of no less than 50Hz. These raw airflow parameters include at least the total incoming pressure, static pressure, angle of attack, sideslip angle, and three-dimensional wind speed calculated from ultrasonic time-of-flight data. The components, where the original airflow parameters refer to the direct output of the sensor without any filtering or compensation processing; finally, the original airflow parameters are processed by body motion compensation and outlier removal. Body motion compensation uses the three-axis angular velocity and linear acceleration data of the UAV synchronously recorded by the inertial measurement unit. By coordinate transformation, the relative airflow velocity in the sensor fixed coordinate system is subtracted from the linear velocity component caused by the body rotation and the airflow inertial term induced by the body acceleration to obtain the absolute wind vector relative to the ground. Outlier removal is carried out by a combination of median filtering and the Laida criterion. First, a 7-point sliding median filter is used to remove impulse noise. Then, the Laida criterion is used to judge the median filter residual and remove outlier sampling points that exceed 3 times the standard deviation. The missing values ​​after removal are filled by linear interpolation, thus obtaining the measured low-altitude wind field data after removing the interference of the UAV's own motion.

[0028] It should be noted that the measured low-altitude wind field data in this application refers to the real atmospheric three-dimensional wind vector time series at each spatiotemporal point of the preset logistics route, decoupled from the actual UAV motion state. In this scheme, by equipping the actual UAV with atmospheric data sensors and undergoing rigorous installation error calibration, airframe motion compensation, and outlier removal, the real atmospheric three-dimensional wind vector time series decoupled from the UAV's own motion state can be obtained. This data directly reflects the real spatiotemporal variation characteristics of the low-altitude wind field during the actual flight of the preset logistics route, providing highly reliable observational constraints for the subsequent dynamic assimilation and correction of the low-altitude wind field numerical model. This significantly reduces the systematic bias of pure numerical simulation in the near-surface wind field simulation under complex mountainous terrain conditions, and enhances the basic guarantee capability for the synchronization and consistency between the virtual and real wind fields and the real wind fields.

[0029] In some embodiments, reference Figure 2 As shown in the figure, this is an exemplary flowchart for determining the virtual-real mapped wind field according to this application. In the wind field dynamic assimilation module, the low-altitude wind field measured data during the test flight of a real UAV performing a preset logistics route in the target mountainous area is used as a constraint to dynamically assimilate and correct the low-altitude wind field numerical model, thereby obtaining a virtual-real mapped wind field that is synchronized with the real wind field. This is achieved through the following steps: In step 2001, the simulated wind field values ​​of the low-altitude wind field numerical model at the corresponding spatiotemporal points of the test flight route are obtained, and the measured low-altitude wind field data and the simulated wind field values ​​are spatiotemporally aligned to construct an assimilated observation increment field. In step 2002, the state variables of the low-altitude wind field numerical model are subjected to sequential assimilation correction based on ensemble Kalman filtering, with the assimilation observation increment field as a constraint. In the assimilation process, the micro-topographic feature parameter field is introduced as a localized weighting factor of the error covariance to obtain the assimilated low-altitude wind field state variables. In step 2003, the assimilated low-altitude wind field state quantity is time-advanced and spatially interpolated to generate a virtual-real mapped wind field covering the entire area of ​​the preset logistics route.

[0030] In specific implementation, firstly, based on the timestamp and three-dimensional position coordinates of each sampling point in the real UAV test flight data, the simulated wind field value of the corresponding spatiotemporal point is obtained through inverse distance weighted interpolation in the unstructured computational grid of the low-altitude wind field numerical model. The measured low-altitude wind field data and the simulated wind field value at the same spatiotemporal point are then subtracted one by one according to the three components of the wind vector, forming an assimilated observation increment field composed of three elements: observation point location, observation time, and wind vector increment. The assimilated observation increment field refers to the vector set of the deviation between the measured wind field and the numerically simulated wind field at discrete observation spatiotemporal points. Then, using the assimilated observation increment field as a constraint, a sequential assimilation correction based on ensemble Kalman filtering is applied to the state variables of the low-altitude wind field numerical model. First, Gaussian perturbations are applied to the incoming flow boundary conditions and turbulence parameters of the low-altitude wind field numerical model to generate an ensemble forecast sample containing 80–120 members. The ensemble forecast is then used… The background error covariance of the sample statistical state variables is calculated, and then the Kalman gain matrix is ​​calculated time-by-time according to the observation time sequence. At the same time, the locations in the micro-topographic feature parameter field where the slope change rate exceeds a preset threshold are used as localized weighting factors of the error covariance, so that the observation increment only updates the state variables within the horizontal correlation radius around the location of the assimilated observation increment, which is adapted to the topographic change characteristics, to obtain the assimilated low-altitude wind field state quantity. Here, the assimilated low-altitude wind field state quantity refers to the wind field variable estimate on the computational grid node after being corrected by the observation data. Finally, the assimilated low-altitude wind field state quantity is updated by time advancement starting from the start time of the test flight period. At each advancement time step, the wind field state quantity on the unstructured computational grid node is interpolated to the three-dimensional spatial range covered by the preset logistics route to generate a virtual-real mapped wind field with a time resolution matching the actual test flight sampling frequency.

[0031] It should be noted that the virtual-real mapped wind field in this application refers to a three-dimensional dynamic wind field data volume that maintains a temporal synchronization and spatial continuous mapping relationship with the real wind field. In this embodiment, the assimilation observation increment field between the measured low-altitude wind field data and the simulated value of the low-altitude wind field numerical model is used as a constraint. In the sequential assimilation correction process based on ensemble Kalman filtering, the micro-topographic feature parameter field is used as a localized weighting factor of the error covariance. This allows the correction amount of the observed data to the wind field state to be adaptively limited to the local influence range with topographic physical significance according to the spatial variation characteristics of the mountain slope, aspect, and undulation. Thus, while maintaining the physical consistency of the numerical model, the dynamic sequential correction of the numerical simulation deviation by the measured data is realized, generating a virtual-real mapped wind field with real flight test observation temporal synchronization and coverage of the entire preset logistics route. This effectively solves the coupling contradiction between the spatial sparsity of the existing measured wind field data and the systematic deviation of the near-surface wind field in the numerical simulation of mountains.

[0032] The consistency deviation analysis module 300 is used to load the virtual-real mapped wind field into the three-dimensional digital twin scene, drive the virtual drone to perform full-process flight simulation according to the preset logistics route, and perform consistency deviation analysis between the flight attitude response sequence of the virtual drone and the flight status data of the real drone during the simulation.

[0033] In the consistency deviation analysis module, loading the virtual-real mapped wind field into the three-dimensional digital twin scene and driving the virtual drone to perform a full-process flight simulation according to the preset logistics route is achieved through the following steps: The virtual-real mapped wind field is discretized and mapped according to the spatial grid nodes of the three-dimensional digital twin scene, and a wind field data query structure with the coordinate system of the three-dimensional digital twin scene is established. During the virtual drone flight simulation, the wind vector at the corresponding position is obtained by interpolation from the wind field data query structure based on the real-time position of the virtual drone on the preset logistics route. The wind vector is loaded as an aerodynamic external force input into the six-degree-of-freedom dynamic model of the virtual UAV, driving the virtual UAV to complete the entire flight simulation process from takeoff, cruise, obstacle avoidance descent to landing.

[0034] In specific implementation, firstly, the wind vectors of the unstructured computational grid nodes at each time step in the virtual-real mapped wind field are mapped to the regular spatial grid nodes corresponding to the basic terrain grid of the 3D digital twin scene through radial basis function interpolation. A wind field data query structure is established, indexed by a spatial octree and hierarchically by time sampling frames. Each leaf node of the wind field data query structure stores the three wind speed components and turbulent kinetic energy of the corresponding spatial volume element within the current time frame. The wind field data query structure refers to a hierarchical index data structure used for quickly retrieving wind vector data by spatial location and time frame. Then, during the virtual UAV flight simulation, based on the real-time position coordinates of the virtual UAV on the preset logistics route, the spatial volume element at that location is first located using the spatial octree index. Then, the wind vectors at the eight corner points of that spatial volume element are trilinearly interpolated, and linear time interpolation is performed between adjacent time frames in conjunction with the current simulation time to obtain the wind vectors at the corresponding location and time. The wind vector refers to a vector composed of three orthogonal wind speed components and turbulent kinetic energy. Finally, the wind vector is loaded as an aerodynamic external force input into the six-degree-of-freedom dynamic model of the virtual UAV. In the calculation of aerodynamic external forces in the six-degree-of-freedom dynamic model, the relative airspeed is first obtained by the difference between the wind vector and the current flight speed vector of the virtual UAV. The wind vector is then loaded as an aerodynamic environment input into the six-degree-of-freedom dynamic model of the virtual UAV. The relative airspeed is obtained by the difference between the wind vector and the current flight speed vector of the virtual UAV. Then, the aerodynamic force and torque are calculated by combining the relative airspeed with the aerodynamic parameters of the virtual UAV. The aerodynamic force and torque, along with the rotor thrust and gravity, are substituted into the six-degree-of-freedom rigid body motion equation for numerical integration, driving the virtual UAV to complete the entire flight simulation from takeoff, cruise, obstacle avoidance descent to landing. The six-degree-of-freedom dynamic model refers to the dynamic equation system describing the change of the UAV's motion state with time in six degrees of freedom: translation along three axes and rotation around three axes.

[0035] It should be noted that the full-process flight simulation in this application refers to a continuous simulation process in which, after loading a virtual-real mapped wind field into a three-dimensional digital twin scene, the virtual UAV's six-degree-of-freedom dynamic model is used as the core, and the complete flight stages, including takeoff, climb, cruise, obstacle avoidance descent, approach, and landing, are executed sequentially according to a preset logistics route. In each flight stage, the wind vector provided by the virtual-real mapped wind field is continuously used as an aerodynamic external force input for dynamic calculation. This process enables the wind field excitation experienced by the virtual UAV in the simulation to remain synchronized with the actual test flight process in terms of time series and spatial distribution. This ensures that the consistency deviation analysis between the virtual UAV's flight attitude response sequence and the actual UAV's flight state data is established under the same wind field input conditions, avoiding the introduction of false deviations due to inconsistent wind field boundary conditions, and improving the convergence accuracy and reliability of subsequent back-correction of aerodynamic parameters and flight control strategy models.

[0036] In the consistency deviation analysis module, the consistency deviation analysis between the flight attitude response sequence of the virtual UAV and the flight state data of the real UAV during the simulation is achieved through the following steps: By aligning the flight attitude response sequence of the virtual drone with the flight status data recorded by the real drone during test flights on the same preset logistics route according to the timestamp, a spatiotemporally aligned sample set is constructed. The time-frequency features of attitude angle, angular velocity, linear acceleration and airspeed are extracted from the virtual attitude response sequence and real flight state data in the same spatiotemporal aligned sample set, respectively, to obtain the virtual response feature set and the real flight feature set; Calculate the multi-scale dynamic time warping distance between the virtual response feature set and the real flight feature set under the same flight phase label, and map the multi-scale dynamic time warping distance as a consistency deviation index.

[0037] In practice, firstly, using the timestamp of each sampling point in the real UAV test flight data as a benchmark, the flight attitude response sequence of the virtual UAV is linearly interpolated and resampled according to the same timestamp, so that the two correspond one-to-one on the time axis. Then, a secondary verification is performed based on the spatial position of the track points on the preset logistics route, eliminating sampling points that cannot form a valid correspondence due to virtual UAV track deviations exceeding the preset spatial tolerance. This constructs a homogeneous spatiotemporally aligned sample set. The preset spatial tolerance is 1.5 times the wingspan of the real UAV, which is not limited here. A homogeneous spatiotemporally aligned sample set refers to pairs of samples with the same timestamp and the same track spatial position. The organization generates a set of virtual and real flight state data. Then, for the virtual attitude response sequence and real flight state data in the same spatiotemporally aligned sample set, each flight stage subsequence is sequentially extracted using a sliding time window of a preset length. For each subsequence, time-domain statistics and frequency-domain features of attitude angle, angular velocity, linear acceleration, and airspeed are extracted. The time-domain statistics include the mean and standard deviation. The frequency-domain features are obtained by extracting the power spectral density peak value and its corresponding frequency in the 0.5Hz to 10Hz frequency band after performing a fast Fourier transform on the subsequence. The time-frequency features of each flight stage subsequence are merged according to the flight stage label to obtain the virtual attitude response sequence. The virtual response feature set and the real flight feature set are used. The virtual response feature set refers to the ordered set of time-domain statistics and frequency-domain features of the virtual UAV under each flight stage label, while the real flight feature set refers to the ordered set of time-domain statistics and frequency-domain features of the real UAV under the same flight stage label. Finally, under the same flight stage label, feature vector sequences are constructed based on the time-domain statistics and frequency-domain features of four types of features (attitude angle, angular velocity, linear acceleration, and airspeed) between the virtual response feature set and the real flight feature set. For each feature vector sequence, the multi-scale dynamic time warping distance is calculated. The multi-scale dynamic time warping distance is calculated by first... The dynamic time warping distance is calculated at the sampling scale. Then, the feature vector sequence is downsampled stepwise by downsampling factors of 1 / 2 and 1 / 4 respectively. The dynamic time warping distance is then calculated again at the coarse scale. The dynamic time warping distances at the original scale, 1 / 2 scale and 1 / 4 scale are weighted and summed with weight coefficients of 0.5, 0.3 and 0.2 respectively to obtain the multi-scale dynamic time warping distance of this flight phase. Finally, the multi-scale dynamic time warping distances of all six flight phases are weighted and averaged according to the proportion of the duration of each flight phase to the total process time. Then, the averaged distances are normalized to the 0-1 interval by the maximum and minimum values ​​to obtain the consistency deviation index.

[0038] It should be noted that the consistency deviation index in this application refers to a comprehensive scalar used to quantitatively measure the difference in dynamic response between virtual and real drones under the same wind field conditions. This embodiment differs from the prior art in that it directly calculates point-by-point error or simple correlation coefficient on a single time scale using the flight status data of virtual and real drones. Instead, it first performs double alignment of the two data points by timestamp and spatial location and removes samples with excessive trajectory spatial deviation. Then, it uses a sliding time window to extract the mean and standard deviation of four features—attitude angle, angular velocity, linear acceleration, and airspeed—in the time domain, as well as the peak power spectral density in the frequency domain for each flight stage subsequence. Based on this, the dynamic time-warped distance of the feature vector sequence is calculated at the original scale, 1 / 2 scale and 1 / 4 scale respectively, and the sum is weighted by weight coefficients of 0.5, 0.3 and 0.2. Finally, the weighted average is normalized according to the proportion of the flight stage duration. This makes the consistency deviation index able to reflect the comprehensive deviation between virtual and real UAVs in terms of time domain statistical characteristics, frequency domain response characteristics, dynamic matching degree of multiple time scales and differences in the importance of different flight stages. It overcomes the shortcomings of existing technologies where a single distance metric or a single feature dimension cannot fully characterize the consistency of flight dynamic response under the excitation of complex mountain low-altitude wind fields.

[0039] The reverse calibration module 400 is used to reversely adjust the aerodynamic parameters and flight control strategy model of the virtual UAV based on the consistency deviation analysis results, until the dynamic response deviation between the virtual UAV and the real UAV under the same wind field conditions is less than a preset threshold.

[0040] In the reverse calibration module, the aerodynamic parameters and flight control strategy model of the virtual UAV are adjusted in reverse based on the consistency deviation analysis results until the dynamic response deviation between the virtual UAV and the real UAV under the same wind field conditions is less than a preset threshold. This is achieved through the following steps: Using the consistency deviation index as the fitness function, a global sensitivity analysis is performed on the aerodynamic parameters and adjustable parameters in the flight control strategy model of the virtual UAV, and a subset of key parameters whose contribution rate to the deviation exceeds a preset proportion is selected. Using the subset of key parameters as optimization variables and minimizing the consistency deviation index as the optimization objective, a Bayesian optimization strategy based on a surrogate model is used to iteratively correct the aerodynamic parameters and flight control strategy model. After each iteration and correction, the virtual drone is re-driven to perform a full-process flight simulation in the virtual-real mapped wind field and the consistency deviation index is recalculated until the consistency deviation index is less than the preset threshold.

[0041] In specific implementation, firstly, using the consistency deviation index as the fitness function, a global sensitivity analysis is performed on the aerodynamic parameters of the virtual UAV and the adjustable parameters in the flight control strategy model. The adjustable parameters include lift slope, drag coefficient, pitch moment derivative, rotor thrust coefficient, and attitude control gain, angular velocity damping coefficient, and trajectory tracking integral coefficient in the flight control strategy model. Using the variance-based Sobol global sensitivity analysis method, 512 uniformly distributed samples are generated for each adjustable parameter within a preset value range, and each sample is substituted into the corresponding consistency deviation index in the full-process flight simulation calculation. The independent and interactive contributions of each parameter to the consistency deviation index are decomposed using the Sobol first-order sensitivity index and the total sensitivity index. Parameters with a first-order sensitivity index or total sensitivity index exceeding 0.1 are selected to form a key parameter subset. This key parameter subset refers to the minimum combination of adjustable parameters that most significantly contribute to the dynamic response deviation between the virtual and real UAVs. Then, using this key parameter subset as the optimization variable and minimizing the consistency deviation index as the optimization objective, a Bayesian optimization strategy based on a surrogate model is used to iteratively correct the aerodynamic parameters and flight control strategy model. Initially, the parameters are... A Kriging surrogate model is constructed using 30 sets of key parameter subsets and their corresponding consistency deviation indices. Then, an expected improvement acquisition function is used to find candidate parameter points within the value space of the key parameter subsets that maximize the expected improvement. These candidate parameter points are substituted into the full-process flight simulation to calculate the actual consistency deviation index, and the Kriging surrogate model is updated. This process is iterated until the consistency deviation index is less than a preset threshold of 0.05, or the surrogate model predicts an optimal value with an improvement of less than 0.01 for five consecutive iterations and the current consistency deviation index no longer decreases. Iteration stops when the consistency deviation index is less than the preset threshold of 0.05, resulting in a virtual UAV aerodynamic parameters and flight control strategy model that meets the consistency requirements. Finally, after each iteration correction, the virtual UAV is re-driven to perform a full-process flight simulation in the virtual-real mapped wind field, and the consistency deviation index is recalculated. Iteration stops when the consistency deviation index is less than the preset threshold of 0.05, resulting in a virtual UAV aerodynamic parameters and flight control strategy model that meets the dynamic response deviation requirements of the real UAV under the same wind field conditions. The preset threshold refers to a pre-defined quantitative limit used to determine whether the dynamic response consistency between the virtual UAV and the real UAV meets the standard.

[0042] The working condition verification module 500 is used to inject artificially set extreme gust disturbances and mountain terrain-induced eddy current combination working conditions into the virtual-real mapped wind field, based on the adjusted aerodynamic parameters and flight control strategy model of the virtual UAV, to drive the virtual UAV to perform risk logistics flight simulation, and verify the safety margin of the preset logistics route under the combined working conditions during the simulation.

[0043] In the operational condition verification module, using the adjusted aerodynamic parameters and flight control strategy model of the virtual UAV as a baseline, artificially set extreme gust disturbances and mountain terrain-induced eddy current combinations are injected into the virtual-real mapped wind field to drive the virtual UAV to perform risk logistics flight simulation. This is achieved through the following steps: Identify the terrain-induced eddy high-incidence sections on the preset logistics route from the virtual and real mapped wind field, and superimpose artificially set extreme gust disturbance time histories on the terrain-induced eddy high-incidence sections to generate a combined disturbance wind field. The combined disturbed wind field is injected into the wind field data query structure of the three-dimensional digital twin scene to replace the original wind vector field of the terrain-induced eddy high-incidence section, so as to obtain a high-risk simulated wind field. Using the adjusted aerodynamic parameters and flight control strategy model of the virtual drone as a baseline, the high-risk simulated wind field is loaded, driving the virtual drone to perform continuous high-risk flight simulation from cruise, obstacle avoidance descent to landing according to the preset logistics route.

[0044] In specific implementation, firstly, high-incidence sections of terrain-induced eddies on the preset logistics route are identified from the virtual-real mapped wind field. The identification is based on the spatial distribution of time-averaged turbulent kinetic energy calculated on the time-averaged wind speed field of the virtual-real mapped wind field. Sections with time-averaged turbulent kinetic energy exceeding twice the average of the entire field, located on the leeward side of a mountain or at the exit of a canyon, and extending continuously for more than 30 meters along the longitudinal direction of the route are marked as high-incidence sections of terrain-induced eddies. At the same time, an artificially set extreme gust disturbance time history is superimposed on this section. The extreme gust disturbance time history is generated using a "1-cosine" gust model. The peak gust wind speed is taken as 1.5 times the average wind speed of this section in the virtual-real mapped wind field and not less than 12 m / s. The spatial scale of the gust is taken as twice the wingspan of the UAV. The duration of the gust is taken as 3 seconds. The gust direction is perpendicular to the prevailing wind direction of this section. The extreme gust disturbance time history is superimposed frame by frame onto the original wind vector field of the terrain-induced eddy high-incidence area to generate a combined disturbance wind field. This combined disturbance wind field refers to a wind field data volume for high-risk simulation testing formed by superimposing artificially set extreme gust disturbances onto a virtual-real mapped wind field. Then, the combined disturbance wind field is injected into the wind field data query structure of the 3D digital twin scene. Using the spatial octagonal leaf node corresponding to the spatial range of the terrain-induced eddy high-incidence area as the replacement unit, the wind vectors of each time frame in the original replacement unit are replaced one by one with the wind vectors of the corresponding spatial location and time frame of the combined disturbance wind field. An inverse distance is applied to the spatial transition zone between the replacement unit and the unreplaced area. After weighted smoothing, a high-risk simulated wind field is obtained. The high-risk simulated wind field refers to the complete simulated wind field data volume that has been replaced by combined disturbance wind fields and spatially smoothed in the wind field data query structure. Finally, using the adjusted aerodynamic parameters and flight control strategy model of the virtual UAV as the baseline, the high-risk simulated wind field is loaded, and the virtual UAV is driven to perform continuous high-risk flight simulation from cruise, obstacle avoidance descent to landing according to the preset logistics route. The continuous high-risk flight simulation does not interrupt the dynamic calculation when switching flight phases, and continuously advances the six-degree-of-freedom motion state of the virtual UAV with the same numerical integration step size as the full-process flight simulation, thereby obtaining the continuous flight state response of the virtual UAV under combined working conditions.

[0045] It should be noted that the high-risk logistics flight simulation in this application refers to replacing the original wind vector field in the terrain-induced eddy current high-incidence section of the virtual UAV with a combined disturbed wind field superimposed with artificially set extreme gust disturbance time histories, based on the aerodynamic parameters and flight control strategy model of the virtual UAV that has passed the virtual-real consistency calibration. The virtual UAV is then driven to perform a continuous flight dynamics calculation process from cruise, obstacle avoidance descent to landing in the high-risk simulated wind field according to the preset logistics route. The purpose is to assess the attitude stability, track maintenance capability, power reserve and near-ground safety boundary of the virtual UAV when performing logistics tasks under extreme wind conditions that are difficult to cover in real test flights, and to provide flight state data under extreme conditions for the safety margin assessment of the preset logistics route.

[0046] In the operating condition verification module, the safety margin of the preset logistics route under the combined operating conditions is verified during the simulation process using the following steps: During the high-risk logistics flight simulation, the attitude angle deviation, track deviation, and actuator margin of the virtual UAV under the combined disturbance wind field are extracted in real time to obtain the flight safety status parameters. The flight safety status quantity is compared with the preset logistics flight safety boundary conditions on a time-by-time basis, and the number of boundary violations and the duration of boundary violations are counted to obtain the safety boundary compliance degree. Using the safety boundary compliance as a quantitative basis, and combining the minimum track interval of the virtual drone in each segment of the preset logistics route, the safety margin of the preset logistics route under the combined operating conditions is calculated.

[0047] In specific implementation, firstly, during the high-risk logistics flight simulation, the flight safety state parameters of the virtual UAV under the combined disturbance wind field are extracted in real time using the same simulation step size as the dynamics calculation. These flight safety state parameters include attitude angle deviation, track deviation, and actuator margin. The attitude angle deviation is the difference between the virtual UAV's real-time pitch and roll angles and the expected attitude angles of the corresponding track points on the preset logistics route. The track deviation is the horizontal and vertical distance from the virtual UAV's real-time position to the centerline of the preset logistics route. The actuator margin is the ratio of the real-time output power of each motor of the virtual UAV to its rated maximum power, with the maximum power among the motors being the largest. The above three values ​​are combined at each simulation time step to form a flight safety state quantity. The flight safety state quantity refers to a real-time quantitative parameter set used to describe the degree to which the virtual UAV deviates from a safe flight state in a high-risk simulation. Then, the flight safety state quantity is compared time-by-time with preset logistics flight safety boundary conditions. The preset logistics flight safety boundary conditions are: attitude angle deviation not exceeding 15°, track deviation not exceeding half the width of a preset flight path corridor (with the corridor half-width being 5 meters), and actuator margin not exceeding 0.9. When any item in the flight safety state quantity exceeds the corresponding boundary at any time, that time is determined to be an out-of-bounds moment. The total value is then statistically analyzed. In a high-risk logistics flight simulation, the number of boundary violations and the duration of each violation are used to determine the boundary violation frequency. The ratio of the total number of boundary violations to the total number of simulation time steps is used as the boundary violation frequency, and the longest consecutive boundary violation duration is used as the maximum boundary violation duration. This yields the safety boundary compliance rate, which refers to the degree to which the flight safety state quantity, expressed as a pair of boundary violation frequency and maximum boundary violation duration, satisfies the preset safety boundary conditions. Finally, based on the safety boundary compliance rate and combined with the minimum track interval of the virtual UAV in each segment of the preset logistics route, the safety margin of the preset logistics route under the combined operating conditions is calculated. The minimum distance between the real-time position of the virtual UAV and the surface of the mountain and the surface of the artificial structure within each flight segment is taken as the interval. The cross-boundary frequency and the maximum cross-boundary duration in the safety boundary compliance are converted into a safety score between 0 and 1 by a linear mapping method. When the cross-boundary frequency is 0, the item scores 1 point, and when the cross-boundary frequency reaches 0.05, the item scores 0 points. When the maximum cross-boundary duration is 0, the item scores 1 point, and when it reaches 3 seconds, the item scores 0 points. The arithmetic mean of the two safety scores is multiplied by the ratio of the minimum flight path interval to the preset minimum safety interval. The preset minimum safety interval is 5 meters. The safety margin of the preset logistics route under the combined working conditions is obtained.

[0048] It should be noted that the safety margin in this application refers to a quantitative indicator used to determine whether a preset logistics route can be safely executed under extreme wind conditions, after comprehensively considering flight attitude safety and near-ground separation safety. This solution combines the attitude angle deviation, track deviation, and actuator margin extracted in real time from high-risk logistics flight simulation into a flight safety state quantity, and compares it with the preset logistics flight safety boundary conditions time by time. The safety boundary compliance is constructed using two indicators: the frequency of exceeding the boundary and the maximum duration of exceeding the boundary. The safety boundary compliance is then integrated with the minimum track separation of each segment to form a comprehensive safety margin indicator that simultaneously reflects four types of safety elements: attitude stability, track maintenance, power reserve, and near-ground separation. This allows for a more comprehensive exposure of the instantaneous and continuous risk of virtual UAVs exceeding the boundary on the preset logistics route under combined disturbed wind fields, avoiding the problem in existing technologies where the safety margin is judged based solely on a single extreme value, which may overlook the risk of continuous instability or track drift accumulation.

[0049] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0050] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A simulation and verification platform for low-altitude logistics in complex mountainous areas using unmanned aerial vehicles (UAVs), characterized in that: The platform includes: The twin scene construction module is used to construct a three-dimensional digital twin scene of the target mountainous area and to establish a low-altitude wind field numerical model containing micro-topographic disturbance features in the three-dimensional digital twin scene. The wind field dynamic assimilation module is used to dynamically assimilate and correct the low-altitude wind field numerical model by using the measured low-altitude wind field data of the actual UAV during the test flight of the preset logistics route in the target mountainous area as a constraint, so as to obtain a virtual-real mapped wind field that is synchronized with the real wind field. The consistency deviation analysis module is used to load the virtual-real mapped wind field into the three-dimensional digital twin scene, drive the virtual drone to perform full-process flight simulation according to the preset logistics route, and perform consistency deviation analysis between the flight attitude response sequence of the virtual drone and the flight status data of the real drone during the simulation process. The reverse calibration module is used to adjust the aerodynamic parameters and flight control strategy model of the virtual drone in reverse according to the consistency deviation analysis results, until the dynamic response deviation between the virtual drone and the real drone under the same wind field conditions is less than a preset threshold. The working condition verification module is used to inject artificially set extreme gust disturbances and mountain terrain-induced eddy currents into the virtual-real mapped wind field as a baseline using the adjusted aerodynamic parameters and flight control strategy model of the virtual UAV. This drives the virtual UAV to perform risk logistics flight simulation and verifies the safety margin of the preset logistics route under the combined working conditions during the simulation.

2. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the twin scene construction module, constructing a 3D digital twin scene of the target mountainous region specifically includes: Acquire multi-source geospatial data of the target mountainous region, register and fuse the multi-source geospatial data, and generate a basic terrain grid of the target mountainous region; The basic terrain grid is subjected to semantic segmentation and classification of land features, and the surface cover type of the mountain and the boundary of artificial structures are extracted to generate a terrain grid with semantic labels. Based on the terrain mesh with semantic tags, a 3D static scene model that is consistent with the geometric appearance of real terrain features is constructed. The surface reflectance and meteorological illumination data of the target mountainous area are obtained, and the material illumination mapping is performed on the three-dimensional static scene model to generate a three-dimensional digital twin scene with realistic textures.

3. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the twin scene construction module, establishing a low-altitude wind field numerical model containing micro-topographic disturbance features in the three-dimensional digital twin scene specifically includes: Extract the digital elevation model of the target mountainous area in the three-dimensional digital twin scene, and calculate the terrain slope, aspect and undulation of the digital elevation model to obtain the micro-topographic feature parameter field; Using the micro-topographic feature parameter field as the lower boundary condition, a computational fluid dynamics grid is constructed for the low-altitude logistics flight altitude layer. The computational fluid dynamics grid is locally refined in the near-wall and ridge regions of the mountain. Meteorological background wind field data of the target mountainous area is obtained, and the meteorological background wind field data is loaded into the computational fluid dynamics grid as the incoming flow boundary condition. The Reynolds-averaged Navier-Stokes equations are solved to obtain a low-altitude wind field numerical model that includes the characteristics of mountain flow around, canyon acceleration and leeward vortex.

4. The complex mountain UAV low-altitude logistics simulation and verification platform integrating virtual and real elements as described in claim 1, characterized in that, In the wind field dynamic assimilation module, the measured low-altitude wind field data during the test flight of the real UAV on the target mountainous area executing a preset logistics route specifically includes: An atmospheric data sensor is mounted on a real drone, and the installation error between the measurement reference of the atmospheric data sensor and the coordinate system of the real drone is calibrated to obtain the sensor calibration parameters. Control the real drone to perform test flights along the preset logistics route, and use the atmospheric data sensor to collect raw airflow parameters during the flight at a preset sampling frequency; The original airflow parameters were processed by body motion compensation and outlier removal to obtain measured low-altitude wind field data after removing interference from the UAV's own motion.

5. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the wind field dynamic assimilation module, the low-altitude wind field measurement data during the test flight of a real UAV executing a preset logistics route in the target mountainous area is used as a constraint to dynamically assimilate and correct the low-altitude wind field numerical model, resulting in a virtual-real mapped wind field that is synchronized with the real wind field. Specifically, this includes: Obtain the simulated wind field values ​​of the low-altitude wind field numerical model at the corresponding spatiotemporal points of the test flight route, and align the measured low-altitude wind field data with the simulated wind field values ​​in time and space to construct an assimilated observation increment field. Using the assimilated observation increment field as a constraint, a sequential assimilation correction based on ensemble Kalman filtering is applied to the state variables of the low-altitude wind field numerical model. In the assimilation process, a micro-topographic feature parameter field is introduced as a localized weighting factor for the error covariance to obtain the assimilated low-altitude wind field state variables. The assimilated low-altitude wind field state variables are subjected to time advancement and spatial interpolation to generate a virtual-real mapped wind field covering the entire area of ​​the preset logistics route.

6. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the consistency deviation analysis module, loading the virtual-real mapped wind field into the three-dimensional digital twin scene and driving the virtual drone to perform a full-process flight simulation according to the preset logistics route specifically includes: The virtual-real mapped wind field is discretized and mapped according to the spatial grid nodes of the three-dimensional digital twin scene, and a wind field data query structure with the coordinate system of the three-dimensional digital twin scene is established. During the virtual drone flight simulation, the wind vector at the corresponding position is obtained by interpolation from the wind field data query structure based on the real-time position of the virtual drone on the preset logistics route. The wind vector is loaded as an aerodynamic external force input into the six-degree-of-freedom dynamic model of the virtual UAV, driving the virtual UAV to complete the entire flight simulation process from takeoff, cruise, obstacle avoidance descent to landing.

7. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the consistency deviation analysis module, the consistency deviation analysis of the flight attitude response sequence of the virtual UAV and the flight state data of the real UAV during the simulation process specifically includes: By aligning the flight attitude response sequence of the virtual drone with the flight status data recorded by the real drone during test flights on the same preset logistics route according to the timestamp, a spatiotemporally aligned sample set is constructed. The time-frequency features of attitude angle, angular velocity, linear acceleration and airspeed are extracted from the virtual attitude response sequence and real flight state data in the same spatiotemporal aligned sample set, respectively, to obtain the virtual response feature set and the real flight feature set; Calculate the multi-scale dynamic time warping distance between the virtual response feature set and the real flight feature set under the same flight phase label, and map the multi-scale dynamic time warping distance as a consistency deviation index.

8. The virtual-real fusion simulation and verification platform for low-altitude logistics of complex mountain UAVs as described in claim 1, characterized in that, In the reverse calibration module, the aerodynamic parameters and flight control strategy model of the virtual UAV are adjusted in reverse according to the consistency deviation analysis results until the dynamic response deviation between the virtual UAV and the real UAV under the same wind field conditions is less than a preset threshold. Specifically, this includes: Using the consistency deviation index as the fitness function, a global sensitivity analysis is performed on the aerodynamic parameters and adjustable parameters in the flight control strategy model of the virtual UAV, and a subset of key parameters whose contribution rate to the deviation exceeds a preset proportion is selected. Using the subset of key parameters as optimization variables and minimizing the consistency deviation index as the optimization objective, a Bayesian optimization strategy based on a surrogate model is used to iteratively correct the aerodynamic parameters and flight control strategy model. After each iteration and correction, the virtual drone is re-driven to perform a full-process flight simulation in the virtual-real mapped wind field and the consistency deviation index is recalculated until the consistency deviation index is less than the preset threshold.

Images