A method and system for unmanned aerial vehicle path planning in strong wind environment

Abstract

The present application relates to the technical field of unmanned aerial vehicle path planning, and provides a kind of unmanned aerial vehicle path planning method and system under strong wind environment, including with a small amount of unmanned aerial vehicle cooperative perception as core, the wind speed vector difference measured by adjacent unmanned aerial vehicle is divided by the distance between machines to obtain shear edge, and is woven into wind field shear network, and discrete single-point measurement is converted into topological structure with spatial gradient constraint.Subsequently, the position of each machine is used as an anchor point to extrapolate along the incoming flow direction, and the inclination of the windward face and leeward face of the obstacle and the edge trend are extracted by means of radar point cloud to make local correction, to obtain the predicted wind field considering the continuity of incoming flow and topographic disturbance. Finally, according to the wind shear distribution, the positions exceeding the threshold value are connected as dangerous areas, which are mapped to the original route and the path points falling into the area are offset corrected to generate a safe flight path. Without pre-stored meteorological model and heavy calculation, a small amount of unmanned aerial vehicles can identify danger in real time and dynamically change course under strong wind.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) path planning technology, and more specifically, to a method and system for UAV path planning in strong wind environments. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] Unmanned aerial vehicle (UAV) path planning technology is a core component ensuring the safe and efficient execution of missions under complex weather conditions. Wind fields, as a key environmental factor affecting UAV flight stability and energy consumption, directly impact the reliability of path planning through their perception and prediction capabilities. In recent years, research on wind field information acquisition and utilization has primarily focused on two approaches.

[0004] One approach focuses on local wind field perception using a single UAV platform. By mounting wind speed and direction sensors on the aircraft, it measures the airflow conditions in a small area around the aircraft in real time and uses this information to avoid obstacles and adjust its attitude. This type of solution is simple in structure and has low deployment costs, and can cope with sudden local airflow disturbances to a certain extent. However, limited by the single-aircraft observation perspective, it is difficult to obtain information on the gradient distribution of the wind field in the spatial dimension, and it lacks the ability to predict the wind field change trends in areas not yet reached. Therefore, it has the limitation of slow response in path planning in strong wind environments.

[0005] Another approach focuses on large-scale UAV swarm systems. This involves using a dense sampling network of dozens of aircraft to conduct simultaneous multi-point observations of the target airspace, and then reconstructing a relatively complete wind field distribution through data fusion and fluid dynamics modeling. While this approach offers advantages in spatial coverage and reconstruction accuracy, it relies heavily on the aggregation and processing of massive amounts of sampled data, resulting in significant computational overhead. For real-time changing weather conditions, especially rapidly evolving hazardous airflow areas, it often struggles to complete identification and early warning within a limited timeframe. Furthermore, the organization and maintenance costs of large-scale swarms limit its applicability in small to medium-sized mission scenarios.

[0006] In summary, single-drone solutions cannot support spatial gradient inference due to insufficient observation dimensions, while the dense sampling logic used in large-scale clusters is difficult to directly transfer under extremely sparse sampling conditions. This results in either a lack of proactive identification of dangerous areas ahead when facing dynamic wind fields, or difficulty in timely correction of flight paths based on wind field topological characteristics. Therefore, how to identify dangerous areas in the wind field ahead and optimize routes using data from a small number of UAVs, without relying on pre-existing meteorological models or introducing heavy hydrodynamic calculations, has become a technical problem that requires further research in this field. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for unmanned aerial vehicle (UAV) path planning in strong wind environments to improve the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for UAV path planning in strong wind conditions, comprising multiple UAVs, each equipped with a wind speed and direction sensor and radar, the method comprising: S101, Obtain the target location, plan a first route based on the target location, send the first route to each drone, control each drone to take off and fly along the first route, and maintain the distance between adjacent drones within a preset range; S102 uses wind speed and direction sensors on each drone to collect wind speed and direction information at the location of each drone in real time, and uses radar to collect three-dimensional point cloud data within a preset range in front of each drone, and sends the wind speed and direction information and three-dimensional point cloud data to the terminal in real time. S103, for each pair of adjacent UAVs, the corresponding wind speed and wind direction information is decomposed into components along the direction of the line connecting the adjacent UAVs and components perpendicular to the direction of the line. The difference between the components along the line and the difference between the components perpendicular to the line are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent UAVs. The wind speed vector difference is divided by the distance between the adjacent UAVs to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent UAVs to form a shear edge. All shear edges constitute the wind field shear network. S104 performs surface reconstruction processing on the three-dimensional point cloud data collected by the radar, identifies the three-dimensional geometric shape of obstacles in the area in front, and extracts the windward tilt angle, leeward tilt angle and edge orientation of the corresponding obstacles from the three-dimensional geometric shape. S105: Using the real-time locations of each UAV as wind field sampling points, wind speed and direction information are collected. The wind field shear variables of each shear edge in the wind field shear network are used as wind field spatial gradient constraints. The unsampled area ahead is extrapolated along the direction of incoming flow indicated by the wind direction information to obtain the initial predicted wind field. Based on the windward tilt angle, leeward tilt angle, and edge orientation corresponding to different locations, the wind speed and direction of the area around the corresponding obstacle in the initial predicted wind field are corrected based on the preset wind simulation model to obtain the corrected predicted wind field. Based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance, the predicted wind field shear distribution in the area ahead is determined. S106, connect the locations in the predicted wind field shear distribution where the wind speed shear component exceeds the preset wind speed shear threshold or the wind direction shear component exceeds the preset angle to form a danger zone; map the danger zone onto the first route, and perform offset correction on the path points in the danger zone in the first route to generate a corrected route that avoids the danger zone.

[0008] Furthermore, the first route is planned based on the target location, specifically including: The takeoff position of each drone is obtained as the starting point of the path. The three-dimensional geometry is converted into a rasterized spatial occupancy representation. Each grid in the spatial occupancy representation corresponds to a three-dimensional space and marks whether the three-dimensional space is occupied by terrain or buildings. Using the grid cell where the path starts as the initial search node, the nodes of adjacent grid cells are expanded along the direction toward the target location. This includes: reading the corresponding occupancy mark for each candidate node obtained from the expansion, discarding the candidate nodes marked as occupied within a preset range, and retaining the remaining candidate nodes as the starting nodes for the next round of node expansion. Repeat the node expansion until the expanded node falls into the grid where the target location is located. The sequence of preserved nodes connected sequentially from the initial search node to the node that falls into the grid where the target location is located is used as the reference track. The reference track is offset according to the relative position of each UAV in the formation to generate the first route corresponding to each UAV.

[0009] Furthermore, the windward tilt angle, leeward tilt angle, and edge orientation of the corresponding obstacle are extracted from the three-dimensional geometry, specifically including: Based on the direction of the incoming flow indicated by wind speed and wind direction information, among the surfaces contained in the three-dimensional geometry, the surface with the normal pointing to the direction of the incoming flow is selected as the windward surface, and the surface with the normal pointing away from the direction of the incoming flow is selected as the leeward surface. The angle between the windward side and the horizontal plane is taken as the windward side inclination angle, and the angle between the leeward side and the horizontal plane is taken as the leeward side inclination angle; the intersection line between adjacent surfaces in the three-dimensional geometry is identified as the edge, and the projection direction of the edge on the horizontal plane is taken as the edge direction.

[0010] Furthermore, following the incoming flow direction indicated by the wind direction information, the unsampled area ahead is extrapolated to obtain the initial predicted wind field, which specifically includes: Using the real-time location of each drone as a wind field sampling point, the sampling is advanced segment by segment from each sampling point to the unsampled area in front, following the direction of the incoming flow indicated by the wind direction information. The influence propagation distance corresponding to each segment is determined based on the wind speed, and the wind speed is directly proportional to the influence propagation distance. For each unsampled location, a wind field sampling point upstream of that location and along the direction of the incoming flow is selected. The wind speeds collected by the selected wind field sampling points are attenuated according to their distances from the unsampled location and then superimposed to obtain the predicted wind speed at the unsampled location. The wind direction information along the direction of the incoming flow is assigned to the unsampled location as the predicted wind direction. The predicted wind speeds and predicted wind directions of all unsampled locations constitute the initial predicted wind field.

[0011] Furthermore, based on a preset wind simulation model, the wind speed and direction in the area surrounding the corresponding obstacle in the initial predicted wind field are corrected to obtain the corrected predicted wind field, specifically including: The grid points located within a preset range around the obstacle in the initial predicted wind field are taken as points to be corrected; for each point to be corrected, the wind speed vector of the point to be corrected in the initial predicted wind field is decomposed into tangential and normal components according to the orientation relationship between the point to be corrected and the surface of the obstacle. When any point to be corrected is on the windward side, the tangential component is deflected along the windward tilt angle, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the windward side and added to the deflected tangential component. When any point to be corrected is on the leeward side, the tangential component is deflected along the leeward tilt angle, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the corresponding leeward side, and then inverted and added to the deflected tangential component. When the point to be corrected is located in the edge region, the tangential component is deflected along the edge direction, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the edge and added to the deflected tangential component; the vector obtained by the addition is used as the corrected wind speed vector of the point to be corrected, and the direction of the wind speed vector is used as the corrected wind direction; the corrected wind field is composed of the corrected wind speed vectors and wind directions of all points to be corrected.

[0012] Furthermore, based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance, the predicted wind field shear distribution in the foreground area is determined, specifically including: The area in front of the corrected predicted wind field is divided into sequentially adjacent spatial grids. The center point of each spatial grid is taken as the wind field value location. For adjacent spatial grids, the corrected wind speed vector at the corresponding center point is read, and the difference between the corrected wind speed vectors is obtained. Divide the wind speed vector difference by the spatial distance between the two corresponding center points to obtain the wind field shear parameters of the adjacent spatial grids; attach the wind field shear parameters to the common edge between the adjacent spatial grids, and the wind field shear parameters on all the common edges constitute the predicted wind field shear distribution of the front area.

[0013] Furthermore, offset corrections are made for waypoints located within the danger zone in the first route, including: Traverse all path points on the first route and mark the path points that fall into the danger zone as target correction points; for each target correction point, read the predicted wind field shear distribution around the target correction point in the corrected predicted wind field, and determine the offset direction along the direction with the minimum wind field shear parameter. The offset distance is determined based on the distance from the target correction point to the boundary of the danger zone, and the target correction point is moved out of the danger zone along the offset direction; the offset path points are connected with adjacent path points in sequence to generate the correction route.

[0014] Secondly, this application also provides a UAV path planning system for strong wind environments, including: The path planning module is used to obtain the target location, plan a first route based on the target location, send the first route to each drone, control each drone to take off and fly along the first route, and maintain the distance between adjacent drones within a preset range. The data acquisition module is used to collect wind speed and direction information of each drone in real time through the wind speed and direction sensors of each drone, collect three-dimensional point cloud data of a preset range in front of each drone through radar, and send the wind speed and direction information and three-dimensional point cloud data to the terminal in real time. The shear network module is used to decompose the wind speed and wind direction information of each pair of adjacent UAVs into components along the direction of the line connecting the adjacent UAVs and components perpendicular to the direction of the line. The difference between the components along the line and the difference between the components perpendicular to the line are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent UAVs. The wind speed vector difference is divided by the distance between the adjacent UAVs to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent UAVs to form a shear edge. All shear edges constitute the wind field shear network. The obstacle recognition module is used to perform surface reconstruction processing on the three-dimensional point cloud data collected by radar, identify the three-dimensional geometric shape of obstacles in the area in front, and extract the windward tilt angle, leeward tilt angle and edge orientation of the corresponding obstacles from the three-dimensional geometric shape. The wind field prediction module collects wind speed and direction information using the real-time locations of each UAV as wind field sampling points. It uses the wind field shear variables of each shear edge in the wind field shear network as wind field spatial gradient constraints. It extrapolates the unsampled area ahead along the direction of incoming flow indicated by the wind direction information to obtain the initial predicted wind field. Based on the windward tilt angle, leeward tilt angle, and edge orientation corresponding to different locations, it corrects the wind speed and direction of the area around the corresponding obstacle in the initial predicted wind field based on a preset wind simulation model to obtain the corrected predicted wind field. Based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance, it determines the predicted wind field shear distribution in the area ahead. The path correction module is used to connect the locations in the predicted wind field shear distribution where the wind speed shear component exceeds a preset wind speed shear threshold or the wind direction shear component exceeds a preset angle to form a danger zone; map the danger zone onto the first route, and perform offset correction on the path points in the first route located in the danger zone to generate a corrected route that avoids the danger zone.

[0015] Thirdly, this application also provides an electronic device, including: Memory, used to store computer programs; A processor for implementing the method steps as described in the first aspect when executing the computer program.

[0016] Fourthly, this application also provides a readable storage medium storing a computer program that, when executed by a processor, implements the method steps of the first aspect.

[0017] The beneficial effects of this invention are as follows: This invention uses collaborative perception by a small number of UAVs as its core method. It constructs a wind field shear network by leveraging the wind speed differences among multiple UAVs, and combines this with radar three-dimensional geometric information to achieve wind field prediction and hazard avoidance in unsampled areas ahead.

[0018] Under conditions of extremely sparse sampling points, the vector difference between wind speed and direction measured by adjacent UAVs is first divided by the distance between the UAVs to obtain the shear edge characterizing the spatial rate of change of the wind field. All shear edges are then woven into a wind field shear network. This network transforms discrete single-point wind speed measurements into a topological structure with spatial gradient constraints, enabling a small number of UAVs to capture the spatial trend of wind field changes. This fills the gaps in the ability of a single UAV to acquire gradient information and the failure of clustered dense sampling logic on a small number of platforms.

[0019] Based on this, the scheme uses the real-time location of each UAV as the sampling anchor point, extrapolates the unsampled area in front along the direction of the incoming flow, and generates an initial predicted wind field. At the same time, it uses radar point cloud to reconstruct the three-dimensional geometry of the obstacle, extracts the windward tilt angle, leeward tilt angle and edge direction, and makes local corrections to the wind speed and wind direction in the area around the obstacle according to the preset wind simulation model, so as to obtain a corrected predicted wind field that takes into account both the continuity of the incoming flow and the terrain disturbance effect.

[0020] Finally, based on the ratio of the wind speed vector difference to the spatial distance between adjacent locations in the corrected predicted wind field, the predicted wind shear distribution is calculated. Locations where the wind speed or direction shear exceeds the threshold are connected as danger zones and mapped to the original planned route. Path points that fall into danger zones are offset and corrected to generate safe and flyable corrected routes. The entire method does not require pre-stored meteorological models or heavy fluid dynamics calculations; only six to seven UAVs are needed to identify danger zones ahead in real time and dynamically adjust flight paths in strong wind environments. Attached Figure Description

[0021] Figure 1 A flowchart of a UAV path planning method in a strong wind environment provided by the present invention; Figure 2 This is a schematic diagram of the initial predicted wind field in this invention; Figure 3This is a schematic diagram of the modified predicted wind field in this invention; Figure 4 This is a schematic diagram illustrating the prediction of wind shear distribution in this invention; Figure 5 This is a schematic diagram of generating the correction route in this invention; Figure 6 This is a schematic diagram of a drone path planning system for strong wind environments provided by the present invention.

[0022] In the diagram: 201, Path Planning Module; 202, Data Acquisition Module; 203, Shear Network Module; 204, Obstacle Identification Module; 205, Wind Field Prediction Module; 206, Path Correction Module. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a path planning method for unmanned aerial vehicles (UAVs) in strong wind conditions, comprising multiple UAVs, each equipped with a wind speed and direction sensor and radar. The method includes: S101: Obtain the target location, plan a first route based on the target location, send the first route to each drone, control each drone to take off and fly along the first route, and maintain the distance between adjacent drones within a preset range.

[0025] Specifically, a multi-UAV collaborative formation approach is adopted. The principle behind this approach is that a single UAV can only collect wind field information covering its own location. However, wind fields in strong wind environments exhibit significant spatial inhomogeneity. By maintaining the distance between adjacent UAVs within a preset range, the overall spatial sampling density of the wind field for the formation can be guaranteed, while avoiding excessive spacing that would lead to formation dispersion and the inability to construct a continuous wind field shear network. Taking a formation containing six UAVs as an example, the preset range can be set to 80 to 120 meters. This spacing satisfies the spatial resolution requirements for subsequent wind field gradient calculations and provides a reasonable spatial baseline for calculating the wind speed vector difference between adjacent UAVs in subsequent steps.

[0026] When planning the first route based on the target location, the takeoff positions of each UAV are first obtained as the starting points of the path. The 3D geometry is then converted into a gridded spatial occupancy representation. Each grid in the spatial occupancy representation corresponds to a 3D space and is marked as to whether that 3D space is occupied by terrain or buildings. The principle behind this step is that the flight space of the UAV is a continuous 3D space. Directly performing path search calculations in continuous space is computationally intensive and it is difficult to uniformly determine obstacles. However, by discretizing the 3D geometry (including static obstacles such as mountains, buildings, and bridges) between the takeoff area and the target location into cubic grids with uniform side lengths, the path planning problem in continuous space can be transformed into a node search problem in discrete grids. In specific implementation, the grid side length can be set to 1 meter, and each grid is marked with a binary symbol to indicate whether it is occupied. A mark of 1 indicates that the 3D space is occupied by terrain or buildings, and a mark of 0 indicates that the grid is a flyable airspace. When expanding subsequent nodes, collision detection can be completed simply by reading the occupancy mark of the corresponding grid, without performing complex 3D geometric intersection calculations, thus significantly improving the real-time performance of path planning.

[0027] After constructing the spatial occupancy representation, the grid cell where the path starts is used as the initial search node. Node expansion is performed on adjacent grid cells along the direction towards the target location. This includes: reading the corresponding occupancy marker for each expanded candidate node; discarding candidate nodes marked as occupied within a preset range; and retaining the remaining candidate nodes as the starting nodes for the next round of node expansion. The expansion is performed along the direction towards the target location, rather than indiscriminately expanding all neighboring grid cells. This is because a heuristic search strategy is introduced, prioritizing the expansion based on the direction of the grid cell containing the target location, thereby significantly reducing the number of invalid search nodes and shortening path planning time. When selecting candidate nodes, not only are grid cells marked as occupied discarded, but candidate nodes within a preset range around the occupied grid cells are also discarded. This preset range can be set to four grid cell sides. The principle behind this is that the drone itself has a certain physical size, and its flight attitude may deviate in strong winds. By reserving a safety margin, the planned path can be prevented from hugging the surface of obstacles, thus reducing the risk of collision. For example, if a candidate node is only three meters away from an occupied building grid, it will be discarded even if the candidate node itself is marked as unoccupied, in order to ensure that the drone and the building always maintain a safe distance of no less than five meters.

[0028] The node expansion is then repeated until the expanded nodes fall into the grid of the target location. The sequence of nodes connected sequentially from the initial search node to the node falling into the grid of the target location is used as the reference track. The reference track is offset according to the relative position of each UAV in the formation to generate the first route corresponding to each UAV. The reference track represents the flight path of the formation center, and the UAVs in the formation need to maintain a preset relative position relationship. For example, the lead UAV is located in the center of the formation, and the other four UAVs are located 100 meters in front, behind, to the left, and to the right of the lead UAV, respectively. Therefore, by translating the reference track along the relative position vector of each UAV, the first route of each UAV can be obtained. The advantage of this generation method is that it ensures that the formation as a whole moves along the same reference track, and that the flight paths of each UAV are parallel and the spacing is stable. This lays the spatial foundation for the synchronous collection of wind field information by each UAV at different spatial positions in subsequent steps, and also ensures that the distance between adjacent UAVs can be maintained within a preset range.

[0029] S102 uses wind speed and direction sensors on each drone to collect wind speed and direction information at the location of each drone in real time, and uses radar to collect three-dimensional point cloud data within a preset range in front of each drone, and sends the wind speed and direction information and three-dimensional point cloud data to the terminal in real time.

[0030] Specifically, each drone is equipped with an ultrasonic three-dimensional wind speed and direction sensor. Its principle lies in utilizing the characteristic that the speed of ultrasound in the air varies with wind speed. Two pairs of orthogonally arranged ultrasonic transducers measure the wind speed components in two orthogonal directions on the horizontal plane, and a vertical transducer measures the vertical wind speed component, thus synthesizing the magnitude and direction of the three-dimensional wind speed vector. The sensor's sampling frequency is set to 10 Hz, meaning it collects wind speed and direction information every 100 milliseconds.

[0031] During the data collection process, each drone only collects wind speed and direction information at its own location. This is because the measurement range of a single sensor is limited, reflecting only the local wind field state at its spatial point. However, by having multiple drones in a formation simultaneously collect data at different spatial locations, distributed wind field sampling data covering a certain spatial area can be obtained, thus compensating for the insufficient spatial sampling capability of a single drone. Taking a formation of six drones as an example, each drone simultaneously collects wind speed and direction information at six different spatial locations. The collected information includes wind speed magnitude, horizontal wind direction angle, and vertical tilt angle. These data collectively form the basic input for calculating the wind speed vector difference between adjacent drones in subsequent steps.

[0032] Simultaneously, the radars mounted on each UAV collect 3D point cloud data within a preset range in front of each UAV. Specifically, millimeter-wave 3D imaging radar is used. The principle behind this is that millimeter-wave radar can transmit frequency-modulated continuous waves and receive echoes. Based on the echo delay and Doppler shift, it calculates the target's range, velocity, and angle information. Then, it measures the target's elevation angle using the phase difference of a multi-channel receiving array, thereby generating point cloud data containing 3D coordinate information including range, azimuth, and elevation angles. The radar's scanning frequency is set to five hertz, meaning it completes a 3D point cloud acquisition of the forward area every 200 milliseconds, obtaining 3D coordinate data for no fewer than two thousand spatial points each time.

[0033] After completing the above data collection, wind speed and direction information, along with 3D point cloud data, are transmitted to the terminal in real time. Specifically, each UAV establishes a communication link with the terminal via its onboard wireless communication module using time-division multiple access. During data transmission, the wind speed and direction information of each UAV, along with the 3D point cloud data, are encapsulated in a unified data packet format. The data packet includes the UAV number, data collection timestamp, wind speed value, wind direction angle, vertical tilt angle, and a point cloud data array. Upon receiving the data packet, the terminal performs alignment processing based on the timestamp, ensuring that data collected by each UAV at the same time can be matched.

[0034] S103, for each pair of adjacent UAVs, the corresponding wind speed and wind direction information is decomposed into components along the direction of the line connecting the adjacent UAVs and components perpendicular to the direction of the line. The difference between the components along the line and the difference between the components perpendicular to the line are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent UAVs. The wind speed vector difference is divided by the distance between the adjacent UAVs to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent UAVs to form a shear edge. All shear edges constitute the wind field shear network.

[0035] Specifically, after receiving real-time wind speed and direction information from each drone, the terminal first identifies all adjacent drone pairs based on the real-time position coordinates of each drone. The principle is that wind shear essentially reflects the rate of change of wind field physical quantities in space. Only the wind speed difference between spatially adjacent sampling points has the physical meaning of a local gradient. If drone pairs that are too far apart are selected for calculation, the resulting difference will be mixed with large-scale wind field changes and will not be able to characterize local shear features. Taking a formation of six drones as an example, the six drones are positioned in the front, back, left, right, up, and down directions, and any one drone can act as the host. Adjacency is determined based on whether the real-time distance between the drones falls within a preset range. Each drone forms an adjacent pair with four drones in its orthogonal direction, and does not form an adjacent pair with drones positioned in the opposite direction if the distance exceeds the preset range, resulting in a total of twelve pairs of adjacent drone combinations.

[0036] For each pair of adjacent UAVs, the terminal first calculates the line vector connecting the two UAVs based on their real-time position coordinates, and then normalizes this line vector to obtain the unit direction along the connecting line, which is expressed as:

[0037]

[0038]

[0039] In the formula, For drones Pointing to drones The connecting vector; , drones and drones 3D position vector; , , For drones 3D coordinates; , , Let J be the three-dimensional coordinates of UAV j; For adjacent drones Spatial distance between j; For drones arrive A three-dimensional unit vector in the direction of the line.

[0040] Next, construct a unit direction perpendicular to the connecting line in the horizontal plane, its expression is:

[0041] In the formula, It is a unit vector constructed in the horizontal plane and perpendicular to the horizontal projection direction of the connecting line; The X component of the unit vector along the direction of the connecting line; The Y component is the unit vector along the direction of the connecting line; the denominator is the magnitude of the horizontal projection of the connecting line, used for normalization.

[0042] The principle is that any vector in a plane can be uniquely decomposed along two orthogonal directions. By projecting the wind speed vector onto these two orthogonal directions—along and perpendicular to the connecting line—the rate of change of the wind field along the formation baseline and perpendicular to the baseline can be examined separately. The physical meanings of these two directions are clear and uncoupled. Subsequently, the wind speed vector for each UAV, synthesized from wind speed magnitude and wind direction angle, is projected onto both the direction along and perpendicular to the connecting line, yielding the component along and perpendicular to the connecting line for each UAV. The component along the connecting line directly reflects the stretching or compressing effect of the wind field along the line connecting the two UAVs, while the component perpendicular to the connecting line reflects the shearing effect of the wind field along the baseline. These correspond to the normal strain rate and shear strain rate in fluid mechanics, respectively, providing a basis for the subsequent physical attribution of wind field shear.

[0043] The expression for the component along the connecting line direction is:

[0044] The expression for the direction component of the perpendicular connection is:

[0045] In the formula, For drones The projection component of the wind speed vector along the connecting line reflects the stretching or compression effect of the wind field along the formation baseline. For drones The projection component of the wind speed vector in the vertical line direction, i.e., the horizontal plane, reflects the shearing effect of the wind field along the baseline. For drones The three-dimensional wind speed vector; , , For drones The wind speed vector has three components: X, Y, and Z. It is a unit vector along the direction of the connecting line; It is a unit vector in the horizontal plane perpendicular to the direction of the connecting line, and its Z component is zero; , , These are the three components of the unit vector along the direction of the connecting line; , These are the X and Y components of the unit vector in the direction of the perpendicular line.

[0046] After component decomposition, the terminal calculates the difference between the components along the connecting line and the difference between the components perpendicular to the connecting line between adjacent UAVs of the same pair, expressed as follows:

[0047]

[0048] In the formula, For adjacent drones and The difference in wind speed components along the direction of the connecting line; For adjacent drones and The difference in wind speed components along the perpendicular line connecting the two sides; , drones and drones The projected component of the wind speed along the direction of the line connecting them; , drones and drones The projected component of the wind speed along the vertical line.

[0049] The principle is that the component differences in the two orthogonal directions represent the spatial variation of the wind field in those two directions, respectively. Calculating them separately avoids the directional information being masked by the scalar difference when directly subtracting the original wind speed vector. In practice, the component along the connecting line of the second UAV is subtracted from the component along the connecting line of the first UAV to obtain the difference in the components along the connecting line; similarly, the component perpendicular to the connecting line of the second UAV is subtracted from the component perpendicular to the connecting line of the first UAV to obtain the difference in the components perpendicular to the connecting line.

[0050] The two differences are then combined to obtain the wind speed vector difference between adjacent drones. The principle is that the direction along the connecting line and the direction perpendicular to the connecting line are orthogonal, and the difference vectors in these two directions are independent. According to the parallelogram law of vector composition, they can be combined into a single resultant vector, the magnitude of which represents the overall difference in wind speed vectors between the locations of the two drones. In practice, the difference in the component along the connecting line and the difference in the component perpendicular to the connecting line are used as coordinate values ​​in two orthogonal directions, and the magnitude of the wind speed vector difference is calculated using the Pythagorean theorem. The advantage of this synthesis method is that the resulting wind speed vector difference integrates the changes in the wind field in two orthogonal directions, comprehensively characterizing the spatial non-uniformity of the wind field between adjacent drones.

[0051] After obtaining the wind speed vector difference, divide the wind speed vector difference by the distance between adjacent UAVs to obtain the corresponding wind field shear variable, the expression of which is:

[0052] In the formula, For adjacent drones and The wind field shear variable on the line connecting them is physically defined as the change in the wind speed vector over a unit spatial distance. For adjacent drones and The magnitude of the vector difference in wind speed between them; For adjacent drones and The spatial distance between them.

[0053] The principle is that wind shear is physically defined as the change in wind speed vector over a unit spatial distance. The absolute value of the wind speed vector difference alone cannot determine the severity of the shear; it needs to be normalized by dividing by the spatial distance between the two sampling points to obtain shear parameters with spatial gradient dimensions. The distance between adjacent UAVs is calculated from their real-time position coordinates, and this distance remains stable within the range of 80 to 120 meters while maintaining a preset formation spacing.

[0054] Wind field shear variables are attached to the lines connecting adjacent drones to form shear edges. The principle is that the lines connecting adjacent drones represent the spatial baseline between wind field sampling points, and the wind field shear variable is a physical quantity calculated along this baseline. Attaching it to the lines allows the shear information to be directly correlated with spatial location. In practice, the real-time positions of the two drones are used as the two endpoints of the edge, and the wind field shear variable is used as the weight attribute of the edge, thus constructing a shear edge with shear weights.

[0055] Finally, all shear edges form a wind field shear network. The principle is that multiple UAVs in the formation form a distributed sampling topology in space. Each pair of adjacent UAVs corresponds to a shear edge, and all shear edges are interconnected with UAV positions as nodes, collectively forming a graph network covering the spatial region where the formation is located. Each node in this network represents the real-time wind field sampling position of a UAV, and each edge represents the wind field shear intensity on that spatial baseline. The wind field shear network organizes discrete multi-UAV sampling data into structured data with spatial topological relationships, visually presenting the spatial distribution of wind field shear around the formation. This provides wind field spatial gradient constraints for extrapolating unsampled areas along the incoming flow direction in subsequent steps.

[0056] S104 performs surface reconstruction processing on the three-dimensional point cloud data collected by the radar, identifies the three-dimensional geometric shape of obstacles in the area in front, and extracts the windward tilt angle, leeward tilt angle and edge orientation of the corresponding obstacles from the three-dimensional geometric shape.

[0057] Specifically, after receiving the 3D point cloud data sent by the radars of each UAV, the terminal first preprocesses the point cloud data. In practice, statistical filtering is used to remove outliers caused by radar multipath reflections and noise interference. After preprocessing, a neighborhood-based principal component analysis method is used to estimate the normal vector for each point. Specifically, for each point, its nearest neighbors within a radius are selected, and the surface normal direction at that point is obtained through covariance matrix decomposition; its expression is as follows:

[0058]

[0059] In the formula, For point The 3x3 covariance matrix of the point set within the neighborhood; For point The total number of nearest neighbors within the neighborhood; For point Within the neighborhood The three-dimensional position vectors of the nearest neighbor points; This is the three-dimensional position mean vector of all nearest neighbor points in the neighborhood; This is the transpose of the vector difference, used to construct the outer product matrix; For point The surface unit normal vector at that location is the eigenvector corresponding to the smallest eigenvalue of the covariance matrix; The smallest eigenvalue of the covariance matrix corresponds to the direction with the smallest variance in the tangent plane of the surface, i.e., the normal.

[0060] The principle is that the local neighborhood of each point in the point cloud can be approximated as a tiny tangent plane. The normal vector of the tangent plane represents the orientation of the surface where the point is located. Estimating the normal vector is a prerequisite for distinguishing the windward and leeward sides. Subsequently, the Poisson surface reconstruction algorithm is used to convert the point cloud with normal vector information into a continuous triangular mesh surface. The principle is that by constructing an indicator function and solving the Poisson equation, the discrete point cloud can be interpolated into a closed and topologically consistent three-dimensional surface model, thereby obtaining the complete three-dimensional geometry of the obstacle.

[0061] Specifically, extracting the windward tilt angle, leeward tilt angle, and edge orientation of the corresponding obstacle from the three-dimensional geometry includes: Based on the incoming flow direction indicated by wind speed and direction information, among the surfaces contained in the three-dimensional geometry, the surface with its normal pointing towards the incoming flow direction is selected as the windward surface, and the surface with its normal pointing away from the incoming flow direction is selected as the leeward surface. In specific implementation, firstly, based on the wind speed and direction information collected by the wind speed and direction sensors of each UAV at the current moment, the average incoming flow direction vector of the formation's location is synthesized; the expression for the average incoming flow direction vector is:

[0062] In the formula, This is the average incoming flow direction unit vector for the area where the formation is located; For drones The three-dimensional wind speed vector at the location; The total number of drones in the formation. The numerator is the vector sum of the wind speed vectors of each drone, and the denominator is the magnitude of the vector sum, used for normalization to obtain the unit direction.

[0063] Subsequently, for each triangular mesh facet contained in the 3D geometry, its normal vector is read and the angle between this normal vector and the incoming flow direction vector is calculated. If the angle is less than 90 degrees, the facet's normal is determined to point towards the incoming flow direction, and it is included in the windward candidate set. If the angle is greater than 90 degrees, the facet's normal is determined to point away from the incoming flow direction, and it is included in the leeward candidate set. The expression for windward / leeward determination is:

[0064] In the formula, The angle between the normal vector of the k-th triangular mesh facet and the average incoming flow direction vector; For the first The unit normal vector of a triangular mesh facet; This is the average incoming flow direction unit vector for the area where the formation is located; The dot product of two unit vectors, with values ​​ranging from -1 to 1; when the included angle is less than 90 degrees, the surface belongs to the candidate set of the windward side, and when the included angle is greater than 90 degrees, the surface belongs to the candidate set of the leeward side.

[0065] The principle behind this method is that the angle between the surface normal and the incoming flow direction directly reflects the degree to which the surface faces the airflow. An angle less than 90 degrees means the airflow is directly impacting the surface, which is considered a windward receiving area. An angle greater than 90 degrees means the airflow is flowing over the back of the surface, which is considered a leeward bypass area. This screening method can automatically and dynamically determine the windward and leeward sides based on the real-time incoming flow direction, eliminating the need for pre-marking of obstacle orientations and adapting to the frequent changes in incoming flow direction in strong wind environments. Taking a pitched roof building as an example, if the incoming flow direction is from west to east, the angle between the normal of the west wall and the west pitched roof and the incoming flow direction is less than 90 degrees, and they are screened as windward sides. The angle between the normal of the east wall and the east pitched roof and the incoming flow direction is greater than 90 degrees, and they are screened as leeward sides.

[0066] Then, the angle between the windward side and the horizontal plane is calculated as the windward tilt angle, and the angle between the leeward side and the horizontal plane is calculated as the leeward tilt angle. In practice, firstly, all triangular mesh faces belonging to the candidate set of the windward side are merged through region growing. Faces with similar normal directions and spatial adjacency are merged into the same windward plane, and then the plane equation of the windward side is obtained through plane fitting. Similarly, the candidate set of the leeward side is merged and fitted to obtain the plane equation of the leeward side. Then, the angle between the normal vector of the windward plane and the vertical unit vector is calculated, and this angle is subtracted from 90 degrees to obtain the angle between the windward side and the horizontal plane, which is the windward tilt angle. The leeward tilt angle is calculated using the same method. The principle is that the angle between the plane and the horizontal plane is equal to the complementary angle between the plane normal vector and the vertical direction. Solving indirectly through the normal vector avoids errors caused by surface unevenness when directly measuring the plane tilt angle.

[0067] The expression for the angle of inclination of the windward side is:

[0068] The expression for the leeward slope angle is:

[0069] In the formula, The angle between the windward side and the horizontal plane is called the windward side tilt angle. The angle between the leeward side and the horizontal plane is called the leeward side inclination angle. The unit normal vector of the windward plane is obtained by region growing and merging of candidate windward plane patches and fitting it to the plane. The unit normal vector of the leeward plane is obtained by region growing and merging of candidate leeward plane patches and fitting with the plane. It is a vertically upward unit vector; The dot product of the windward normal vector and the vertical direction is equal to the cosine of the angle between them.

[0070] The intersection lines between adjacent surfaces in a 3D geometry are identified as edges, and the projection direction of these edges onto the horizontal plane is determined as the edge direction. Specifically, all triangular mesh edges in the 3D geometry are traversed. When an edge is adjacent to two faces belonging to different surface planes and the angle between the normal vectors of the two planes exceeds a preset angle threshold, the edge is marked as a candidate edge. Spatial connections and line fitting are performed on all candidate edges to obtain complete edge spatial segments. Then, the two endpoints of these edge spatial segments are projected onto the horizontal plane, and connecting the two projection points yields the projection line segment of the edge onto the horizontal plane. The direction indicated by this projection line segment is the edge direction, expressed as follows:

[0071] In the formula, The edge orientation angle is the angle between the projection line segment of the edge onto the horizontal plane and the positive direction of the X-axis, with counterclockwise being positive. , The horizontal coordinate of the first endpoint of the edge space segment; , Let be the horizontal coordinate of the second endpoint of the edge space segment; It is a two-parameter arctangent function, which automatically determines the quadrant based on the signs of X and Y.

[0072] The principle behind this method lies in the fact that the intersection between adjacent surfaces is the location where the geometry of an obstacle undergoes a sudden change in direction. When airflow passes over an edge, it generates significant flow around and vortex shedding. The projection direction of the edge onto the horizontal plane determines the direction of the airflow's lateral deflection along the edge. This identification method can automatically extract all sharp edges on an obstacle that significantly affect the wind field distribution and quantify their spatial orientation, providing a directional basis for subsequent steps to correct the tangential deflection of the wind speed vector in the edge region. Taking the aforementioned pitched roof building as an example, the intersection between the western pitched roof surface and the western wall of the building is identified as an edge. Its projection direction on the horizontal plane is north-south, so the direction of this edge is north-south, and the airflow will undergo tangential deflection along the north-south direction when passing over this edge.

[0073] S105: Using the real-time locations of each UAV as wind field sampling points, wind speed and direction information are collected. The wind field shear variables of each shear edge in the wind field shear network are used as wind field spatial gradient constraints. The unsampled area ahead is extrapolated along the direction of incoming flow indicated by the wind direction information to obtain the initial predicted wind field. Based on the windward tilt angle, leeward tilt angle and edge orientation corresponding to different locations, the wind speed and direction of the area around the corresponding obstacle in the initial predicted wind field are corrected based on the preset wind simulation model to obtain the corrected predicted wind field. Based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance, the predicted wind field shear distribution in the area ahead is determined.

[0074] Specifically, after completing the construction of the wind field shear network and the extraction of obstacle geometric features, the terminal enters the wind field prediction stage for the unsampled area ahead. The principle is that the UAV formation can only acquire measured wind field data at its current location, while areas yet to be reached lack direct observation. Flying solely based on a baseline trajectory could lead to sudden wind shear and attitude instability. Therefore, it is necessary to predict the wind field ahead based on the wind field distribution patterns of the sampled areas and the geometric features of obstacles. This step sequentially includes three sub-processes: initial predicted wind field extrapolation, wind field correction around obstacles, and determination of the predicted wind field shear distribution, forming a coarse-to-fine wind field prediction chain.

[0075] First, an extrapolation of the initial predicted wind field is performed. Using the real-time locations of each UAV as wind field sampling points, the sampling is performed segment by segment forward from each sampling point towards the unsampled area, following the direction of the incoming wind flow indicated by the wind direction information. The propagation distance of the influence corresponding to each segment is determined based on the wind speed, where wind speed is directly proportional to the propagation distance. The expression is:

[0076] In the formula, For drones The influence distance of the wind field information at the sampling point propagating downstream along the direction of the incoming flow; This is a preset scaling factor, with the dimension of time, used to convert wind speed into propagation distance, and can be set according to the turbulence characteristics of the wind field. For drones The wind speed at the location. This formula reflects the physical law that the greater the wind speed, the greater the airflow, and the wider the spatial range of wind field characteristics transmitted along the flow direction.

[0077] In practice, the terminal reads the real-time position coordinates, wind speed, and wind direction information of each UAV, marking the spatial position of each UAV as a wind field sampling point. Simultaneously, it reads the wind field shear variables of each shear edge in the wind field shear network as constraint values ​​for the spatial gradient of the wind field between adjacent sampling points. The principle is that the wind field shear variable of the shear edge represents the rate of change of the wind speed vector on that baseline segment; using it as a gradient constraint allows the extrapolation process to utilize the spatial variation trend of adjacent sampling points, improving the spatial continuity of the extrapolation results. Subsequently, the terminal synthesizes the average incoming flow direction of the formation area based on the wind direction information of each UAV, using this direction as the extrapolation advancement direction, advancing segment by segment from each sampling point. During the advancement, the influence propagation distance is determined based on the wind speed at the current sampling point; the higher the wind speed, the farther the influence of the wind field information at that point propagates downstream. The principle is that higher wind speed means greater airflow, and the wider the spatial range of wind field characteristics transmitted along the flow direction. In practice, the influence propagation distance can be set as the product of the wind speed and a preset proportional coefficient, allowing the extrapolation range to dynamically adapt to the wind speed.

[0078] For each unsampled location, wind field sampling points upstream of that location and along the incoming flow direction are selected. The wind velocities collected by the selected sampling points are attenuated according to their distances from the unsampled location and then superimposed to obtain the predicted wind velocity for that unsampled location. The wind direction information along the incoming flow direction is assigned to the unsampled location as the predicted wind direction. The predicted wind velocities and predicted wind directions of all unsampled locations constitute the initial predicted wind field, as shown below. Figure 2As shown in the diagram. In practice, the terminal discretizes the unsampled area ahead into a three-dimensional grid consistent with the spatial occupancy representation, with the center point of each grid representing an unsampled location. For each unsampled location, sampling points located upstream and whose projected distance along the incoming flow direction is within the influence propagation distance are first selected. The principle is that only sampling points upstream and whose influence range covers the location have a real contribution to the wind field at that location. Subsequently, the spatial distance between each selected sampling point and the unsampled location is calculated. The wind speed is attenuated using a distance attenuation function and then weighted and superimposed. The superposition weight can be normalized according to the wind field shear edge variable corresponding to each sampling point, thereby obtaining the predicted wind speed. The principle is that during the propagation of the wind field along the flow direction, the wind speed amplitude gradually decreases with increasing distance due to turbulent dissipation. The superposition after attenuation can simulate this physical process, making the predicted wind speed spatially smooth. The wind direction is directly assigned to the unsampled location along the incoming flow direction. The principle is that in open areas without obstruction interference, the wind direction remains relatively stable along the flow direction. The initial predicted wind field is formed by the predicted wind speed and predicted wind direction at all unsampled locations, providing a base field for subsequent fine-tuning around obstacles.

[0079] The expression for the distance decay function is as follows:

[0080] In the formula, This is the value of the distance decay function; This represents the spatial distance between the sampling point and the unsampled location. The wind field attenuation characteristic length represents the distance that the wind speed amplitude travels when it attenuates to the negative first power of its original value e along the propagation direction, and is determined by the intensity of turbulent dissipation. It is a natural exponential function. This formula simulates the physical process by which the wind speed amplitude decreases exponentially with increasing distance during the propagation of the wind field along the flow direction due to turbulent dissipation.

[0081] The expression for predicted wind speed at unsampled locations is:

[0082] The upstream sampling point selection criteria are as follows:

[0083] In the formula, Unsampled locations The predicted three-dimensional wind speed vector at the location; The three-dimensional position vector for the unsampled location; For position The set of contributing upstream sampling points; For drones The normalized weight of the corresponding sampling point can be determined based on the shear variable of the wind field shear edge to which the sampling point belongs; Sampling points The distance decay function value to the unsampled location; Unsampled locations and sampling points Spatial distance between them; For drones sampling points 3D position vector; Sampling points The measured three-dimensional wind speed vector at the location; The average incoming flow direction is represented by a unit vector. Sampling points The influence propagation distance. The denominator is a normalization factor to ensure that the sum of the weights of each sampling point is one, so that the predicted wind speed is the weighted average of the wind speeds of each upstream sampling point. The first item in the screening criteria determines the sampling point. Located upstream of the unsampled location, the second determination determines that the influence range of the sampling point covers the unsampled location.

[0084] Next, the initial predicted wind field is corrected based on the preset wind simulation model, including: The first step involves selecting grid points within a predetermined range around the obstacle in the initial predicted wind field as points to be corrected. For each point, based on its orientation relative to the obstacle surface, the wind speed vector at that point in the initial predicted wind field is decomposed into tangential and normal components. The expression is as follows:

[0085]

[0086]

[0087] In the formula, This represents the projection component of the wind speed vector at the point to be corrected onto the tangential direction of the surface. This represents the projection component of the wind speed vector at the point to be corrected onto the surface normal direction. Points to be corrected The initial predicted three-dimensional wind speed vector at the location; The three-dimensional position vector of the point to be corrected; It is the tangential unit vector of the nearest surface of the obstacle at the point to be corrected, and lies in the tangential plane of the surface; It is the outward normal unit vector of the nearest surface of the obstacle at the point to be corrected, perpendicular to the tangent plane of the surface. and They are mutually orthogonal, forming a two-dimensional orthogonal basis, and any vector located in this plane can be uniquely decomposed.

[0088] In practice, the terminal determines the spatial boundary of a preset range around the obstacle based on the three-dimensional geometry of the obstacle identified by S104. This range can be set as an area extending outward from the outer envelope of the obstacle by twenty meters. The principle is that the disturbance of the wind field by the obstacle is mainly concentrated within a certain distance from the wall surface. Beyond this range, the wind field has returned to its inflow state and does not require correction. Grid points falling within this range are marked as points to be corrected. For each point to be corrected, the tangential and normal directions are determined based on its orientation relationship with the nearest surface of the obstacle. The wind speed vector is decomposed by projection along the two orthogonal directions. The principle is that the disturbance of the airflow by the obstacle manifests as tangential wall-attached deflection and normal stagnation or suction. After decomposition, the two effects can be modeled independently to avoid coupling errors.

[0089] The second step is to deflect the tangential component along the windward side when any point to be corrected is located on the windward side. Then, the modulus of the normal component is reduced by the distance between the point to be corrected and the windward side, and added to the deflected tangential component. The expression is:

[0090]

[0091] The tangential deflection direction is:

[0092] In the formula, Points to be corrected The corrected three-dimensional wind speed vector; To correct the tangential component of the wind speed vector; To correct the normal component of the wind speed vector; It is the tangential unit vector after deflection along the tilt angle of the windward surface, and is parallel to the windward plane; The outward normal unit vector of the windward side; The normal distance from the point to be corrected to the windward side; The characteristic attenuation distance of the windward normal blocking effect; This is the normal component reduction coefficient. The smaller the spacing, the greater the reduction. When approaching the wall, the normal component approaches zero, reflecting the normal blocking effect of the windward surface on the airflow. For the angle of inclination of the windward side; The tangential unit vector of the incoming flow direction within the horizontal projection of the windward surface; It is a vertically upward unit vector. This formula applies only when... It takes effect immediately; if the wind field outside this range has already returned to its inflow state, no correction is needed.

[0093] In practice, if the point to be corrected is located on the side of the windward surface where the normal direction points, the tangential component is deflected along the windward surface angle to make it parallel to the windward surface plane. The principle is that after the airflow impacts the windward surface, it is forced to flow along the surface direction. At the same time, the normal component is shortened according to the distance between it and the windward surface. The smaller the distance, the greater the reduction. When it approaches the wall, the normal component approaches zero. The principle is that the normal blocking effect of the windward surface on the airflow increases with decreasing distance, and the airflow cannot penetrate the surface when close to the wall. The corrected wind speed vector is obtained by adding the shortened normal component and the deflected tangential component vector.

[0094] The third step involves deflecting the tangential component along the leeward side when any point to be corrected is located on the leeward side. The modulus of the normal component is then reduced by the distance between the point to be corrected and the corresponding leeward side, inverted, and added to the deflected tangential component. The expression is as follows:

[0095]

[0096] The tangential deflection direction is:

[0097] In the formula, Points to be corrected The corrected three-dimensional wind speed vector; To correct the tangential component of the wind speed vector; To correct the normal component of the wind speed vector; It is the tangential unit vector after being deflected along the leeward tilt angle, and is parallel to the leeward plane; The outward normal unit vector of the leeward side; The normal distance from the point to be corrected to the leeward side; This represents the characteristic attenuation distance of the normal suction effect on the leeward side. The negative sign indicates that the normal component is reversed, and the direction changes from pointing towards the wall to away from the wall, characterizing the effect of airflow being drawn outward from the wall in the leeward recirculation zone. The angle of inclination of the leeward side; It is the tangential unit vector of the incoming flow direction within the horizontal projection of the leeward side; It is a vertically upward unit vector.

[0098] In practice, the tangential component is deflected along the leeward side at the angle of inclination so that it is parallel to the leeward plane; the normal component is first reduced in length according to the spacing and then reversed, that is, the direction changes from pointing towards the wall to away from the wall. The principle is that after the airflow bypasses the obstacle, a backflow zone is formed on the leeward side, and the normal velocity direction is opposite to that on the windward side, which manifests as the effect of airflow being drawn outward from the wall. Reversing the direction can accurately characterize this reverse flow feature.

[0099] Fourth step: When the point to be corrected is located in the edge region, the tangential component is deflected along the edge direction, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the edge, and then added to the deflected tangential component. The expression is:

[0100]

[0101] In the formula, Points to be corrected in the edge region The corrected three-dimensional wind speed vector; To correct the tangential component of the wind speed vector; To correct the normal component of the wind speed vector; The unit vector representing the direction of the edge is determined by the direction of the spatial line segment of the edge, and is equal to... ; The angle of the edge direction; It is the normal unit vector of the edge region, pointing from the point to be corrected to the nearest point on the edge; It is the shortest distance from the point to be corrected to the edge space line segment; The threshold for the influence range of the edge; This is the reduction factor for the normal component; the closer to the edge, the greater the reduction.

[0102] In practice, the edge region can be defined as a range no more than ten meters from the edge spatial line segment identified by S104. The tangential component of the point to be corrected within this region deflects along the edge direction. The principle is that when airflow passes through a sharp edge, it separates and flows laterally around the edge, with the tangential direction changing from the surface orientation to the edge direction. The summed vector is used as the corrected wind speed vector for that point, and the direction of the wind speed vector is used as the corrected wind direction. The corrected wind speed vectors and wind directions of all points to be corrected constitute the corrected predicted wind field, such as... Figure 3 As shown. The beneficial effect of this correction process is that by establishing differentiated correction rules for the three geometric features of the windward side, the leeward side, and the edge, key physical phenomena such as wall deflection, backflow, and edge separation can be simulated with lower computational overhead without relying on computational fluid dynamics numerical solutions. This allows the corrected wind field prediction to take into account both large-scale distribution trends and near-field refined characteristics.

[0103] Finally, the predicted wind shear distribution in the foreground area is determined as follows: The foreground area covered by the corrected predicted wind field is divided into sequentially adjacent spatial grids, with the center point of each grid taken as the wind field value location; for adjacent spatial grids, the corrected wind speed vector at the corresponding center point is read, and the difference between the corrected wind speed vectors is obtained; the wind speed vector difference is divided by the spatial distance between the two corresponding center points to obtain the wind shear parameters of the adjacent spatial grids; the wind shear parameters are attached to the common edges between adjacent spatial grids, and the wind shear parameters on all common edges constitute the predicted wind shear distribution in the foreground area, such as... Figure 4 As shown.

[0104] In practice, the terminal divides the area in front into a cubic spatial grid with sides of five meters. The corrected wind speed vector is read from the center point of each grid. For each pair of adjacent grids, the difference between the corrected wind speed vectors at the two center points is obtained as the wind speed vector difference. Dividing its magnitude by the spatial distance between the two center points yields the wind shear parameter, which characterizes the rate of change of the wind speed vector per unit distance. The wind shear parameter is attached to the common edges between adjacent grids, and the predicted wind shear distribution is formed by the shear parameters on all common edges. The principle and beneficial effect are that it organizes the spatial gradient information of the wind field in the area in front using a graph structure. The shear parameter on each common edge directly reflects the intensity of the wind field at that location, transforming the spatial inhomogeneity of the continuous wind field into structured data that can be directly compared and thresholded. This allows subsequent identification of the S106 hazard area to be completed simply by traversing the shear parameters of each common edge, eliminating the need for complex gradient calculations in continuous space, improving processing efficiency, and providing a quantitative basis for path offset correction.

[0105] Among them, the wind field shear parameters of adjacent spatial grids include , , Their expressions are as follows:

[0106]

[0107]

[0108] In the formula, The modulus of the corrected wind speed vector difference at the center points of adjacent spatial grids a and b; The corrected three-dimensional wind speed vector at the center point of spatial grid a; The corrected 3D wind speed vector at the center point of spatial grid b; The three-dimensional position vector of the center point of spatial grid a; The three-dimensional position vector of the center point of spatial grid b; This represents the spatial distance between the center points of adjacent spatial grids a and b. The wind shear parameter is the wind field shear parameter on the common edge between adjacent spatial grids a and b, which represents the rate of change of the wind speed vector per unit distance.

[0109] S106, connect the locations in the predicted wind field shear distribution where the wind speed shear component exceeds the preset wind speed shear threshold or the wind direction shear component exceeds the preset angle to form a danger zone; map the danger zone onto the first route, and perform offset correction on the path points in the danger zone in the first route to generate a corrected route that avoids the danger zone.

[0110] Specifically, locations where the wind speed shear component exceeds a preset wind speed shear threshold or the wind direction shear component exceeds a preset angle are connected to form a danger zone. In practice, the terminal iterates through the wind field shear parameters on each common edge of the predicted wind field shear distribution obtained in S105, decomposing them into wind speed shear components and wind direction shear components. The preset wind speed shear threshold can be set to 0.5 per second, and the preset angle can be set to 15 degrees. When the wind speed shear component on any common edge exceeds 0.5 per second or the wind direction shear component exceeds 15 degrees per meter, the spatial grid connected to that common edge is marked as a danger grid. Then, adjacent danger grids are sequentially connected and merged into a connected danger zone. The principle is that sudden changes in wind speed amplitude and sharp deflections in wind direction both disturb the UAV's attitude. Setting separate thresholds covers both types of risks, and merging them into a continuous area provides a clear spatial boundary for subsequent avoidance.

[0111] The expression for the wind speed shear component is:

[0112] The wind shear component is:

[0113] In the formula, It represents the wind speed shear component on the common edge between adjacent grids a and b, i.e., the rate of change of wind speed per unit distance; This represents the wind shear component along the common edge between adjacent grids a and b, i.e., the rate of change of wind direction angle per unit distance. The magnitude of the corrected wind speed vector at the center point of grid a is the wind speed. The magnitude of the corrected wind speed vector at the center point of grid b is the wind speed. The dot product of the corrected wind speed vectors at the center points of the two grids; This represents the spatial distance between the center points of adjacent grids a and b.

[0114] The danger zone is mapped onto the first route, and the path points within the danger zone on the first route are offset and corrected to generate a corrected route that avoids the danger zone. The terminal then performs intersection determination between the coordinates of the path points of each UAV's first route generated by S101 and the spatial boundary of the danger zone, thereby accurately locating the flight segments that need adjustment.

[0115] Offset correction is performed on path points located within the danger zone in the first route, including: traversing all path points on the first route and marking those falling into the danger zone as target correction points; point-by-point determination ensures that all segments crossing the danger zone are identified; for each target correction point, the predicted wind field shear distribution around the target correction point is read, and the offset direction is determined along the direction of minimum wind field shear parameter. Specifically, the common edge shear parameter within a 30-meter radius centered on the target correction point is selected, and the average shear parameter in each direction is searched at 10-degree intervals in the horizontal plane. The direction with the minimum value is taken as the offset direction. The principle is that the direction with the minimum shear parameter corresponds to the airspace with the most stable airflow, and offsetting along this direction can place the corrected path in the area with the weakest disturbance; the offset distance is determined based on the distance from the target correction point to the boundary of the danger zone, and the target correction point is moved out of the danger zone along the offset direction. The offset distance is set to 1.2 times the shortest distance from the target correction point to the boundary, to reserve a safety margin on the basis of completely leaving the danger zone and avoid re-entry due to attitude offset; the offset path points are connected sequentially with adjacent path points to generate the corrected route, such as... Figure 5 As shown, maintaining temporal continuity ensures a smooth and continuous flight path. Taking the aforementioned formation as an example, three waypoints located in the danger zone east of the building in the first route are marked as target correction points. After being offset by approximately eighteen meters along the direction of minimum shear parameter on the south side, they are connected to the waypoints before and after the target, forming a corrected route that detours southward, allowing the formation to safely avoid the leeward backflow area.

[0116] like Figure 6 As shown, based on the same inventive concept, this embodiment provides a UAV path planning system for strong wind environments, including: The path planning module 201 is used to obtain the target location, plan a first route based on the target location, send the first route to each UAV, control each UAV to take off and fly along the first route, and maintain the distance between adjacent UAVs within a preset range. The data acquisition module 202 is used to collect wind speed and direction information of each UAV in real time through wind speed and direction sensors of each UAV, collect three-dimensional point cloud data of a preset range in front of each UAV through radar, and send the wind speed and direction information and three-dimensional point cloud data to the terminal in real time. The shear network module 203 is used to decompose the corresponding wind speed and wind direction information of each pair of adjacent UAVs into components along the direction of the line connecting the adjacent UAVs and components perpendicular to the direction of the line connecting the UAVs. The difference between the components along the direction of the line connecting the UAVs and the difference between the components perpendicular to the direction of the line connecting the UAVs are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent UAVs. The wind speed vector difference is divided by the distance between the adjacent UAVs to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent UAVs to form a shear edge. All shear edges constitute the wind field shear network. The obstacle recognition module 204 is used to perform surface reconstruction processing on the three-dimensional point cloud data collected by the radar, identify the three-dimensional geometric shape of the obstacle in the front area, and extract the windward tilt angle, leeward tilt angle and edge orientation of the corresponding obstacle from the three-dimensional geometric shape. The wind field prediction module 205 is used to collect wind speed and direction information using the real-time positions of each UAV as wind field sampling points. It uses the wind field shear variables of each shear edge in the wind field shear network as wind field spatial gradient constraints. It extrapolates the unsampled area ahead along the direction of incoming flow indicated by the wind direction information to obtain the initial predicted wind field. Based on the windward tilt angle, leeward tilt angle and edge orientation corresponding to different positions, it corrects the wind speed and direction of the area around the corresponding obstacle in the initial predicted wind field based on a preset wind simulation model to obtain the corrected predicted wind field. Based on the ratio of the wind speed vector difference between different spatial positions in the corrected predicted wind field to the corresponding spatial distance, it determines the predicted wind field shear distribution in the area ahead. The path correction module 206 is used to connect the locations in the predicted wind field shear distribution where the wind speed shear component exceeds the preset wind speed shear threshold or the wind direction shear component exceeds the preset angle to form a dangerous area; map the dangerous area onto the first route, and perform offset correction on the path points in the dangerous area in the first route to generate a corrected route that avoids the dangerous area.

[0117] Based on the same inventive concept, this embodiment provides an electronic device, including: Memory, used to store computer programs; A processor is used to implement the method steps as described in the first aspect when executing a computer program.

[0118] Based on the same inventive concept, this embodiment provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the method steps as described in the first aspect.

[0119] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0120] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for UAV path planning in strong wind conditions, characterized in that, The method includes multiple drones, each equipped with a wind speed and direction sensor and radar, and comprises: S101, Obtain the target location, plan a first route based on the target location, send the first route to each drone, control each drone to take off and fly along the first route, and maintain the distance between adjacent drones within a preset range; S102, by using the wind speed and direction sensors of each UAV, the wind speed and direction information of the location of each UAV is collected in real time, and the three-dimensional point cloud data of the preset range in front of each UAV is collected by the radar, and the wind speed and direction information and the three-dimensional point cloud data are sent to the terminal in real time. S103, for each pair of adjacent drones, the corresponding wind speed and wind direction information is decomposed into components along the direction of the line connecting the adjacent drones and components perpendicular to the direction of the line connecting the drones. The difference between the components along the line connecting the drones and the difference between the components perpendicular to the line connecting the drones are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent drones. The wind speed vector difference is divided by the distance between the adjacent drones to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent drones to form a shear edge. All the shear edges constitute a wind field shear network. S104, Perform surface reconstruction processing on the three-dimensional point cloud data collected by the radar to identify the three-dimensional geometric shape of the obstacle in the front area, and extract the windward tilt angle, leeward tilt angle and edge orientation of the obstacle from the three-dimensional geometric shape; S105, using the real-time positions of each UAV as wind field sampling points, collect the wind speed and wind direction information; using the wind field shear variables of each shear edge in the wind field shear network as wind field spatial gradient constraints; extrapolate the unsampled area ahead along the incoming flow direction indicated by the wind direction information to obtain the initial predicted wind field; based on the windward inclination angle, leeward inclination angle, and edge orientation corresponding to different positions, correct the wind speed and wind direction of the area around the corresponding obstacle in the initial predicted wind field according to a preset wind simulation model to obtain the corrected predicted wind field; based on the ratio of the wind speed vector difference between different spatial positions in the corrected predicted wind field to the corresponding spatial distance, determine the predicted wind field shear distribution in the area ahead; S106, in the predicted wind field shear distribution, the positions where the wind speed shear component exceeds a preset wind speed shear threshold or the wind direction shear component exceeds a preset angle are connected to form a danger zone; the danger zone is mapped onto the first route, and the path points in the first route located in the danger zone are offset and corrected to generate a corrected route that avoids the danger zone.

2. The method for UAV path planning in strong wind conditions according to claim 1, characterized in that, The planning of the first route based on the target location specifically includes: The takeoff position of each UAV is obtained as the path starting point, and the three-dimensional geometry is converted into a rasterized spatial occupancy representation. In the spatial occupancy representation, each grid corresponds to a three-dimensional space and marks whether the three-dimensional space is occupied by terrain or buildings. Using the grid where the path starts as the initial search node, the adjacent grids are expanded along the direction toward the target location, including: reading the corresponding occupancy mark for each expanded candidate node, discarding the candidate nodes marked as occupied within a preset range, and retaining the remaining candidate nodes as the starting node for the next round of node expansion; The node expansion is repeated until the expanded node falls into the grid where the target location is located. The sequence of nodes that are sequentially connected from the initial search node to the node that falls into the grid where the target location is located is used as the reference track. The reference track is offset according to the relative position of each UAV in the formation to generate the first route corresponding to each UAV.

3. The method for UAV path planning in strong wind conditions according to claim 1, characterized in that, The extraction of the windward slope angle, leeward slope angle, and edge orientation of the corresponding obstacle from the three-dimensional geometry specifically includes: Based on the incoming flow direction indicated by the wind speed and wind direction information, among the surfaces contained in the three-dimensional geometry, the surface with the normal pointing to the direction of the incoming flow is selected as the windward surface, and the surface with the normal away from the direction of the incoming flow is selected as the leeward surface. The angle between the windward side and the horizontal plane is determined as the windward side inclination angle, and the angle between the leeward side and the horizontal plane is determined as the leeward side inclination angle; the intersection line between adjacent surfaces in the three-dimensional geometry is identified as the edge, and the projection direction of the edge on the horizontal plane is determined as the edge direction.

4. The method for UAV path planning in strong wind conditions according to claim 1, characterized in that, The initial predicted wind field is obtained by extrapolating the unsampled area ahead along the incoming flow direction indicated by the wind direction information, specifically including: Using the real-time location of each UAV as a wind field sampling point, the sampling is advanced segment by segment from each sampling point towards the unsampled area along the direction of the incoming flow indicated by the wind direction information; the influence propagation distance corresponding to each segment is determined according to the wind speed, and the wind speed is directly proportional to the influence propagation distance; For each unsampled location, a wind field sampling point located upstream of that location and along the incoming flow direction is selected. The wind speeds collected by the selected wind field sampling points are attenuated according to their distances from the unsampled location and then superimposed to obtain the predicted wind speed at that unsampled location. The wind direction information along the incoming flow direction is assigned to the unsampled location as the predicted wind direction. The predicted wind speeds and predicted wind directions of all unsampled locations constitute the initial predicted wind field.

5. The method for UAV path planning in strong wind conditions according to claim 1, characterized in that, The step of correcting the wind speed and direction in the area surrounding the corresponding obstacle in the initial predicted wind field based on a preset wind simulation model to obtain the corrected predicted wind field specifically includes: The grid points located within a preset range around the obstacle in the initial predicted wind field are taken as points to be corrected; for each point to be corrected, the wind speed vector of the point to be corrected in the initial predicted wind field is decomposed into tangential and normal components according to the orientation relationship between the point to be corrected and the surface of the obstacle. When any of the points to be corrected is on the windward side, the tangential component is deflected along the tilt angle of the windward side, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the windward side and then added to the deflected tangential component. When any of the points to be corrected is on the leeward side, the tangential component is deflected along the leeward tilt angle, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the corresponding leeward side, and then inverted and added to the deflected tangential component. When the point to be corrected is located in the edge region, the tangential component is deflected along the edge direction, and the modulus of the normal component is reduced according to the distance between the point to be corrected and the edge and added to the deflected tangential component; the vector obtained by addition is used as the corrected wind speed vector of the point to be corrected, and the direction of the wind speed vector is used as the corrected wind direction; the corrected wind field is composed of the corrected wind speed vectors and wind directions of all points to be corrected.

6. The method for UAV path planning in strong wind conditions according to claim 5, characterized in that, The determination of the predicted wind field shear distribution in the foreground area based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance specifically includes: The area in front of the corrected predicted wind field is divided into sequentially adjacent spatial grids, and the center point of each spatial grid is taken as the wind field value location; for adjacent spatial grids, the corrected wind speed vector at the corresponding center point is read, and the difference between the corrected wind speed vectors is obtained. Divide the wind speed vector difference by the spatial distance between the two corresponding center points to obtain the wind field shear parameters of the adjacent spatial grids; attach the wind field shear parameters to the common edge between the adjacent spatial grids, and the wind field shear parameters on all the common edges constitute the predicted wind field shear distribution of the front area.

7. The method for UAV path planning in strong wind conditions according to claim 1, characterized in that, The offset correction of path points located within the danger zone in the first route includes: Traverse each path point on the first route and mark the path points that fall into the danger zone as target correction points; for each target correction point, read the predicted wind field shear distribution around the target correction point in the corrected predicted wind field, and determine the offset direction along the direction with the minimum wind field shear parameter. The offset distance is determined based on the distance from the target correction point to the boundary of the danger zone, and the target correction point is moved out of the danger zone along the offset direction; the offset path points are connected to adjacent path points in sequence to generate the correction route.

8. A path planning system for unmanned aerial vehicles (UAVs) in strong wind conditions, employing the path planning method for UAVs in strong wind conditions as described in claim 1, characterized in that... include: The path planning module is used to obtain the target location, plan a first route based on the target location, send the first route to each drone, control each drone to take off and fly along the first route, and maintain the distance between adjacent drones within a preset range. The data acquisition module is used to collect wind speed and direction information of each UAV in real time through the wind speed and direction sensors of each UAV, collect three-dimensional point cloud data of a preset range in front of each UAV through the radar, and send the wind speed and direction information and the three-dimensional point cloud data to the terminal in real time. The shear network module is used to decompose the wind speed and wind direction information of each pair of adjacent UAVs into components along the direction of the line connecting the adjacent UAVs and components perpendicular to the direction of the line connecting the adjacent UAVs. The difference between the components along the line connecting the line and the difference between the components perpendicular to the line connecting the line are calculated respectively. The two differences are combined to obtain the wind speed vector difference between the adjacent UAVs. The wind speed vector difference is divided by the distance between the adjacent UAVs to obtain the corresponding wind field shear variable. The wind field shear variable is attached to the line connecting the adjacent UAVs to form a shear edge. All the shear edges constitute the wind field shear network. The obstacle recognition module is used to perform surface reconstruction processing on the three-dimensional point cloud data collected by the radar, identify the three-dimensional geometric shape of the obstacle in the front area, and extract the windward tilt angle, leeward tilt angle and edge orientation of the obstacle from the three-dimensional geometric shape. The wind field prediction module is used to collect wind speed and direction information using the real-time locations of each UAV as wind field sampling points, and to use the wind field shear variables of each shear edge in the wind field shear network as wind field spatial gradient constraints. It extrapolates the unsampled area ahead along the direction of incoming flow indicated by the wind direction information to obtain the initial predicted wind field. Based on the windward slope angle, leeward slope angle, and edge orientation corresponding to different locations, it corrects the wind speed and direction in the area surrounding the corresponding obstacle in the initial predicted wind field using a preset wind simulation model to obtain the corrected predicted wind field. Based on the ratio of the wind speed vector difference between different spatial locations in the corrected predicted wind field to the corresponding spatial distance, it determines the predicted wind field shear distribution in the area ahead. The path correction module is used to connect the locations in the predicted wind field shear distribution where the wind speed shear component exceeds a preset wind speed shear threshold or the wind direction shear component exceeds a preset angle to form a danger zone; map the danger zone onto the first route, and perform offset correction on the path points in the first route located in the danger zone to generate a corrected route that avoids the danger zone.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to implement, when executing the computer program, a method for unmanned aerial vehicle (UAV) path planning in a strong wind environment as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements a method for unmanned aerial vehicle (UAV) path planning in strong wind conditions as described in any one of claims 1 to 7.

Images