Urban low-altitude unmanned aerial vehicle flight path optimization method and system

Abstract

The application relates to a kind of urban low-altitude unmanned aerial vehicle flight path optimization method and system, it is related to unmanned aerial vehicle path optimization technical field, including: obtaining the three-dimensional building map of target flight area, marking flight starting point and end point position, determine initial flight path;Build fast airflow simulator, obtain micro air flow data in combination with macro air flow data;Based on the micro air flow data of current time, in combination with the performance parameters of target model unmanned aerial vehicle, the path cost of executing initial flight path is calculated;With three-dimensional building map, target model unmanned aerial vehicle as constraint, with the lowest path cost as target, the initial flight path is iteratively optimized, and the optimal flight path is obtained.The application solves the problem that in the traditional unmanned aerial vehicle path planning, the wind direction and wind speed on the route are difficult to predict due to the complex building environment of urban low altitude, the path is planned only from a single distance dimension, the path cannot adapt to the air flow difference of urban low altitude, and the flight safety of unmanned aerial vehicle is affected.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) path optimization, and in particular to a method and system for optimizing the flight path of urban low-altitude UAVs. Background Technology

[0002] With the expansion of drone applications in urban low-altitude scenarios, optimizing the flight paths of urban low-altitude drones has become a key foundation for ensuring flight safety and improving operational efficiency.

[0003] Existing drone path planning methods make decisions based on a single distance dimension. In urban low-altitude environments, the complex building environment makes it difficult to predict wind direction and speed along the flight path, and cannot adapt to the differences in urban low-altitude airflow. Either the flight time and energy consumption are increased due to ignoring the influence of airflow, or flight stability problems are caused by airflow interference, making it difficult to meet the dual requirements of flight safety and operational efficiency. Summary of the Invention

[0004] This application provides a method and system for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs). It overcomes the shortcomings of traditional UAV path planning, which only makes decisions based on a single distance dimension, and enhances the adaptability to differences in urban low-altitude airflow. This not only ensures the flight safety of UAVs but also improves the precision of path planning, achieving a balance between flight safety and operational efficiency.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] In a first aspect, embodiments of this application provide a method for optimizing the flight path of a low-altitude unmanned aerial vehicle (UAV) in an urban area, the method comprising:

[0007] Obtain a 3D building map of the target flight area. Based on the 3D building map, mark the start and end points of the flight. Using the start and end points and the 3D building map as constraints, obtain the shortest flight path as the initial flight path.

[0008] Based on a 3D building map of the target flight area, a fast airflow simulator of the target flight area is obtained. Based on the fast airflow simulator and macro airflow data, micro airflow data of multiple sub-regions in the target flight area are obtained. Both the macro airflow data and the micro airflow data include wind speed and wind direction.

[0009] Based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, combined with the performance parameters of the target UAV, the path cost of the target UAV executing the initial flight path is calculated.

[0010] Using a 3D building map of the target flight area and the target UAV model as constraints, and aiming at the lowest path cost, the initial flight path is iteratively optimized to obtain the optimal flight path.

[0011] Secondly, embodiments of this application provide an urban low-altitude unmanned aerial vehicle (UAV) flight path optimization system, the system comprising:

[0012] The initial path acquisition module is used to acquire a 3D building map of the target flight area. Based on the 3D building map, the flight start and end points are marked. Using the start and end points and the 3D building map as constraints, the shortest flight path is acquired as the initial flight path.

[0013] The airflow data acquisition module is used to obtain a fast airflow simulator of the target flight area based on a three-dimensional building map of the target flight area, and to obtain micro airflow data of multiple sub-regions in the target flight area based on the fast airflow simulator and macro airflow data. Both the macro airflow data and the micro airflow data include wind speed and wind direction.

[0014] The path cost calculation module is used to calculate the path cost of the target UAV executing the initial flight path based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, combined with the performance parameters of the target UAV model.

[0015] The path iteration optimization module is used to iteratively optimize the initial flight path with the constraints of a 3D building map of the target flight area and the target UAV model, and with the goal of minimizing the path cost, to obtain the optimal flight path.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] This application proposes a method and system for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs). Through steps such as constructing a 3D building map, using a fast airflow simulator, calculating path costs, and iterative optimization, it achieves end-to-end optimization from basic modeling to optimal flight path generation. First, a 3D building map of the target flight area is acquired, and the starting and ending points are marked to generate the shortest initial flight path, providing a basic trajectory for subsequent optimization. Next, a fast airflow simulator is trained based on the 3D building map, and micro-airflow data for multiple sub-regions is output in conjunction with macro-airflow data. Then, the path cost of the initial flight path is calculated based on the performance parameters of the target UAV model, quantifying the flight time and energy consumption. Finally, constrained by the 3D building map and UAV performance, the lowest-cost optimal flight path is obtained through multiple rounds of "local node perturbation - generating alternative flight paths - calculating path costs - iterative optimization," and used to control the UAV's flight.

[0018] This application's technical solution ensures path safety through 3D building maps, improves the accuracy of airflow prediction through a fast airflow simulator, quantifies costs through path cost calculation, and iteratively optimizes to approach the optimal solution. It solves the problem that traditional UAV path planning ignores the influence of urban buildings and airflow, resulting in short endurance or low safety. It enables efficient, safe, and low-consumption flight of urban low-altitude UAVs, providing reliable technical support for path planning in urban UAV logistics, inspection, and other scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a method for optimizing the flight path of a low-altitude unmanned aerial vehicle (UAV) in an urban area, as provided in this application embodiment;

[0021] Figure 2 This is a schematic diagram of a flight path optimization system for urban low-altitude unmanned aerial vehicles (UAVs) provided in an embodiment of this application.

[0022] The components represented by each number in the attached diagram are explained below:

[0023] Initial path acquisition module 01, airflow data acquisition module 02, path cost calculation module 03, path iterative optimization module 04. Detailed Implementation

[0024] This application provides a method and system for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs), which solves the technical problem that existing technologies cannot adapt to the differences in urban low-altitude airflow when UAVs plan their paths based on a single distance dimension in complex urban low-altitude building environments, thus affecting UAV flight safety and leading to increased flight time and energy consumption.

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0027] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0028] Example 1, as shown in the appendix Figure 1 As shown, this application provides a method for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs), the method comprising the following steps:

[0029] S110: Obtain a 3D building map of the target flight area. Based on the 3D building map, mark the start and end points of the flight. Using the start and end points and the 3D building map as constraints, obtain the shortest flight path as the initial flight path.

[0030] In this embodiment of the application, in the scenario of complex building environment at low altitude in urban areas, in order to ensure that the flight path of the UAV meets the requirements for avoiding buildings at low altitude in urban areas and has basic path efficiency, it is necessary to first complete the accurate construction of the three-dimensional building map of the target flight area, and then generate the initial flight path in combination with the core constraints.

[0031] Specifically, the target flight area of ​​the UAV is first defined. Based on flight planning software, and combined with the range and shape characteristics of the area, a mapping grid flight path is planned that covers the entire area and scans without omissions.

[0032] The mapping grid route is a dedicated operating route for mapping drones, which can ensure a comprehensive scan of all buildings within the target area.

[0033] Furthermore, by controlling the surveying drone along the planned surveying grid route, the buildings in the target flight area are scanned from multiple perspectives. Through multi-perspective data collection, the triangular grid model and coordinate information of each building are accurately obtained.

[0034] Furthermore, the triangular mesh models of all buildings are arranged and integrated in three-dimensional space according to their actual coordinates to form a complete three-dimensional building map of the target flight area.

[0035] Finally, the starting and ending points of the UAV flight are accurately marked on the 3D building map. Based on the constraints of the 3D building map and the starting / ending point constraints, the shortest flight path is calculated and selected by the path planning algorithm and used as the initial flight path.

[0036] This step involves first accurately constructing a 3D building map to ensure path safety, and then generating an initial flight path with the shortest distance as the core objective. This not only avoids interference from complex low-altitude buildings in the city on the flight, but also lays an efficient foundation for subsequent path optimization based on airflow data, ensuring that the initial flight path is both safe and efficient.

[0037] Step S110 in the method provided in this application embodiment includes:

[0038] Obtain the target flight area of ​​the drone;

[0039] Based on flight planning software and the extent and shape of the target flight area, plan and map the route using a grid.

[0040] Using a surveying drone, based on the surveying grid flight path, the buildings in the target flight area are scanned from multiple perspectives with overlapping data to obtain the triangular mesh model and coordinate information of each building in the target flight area.

[0041] Based on the triangular mesh model and coordinate information of each building in the target flight area, a three-dimensional building map of the target flight area is obtained.

[0042] In this embodiment of the application, in order to achieve accurate optimization of the flight path of the UAV in the urban low-altitude environment, it is necessary to first construct a three-dimensional model that can truly reflect the spatial distribution of urban buildings, so as to serve as the spatial constraint basis for subsequent airflow simulation and path planning.

[0043] First, by analyzing the mission requirements and defining the geographical scope of the UAV, the target flight area of ​​the UAV is obtained. This target flight area clarifies the specific spatial boundaries for subsequent airflow prediction and flight path planning, and is the core basis for all surveying and modeling.

[0044] Furthermore, based on flight planning software and the extent and shape of the target flight area, a surveying grid route is planned.

[0045] Since urban buildings are often distributed in an irregular shape, if random flight routes are used for surveying, blind spots or repeated scanning of buildings may occur. In contrast, the surveying grid flight route is a standardized flight route that is suitable for the needs of building surveying.

[0046] Therefore, by using existing flight planning software, route parameters can be set according to the geometric characteristics of the target flight area. For example, for rectangular areas, orthogonal grid routes are used to ensure uniform spacing of flight routes and coverage without blind spots; for irregularly shaped areas, adaptive grid routes are used to adjust the route direction along the area boundary to fit the shape of the target flight area and achieve complete coverage.

[0047] Meanwhile, the flight planning software will also set the flight altitude in conjunction with the building height to avoid collisions between the surveying drone and the building, and to ensure that the building facade and top details can be clearly captured, providing orderly and safe flight path support for subsequent scanning.

[0048] Specifically, using a surveying drone, based on a planned surveying grid flight path, the buildings in the target flight area are scanned from multiple perspectives to obtain the triangular mesh model and coordinate information of each building.

[0049] Among them, the surveying drone is equipped with high-definition stereo imaging equipment and high-precision positioning device. When flying along the preset grid route, it will collect data on the same building multiple times from different heights and angles to avoid information loss caused by building occlusion from a single perspective.

[0050] Meanwhile, the collected image data will generate a point cloud model of the building through existing image matching algorithms, and then be triangulated into a triangular mesh model. This triangular mesh model can accurately restore the three-dimensional shape of the building, including geometric details such as wall thickness and roof slope.

[0051] In addition, the high-precision positioning device records the coordinate information (latitude, longitude, and altitude) of each key point of the building to ensure that the triangular mesh model corresponds to the real geographic location. Together, they constitute the complete spatial data of a single building, laying the foundation for the subsequent integration of 3D building maps.

[0052] Finally, based on the triangular mesh model and coordinate information of each building in the target flight area, a three-dimensional building map of the target flight area is obtained.

[0053] Specifically, the integration process follows the principle of real coordinate mapping, which means that the triangular mesh model of each building is precisely arranged in a three-dimensional spatial coordinate system according to its corresponding latitude, longitude and altitude, to ensure that the relative distance and height difference between buildings are completely consistent with the actual urban environment.

[0054] For example, adjacent residential buildings on both sides of a street will be spaced 5 meters apart in three-dimensional space based on their actual latitude and longitude differences, and an office building with a height of 30 meters and a shop with a height of 15 meters will have a vertical drop of 15 meters based on their actual altitude.

[0055] Simultaneously, the integrated model will undergo data optimization processing: on the one hand, the jagged edges caused by scanning errors in the triangular mesh model will be eliminated through edge smoothing algorithms in existing technologies; on the other hand, redundant points generated during point cloud matching will be removed through data noise reduction to ensure the clarity and accuracy of the 3D building map.

[0056] The resulting 3D building map not only visually presents the three-dimensional distribution characteristics of urban buildings, but also provides precise building geometric boundaries for the subsequent construction of a rapid airflow simulator, and provides clear obstacle constraints for planning the initial path of drones, thus serving as the spatial basis for optimizing the path of urban low-altitude drones.

[0057] Furthermore, the starting and ending points of the drone flight are clearly marked in the completed 3D building map.

[0058] The starting point is usually the take-off and landing point of the drone, and the ending point is the mission target point of the drone. During the labeling process, it is necessary to combine the real geographic coordinates to ensure that the positions of the starting point and the ending point in the 3D building map correspond completely with the actual scene, so as to avoid the subsequent path planning deviating from the actual needs due to coordinate deviation.

[0059] After the annotation is completed, the path analysis is performed using existing path planning algorithms, with the starting point, ending point, and 3D building map as constraints, to obtain the shortest flight path and use it as the initial flight path.

[0060] Among them, the constraint effect of the 3D building map is reflected in obstacle avoidance, that is, the path planning algorithm will automatically identify the building models in the map to ensure that the planned path will not pass through any buildings, while avoiding narrow gaps between buildings; the constraint effect of the start point and the end point is reflected in orientation planning, to ensure that the analysis path always revolves around the goal from the start point to the end point, and there will be no deviation in direction.

[0061] When analyzing the initial flight path, the path planning algorithm uses a three-dimensional spatial coordinate system and a straight-line distance-first analysis method. That is, it first calculates the straight-line spatial distance between the starting point and the ending point, and then combines the building distribution in the three-dimensional building map to make local adjustments to the straight-line path.

[0062] If the straight path does not pass through a building, it will be used as the initial flight path. If the straight path intersects with a building, it will detour along the edge of the building to ensure that the adjusted path, while avoiding obstacles, has the overall distance closest to the straight-line distance between the start and end points.

[0063] For example, if the starting point is located in the square of Community A (coordinates: 104.06°E, 30.67°N, altitude 500m) and the ending point is located on the roof of Office Building B (coordinates: 104.08°E, 30.68°N, altitude 520m), and there is a 20-story residential building (altitude 500-560m) in the middle of the straight path between the two, then the path planning algorithm will automatically bypass the residential building and plan an initial flight path of "taking off from the square of Community A - flying 50m east of the residential building - arriving at the roof of Office Building B". This initial flight path avoids building obstacles and is closest to the straight distance between the starting point and the ending point.

[0064] S120: Based on the three-dimensional building map of the target flight area, obtain a fast airflow simulator of the target flight area. Based on the fast airflow simulator and macro airflow data, obtain micro airflow data of multiple sub-regions in the target flight area. Both the macro airflow data and the micro airflow data include wind speed and wind direction.

[0065] In this embodiment of the application, in the scenario of complex building environment at low altitude in the city, in order to accurately obtain the airflow state on the flight path of the UAV and avoid the impact of airflow prediction deviation caused by urban building blockage on flight endurance and safety, it is necessary to build a fast airflow simulator that is adapted based on the three-dimensional building map, and then combine macro airflow data to derive micro airflow data, so as to form airflow data support that fits the actual flight environment.

[0066] Specifically, the first step is to extract macroscopic airflow data of the target flight area. This data includes wind direction and speed data over a wide area of ​​the target region, which can be directly obtained by connecting to the real-time monitoring platform of the local meteorological station and can reflect the overall airflow trend of the target region.

[0067] Furthermore, based on the constructed 3D building map of the target flight area, a fast airflow simulator specific to that area is trained to accurately reproduce the impact of urban buildings on airflow.

[0068] Finally, the extracted macroscopic airflow data of the target flight area is input into the trained fast airflow simulator. The simulator will combine the building distribution characteristics in the 3D building map to calculate the flow and turbulence of macroscopic airflow between urban buildings, and finally output the microscopic airflow data of multiple sub-regions in the target flight area.

[0069] This step involves building a rapid airflow simulator based on a 3D building map and then deriving micro-airflow data from macro-airflow data. This not only solves the problem of accurately predicting airflow in urban low-altitude building environments, but also provides real-time and accurate airflow data for subsequent path cost calculations and flight path optimization, ensuring that subsequent path optimization can fully adapt to the actual airflow conditions in urban low-altitude areas.

[0070] Step S120 in the method provided in this application embodiment includes:

[0071] Extract macroscopic airflow data of the target flight area;

[0072] A fast airflow simulator for the target flight area is trained based on a 3D building map of the target flight area.

[0073] Macroscopic airflow data of the target flight area is input into the fast airflow simulator to obtain microscopic airflow data of multiple sub-regions within the target flight area.

[0074] In this embodiment of the application, in order to avoid the impact of wind direction and wind speed prediction deviations on the drone's endurance and flight stability, it is necessary to construct and train a fast airflow simulator adapted to the target area based on a three-dimensional building map, and then combine macro airflow data to deduce the micro airflow state in order to form airflow data support that fits the actual flight scenario.

[0075] First, macroscopic airflow data for the target flight area is extracted. This macroscopic airflow data consists of wind direction and speed data over a large area of ​​the target flight area, which can be obtained directly by connecting to the real-time monitoring platform of the local meteorological station.

[0076] For example, if the target flight area is the core business district of a city, the meteorological station will provide macroscopic data such as the overall wind direction and average wind speed of the area on that day to reflect the overall airflow trend of the target flight area. This data serves as the basis for deriving microscopic airflow data and ensures that the subsequent airflow simulation does not deviate from the overall climate characteristics of the target flight area.

[0077] Furthermore, based on the 3D building map of the target flight area, a fast airflow simulator specifically for that target flight area is trained and constructed to solve the problem of complex and time-consuming calculation of the original airflow simulation in urban low-altitude environments.

[0078] The method provided in this application embodiment, which involves "training a fast airflow simulator for the target flight area based on a three-dimensional building map of the target flight area", includes:

[0079] Based on a 3D building map of the target flight area, combined with a fluid dynamics model, an original airflow simulator was established.

[0080] Based on the wind measurement drone, equipped with airflow sensing equipment, the actual airflow data of multiple sub-regions of the target flight area is measured as sample micro airflow data, and the macro airflow data of the target flight area is acquired in real time as sample macro airflow data;

[0081] Using sample macroscopic airflow data and sample microscopic airflow data, the simulation results of the original airflow simulator are corrected to obtain an airflow simulator for the target flight area.

[0082] Macroscopic airflow data of the target flight area is input into the airflow simulator to obtain predicted airflow data for multiple sub-regions of the target flight area;

[0083] Collect macroscopic airflow data of the target flight area at multiple times, as well as microscopic airflow data of multiple sub-regions predicted by the airflow simulator corresponding to the macroscopic airflow data at multiple times, as a sample predicted airflow dataset;

[0084] Using the sample predicted airflow dataset, the fast airflow simulator is trained to obtain micro airflow data for multiple sub-regions within the target flight area.

[0085] In this embodiment of the application, in order to accurately obtain the airflow status of each sub-region on the UAV flight path, it is necessary to build and train a fast airflow simulator based on three-dimensional building maps and historical airflow sample data, so that it can efficiently capture the mapping relationship between macro airflow and micro airflow, and provide real-time and accurate airflow data support for subsequent UAV path cost calculation and optimization.

[0086] First, a primitive airflow simulator is established based on a 3D building map of the target flight area and a fluid dynamics model.

[0087] Specifically, the 3D building map has accurately reproduced the triangular mesh model of all buildings in the target area. Using this as a spatial framework, the Navier-Stokes equations (the core equations describing the motion of viscous fluids) from fluid mechanics are introduced to build an original airflow simulator.

[0088] The original airflow simulator is an aerodynamic model built based on fluid mechanics formulas. It can calculate the changes in airflow around, turbulence, acceleration or deceleration in the low-altitude urban area based on the spatial distribution of buildings.

[0089] For example, when the macro airflow is "north wind 3m / s", the original airflow simulator can calculate the wind direction deflection angle of the airflow in the gap on the east side of a 25-story building due to the building's obstruction, and the wind speed increase caused by the funnel effect in the airflow in a narrow east-west street.

[0090] However, it should be noted that the original airflow simulator needs to calculate the impact of buildings on airflow point by point, involving a large number of matrix operations and fluid state iterations. The simulation of a single set of macroscopic airflow data can take 8-12 minutes, which cannot meet the real-time requirement of "outputting airflow data within 10 seconds" for UAV path planning. Therefore, further optimization and correction are needed.

[0091] Furthermore, the microscopic airflow data and macroscopic airflow data of the samples were collected by the wind-measuring drone.

[0092] Specifically, a wind-measuring drone equipped with high-precision airflow sensors was selected to conduct on-site airflow measurements in multiple sub-regions within the target flight area.

[0093] The selection of sub-regions needs to cover different building environment types, including several typical sub-regions such as "streets in densely populated high-rise buildings", "between two super high-rise buildings", "surroundings of city squares", "intersections", and "low-rise building clusters". Each sub-region is measured three times at different times, and the average of the three measurement results is taken as the actual airflow data of the sub-region, that is, the sample micro airflow data.

[0094] Meanwhile, by connecting to the real-time data interface of the local meteorological station, macroscopic airflow data (including large-scale wind direction and wind speed) of the target flight area is obtained synchronously for each measurement, which serves as sample macroscopic airflow data to ensure that each set of sample microscopic airflow data can form an "input-output" sample pair with the corresponding sample macroscopic airflow data.

[0095] For example, when the sample macroscopic airflow data is "east wind 2.7 m / s", the sample microscopic airflow data of a certain high-rise building gap sub-area is "east wind southerly 1.8 m / s", the sample microscopic airflow data of a certain intersection sub-area is "east wind 2.4 m / s", and the sample microscopic airflow data of a certain square surrounding sub-area is "east wind 2.6 m / s", thus covering the airflow differences in different building environments.

[0096] Furthermore, the simulation results of the original airflow simulator are corrected using sample macroscopic airflow data and sample microscopic airflow data to obtain an airflow simulator for the target flight area.

[0097] Specifically, all sample macroscopic airflow data are first input into the original airflow simulator one by one to obtain the predicted microscopic airflow data output by the simulator.

[0098] Furthermore, the predicted micro-airflow data is compared with the sample micro-airflow data measured in the field, and the wind speed error between the two is calculated using the formulas "wind speed error = |predicted wind speed - actual wind speed|, wind direction error = |predicted wind direction - actual wind direction|".

[0099] If the wind speed error in a certain sub-region exceeds 0.4 m / s or the wind direction error exceeds 8°, the key parameters in the original airflow simulator are adjusted. For example, the friction coefficient of the high-rise building walls is adjusted from 0.02 to 0.017, and the turbulence coefficient of the building gaps is adjusted from 0.005 to 0.0045. The sample macroscopic airflow data is then re-entered for simulation verification.

[0100] Meanwhile, the correction process is iterated repeatedly until the wind speed error of all sub-regions is ≤0.3m / s and the wind direction error is ≤5°. At this point, the original airflow simulator has been adapted to the building environment characteristics of the target flight area and has become the target flight area airflow simulator (the corrected simulator) with the required accuracy.

[0101] However, although the accuracy of the airflow simulator for the target flight area has been improved, the computational complexity has not been significantly reduced. Simulation of a single set of data still takes 3-5 minutes, and a rapid response model still needs to be built.

[0102] Therefore, it is necessary to further input the macroscopic airflow data of the target flight area into the corrected airflow simulator to obtain predicted airflow data for multiple sub-regions.

[0103] Specifically, macroscopic airflow data will continue to be collected at multiple times in the target flight area, with the collection period covering different weather conditions and different time periods (one set per hour within 24 hours), for 30 consecutive days, resulting in a total of 720 sets of macroscopic airflow data.

[0104] Furthermore, each set of macroscopic airflow data is input into the corrected airflow simulator to obtain the predicted microscopic airflow data of multiple typical sub-regions output by the airflow simulator, so as to ensure that each set of macroscopic airflow data can correspond to the complete microscopic airflow information of each sub-region.

[0105] Furthermore, the collected data are used to form a sample predicted airflow dataset. That is, "one set of macroscopic airflow data and the predicted microscopic airflow data of the corresponding number of sub-regions" are combined into one sample. A total of 720 samples are generated from 720 sets of macroscopic airflow data, forming the sample predicted airflow dataset.

[0106] Meanwhile, the input features for each sample are "macro wind speed and macro wind direction", and the output features are "micro wind speed and micro wind direction of the corresponding sub-region". All features are labeled with their corresponding data types and units to ensure that the dataset has a clear structure and can be directly used for model training.

[0107] Furthermore, a fast airflow simulator is trained using a sample predicted airflow dataset to achieve rapid and accurate output of micro-airflow data for the target flight area.

[0108] The method provided in this application embodiment, which "uses the sample prediction airflow dataset to train the fast airflow simulator and obtains micro-airflow data of multiple sub-regions in the target flight area", includes:

[0109] The rapid airflow simulator is built based on a neural network;

[0110] Based on the 3D building map and the sample predicted airflow dataset, the fast airflow simulator is trained to learn the mapping relationship between macroscopic airflow data and microscopic airflow data of multiple sub-regions until the fast airflow simulator converges.

[0111] Macroscopic airflow data is input into a fast airflow simulator, which predicts and outputs microscopic airflow data for multiple sub-regions of the target flight area.

[0112] In this embodiment of the application, in order to solve the problems of complex calculation and long calculation time of the original airflow simulator and the modified airflow simulator, and at the same time to ensure accurate output of micro airflow data of sub-regions, it is necessary to build a fast airflow simulator based on neural network to meet the dual requirements of UAV path planning for real-time and accurate airflow data, and to provide reliable data support for subsequent path cost calculation and optimization.

[0113] First, a rapid airflow simulator was built based on a neural network. Considering the complex and nonlinear changes in urban low-altitude airflow due to the influence of buildings, and the need for rapid response and prediction, a fully connected neural network was chosen as the basic framework of the simulator.

[0114] The fully connected neural network framework can effectively fit the complex relationship between macroscopic airflow and microscopic airflow in multiple sub-regions through the nonlinear transformation of multi-layer neurons. The model has a simple structure, fast inference speed, and can meet the real-time requirements of path planning.

[0115] Specifically, the fast airflow simulator is designed as follows: the input layer has two neurons, corresponding to the "standardized macro wind speed" and the "standardized macro wind direction" respectively. The wind direction needs to be converted into a value of 0-360° first, and then decomposed into two features, "sin (wind direction radian value)" and "cos (wind direction radian value)," through trigonometric functions to avoid model training bias caused by the periodicity of wind direction.

[0116] Meanwhile, three hidden layers are set: the first layer contains 32 neurons, the second layer contains 64 neurons, and the third layer contains 32 neurons. All of them use the ReLU activation function, which can effectively solve the gradient vanishing problem and improve the model training efficiency and fitting ability.

[0117] Secondly, the output layer has 2N neurons, corresponding to the "standardized micro wind speed" and "standardized micro wind direction" of each sub-region, respectively, to achieve a direct mapping from macro airflow input to micro airflow output in multiple sub-regions. Here, N is the number of sub-regions selected for the target flight area. For example, if N=50, the output layer has 100 neurons.

[0118] Furthermore, based on 3D building maps and sample predicted airflow datasets, a fast airflow simulator is trained to learn the mapping relationship.

[0119] Specifically, the first step is to perform spatial correlation processing on the sample predicted airflow dataset in conjunction with a 3D building map. Multiple sub-regions in the sample predicted airflow dataset have been selected based on the building distribution characteristics of the 3D building map. During training, the spatial attributes of these sub-regions in the 3D building map must be incorporated as implicit constraints into the model training process to ensure that the mapping relationship learned by the model accurately reflects the influence of the actual building environment on airflow in the target area.

[0120] For example, the model needs to autonomously identify the spatial attribute that "a certain sub-region is located between two 30-story buildings" and then learn the pattern that the micro airflow in this type of sub-region is prone to wind direction deflection and wind speed reduction due to building obstruction.

[0121] In the data preprocessing stage, all features in the sample predicted airflow dataset need to be standardized or normalized, that is, the macro wind speed and micro wind speed are normalized to the interval [0, 1] according to the formula "(actual value - sample minimum value) / (sample maximum value - sample minimum value)".

[0122] Meanwhile, the decomposed wind direction features (sin value, cos value) are also normalized to the [0, 1] interval to eliminate training bias caused by differences in the numerical range of different features and improve the stability of model training.

[0123] Furthermore, the dataset is divided into a training set and a test set in an 8:2 ratio. The training set is used for learning model parameters, and the test set is used to evaluate the model's generalization ability. For example, if the sample airflow prediction dataset contains 500 samples, it will be divided into 400 training samples and 100 test samples.

[0124] During training, mean squared error (MSE) is used as the loss function, and the calculation formula is "MSE=1 / (2N)×Σ(predicted feature value-true feature value)". 2 This function quantifies the degree of deviation between the micro-airflow data output by the model and the actual micro-airflow data of the sample.

[0125] For example, if the actual micro wind speed in a certain sub-region is 2.1 m / s and the predicted value is 1.9 m / s, and the actual micro wind direction (cosine value after decomposition) is 0.8 and the predicted value is 0.75, then the wind speed error term for this sub-region is (1.9 - 2.1). 2 =0.04, the wind direction error term is (0.75-0.8) 2 =0.0025, the two errors are included in the total loss value to guide the adjustment of model parameters.

[0126] Meanwhile, the Adam optimizer is used to update the model parameters. The initial learning rate is set to 0.001, and the learning rate is reduced to 0.9 times the original rate every 200 iterations to avoid model oscillation caused by an excessively high learning rate or slow training caused by an excessively low learning rate.

[0127] In addition, the model performance needs to be dynamically evaluated during the training process. That is, the "airflow prediction accuracy" of the test set is calculated after each iteration. The judgment criteria are "micro wind speed prediction error ≤ 0.2m / s and micro wind direction prediction error ≤ 3°". The percentage of sub-regions in the test set that meet the criteria is the accuracy of the current round.

[0128] For example, during the initial 100th training round, the test set accuracy was only 65%, mainly due to a large prediction error in the airflow of the "high-rise dense area sub-region". By adjusting the number of neurons in the hidden layer (increasing the number of neurons in the second layer from 64 to 80) and increasing the number of training rounds, the test set accuracy improved to 88% by the 500th round, and the prediction error of the high-rise dense area sub-region was significantly reduced. When training continued to the 1000th round, the test set accuracy stabilized at over 95% and remained stable for 30 consecutive rounds without significant fluctuations. At this point, the model had fully learned the mapping relationship between macroscopic and microscopic airflow, and the fast airflow simulator training was deemed to have converged.

[0129] Furthermore, macroscopic airflow data is input into a fast airflow simulator that has been trained and converged, and the microscopic airflow data of multiple sub-regions of the target flight area are predicted and output.

[0130] In practical applications, it is only necessary to obtain the macroscopic airflow data of the target flight area at the current moment through the weather station, process it according to the preprocessing method during training (standardization, wind direction decomposition), and then input it into the fast airflow simulator. The simulator can complete the calculation within 10 seconds and output the microscopic airflow data of all sub-regions.

[0131] For example, when the macroscopic airflow is input as "north wind 3.2 m / s", the simulator outputs results such as "sub-region 1 (between high-rise buildings): northerly wind 2.3 m / s", "sub-region 2 (intersection): northerly wind 3.0 m / s", and "sub-region 3 (around the square): northerly wind 2.9 m / s". The error between these results and the output of the corrected airflow simulator is controlled within wind speed ≤ 0.2 m / s and wind direction ≤ 3°. This satisfies the real-time requirements and ensures that the accuracy can support the subsequent path cost calculation, providing key airflow data support for the optimization of urban low-altitude UAV flight paths.

[0132] Finally, the macroscopic airflow data of the target flight area at the current moment is input into the constructed fast airflow simulator. Through the neural network inference calculation of the fast airflow simulator, the microscopic airflow data of multiple sub-regions in the target flight area are output to achieve rapid and accurate acquisition of microscopic airflow data in urban low-altitude environment, providing key airflow data support for subsequent calculation of the path cost of the UAV's initial flight path.

[0133] S130: Based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, and combined with the performance parameters of the target UAV, calculate the path cost of the target UAV executing the initial flight path;

[0134] In this embodiment of the application, in the context of complex urban low-altitude building environment and variable airflow, in order to quantify the comprehensive cost of the UAV executing the initial flight path, it is necessary to combine real-time micro airflow data and UAV performance parameters to calculate the path cost through multi-dimensional calculations, which will serve as the core evaluation criterion for subsequent path iteration optimization.

[0135] Specifically, firstly, based on a fast airflow simulator, micro-airflow data of N sub-regions traversed along the initial flight path of the UAV is acquired. This micro-airflow data consists of wind speed and direction data for the specific areas covered by the initial flight path at the current moment, accurately reflecting the actual airflow conditions faced by the UAV during flight.

[0136] Furthermore, based on the performance parameters of the target UAV model and the micro airflow data of N sub-regions, the estimated time and energy consumption of the UAV in each sub-region are predicted.

[0137] Among them, the performance parameters of the target UAV model need to be clearly defined in advance, including the core indicators such as the climb rate, climb energy consumption, level flight speed, and level flight energy consumption of the UAV model. These parameters directly determine the operating efficiency of the UAV under different airflow and flight conditions.

[0138] At the same time, by combining the flight conditions and micro-airflow data of the initial flight path in each sub-region, the impact of airflow on flight is determined.

[0139] Furthermore, by summing the estimated time and energy consumption of each sub-region, the total time (path time) and total energy consumption (path energy consumption) of the UAV executing the initial flight path can be obtained, thus fully presenting the basic operating cost of the initial flight path.

[0140] Finally, based on preset weights, the path time and energy consumption values ​​are normalized and then weighted and summed to obtain the path cost of the initial flight path.

[0141] This step combines the effects of real-time airflow with the performance of the UAV, transforming the operational costs of the initial flight path into quantifiable path costs. This avoids cost estimation biases caused by ignoring airflow, ensuring that subsequent optimizations can effectively improve the efficiency and economy of UAV flight.

[0142] Step S130 in the method provided in this application embodiment includes:

[0143] Based on a fast airflow simulator, micro-airflow data of N sub-regions traversed along the initial flight path of the UAV are obtained.

[0144] Based on the UAV model and micro airflow data of N sub-regions, the estimated time and energy consumption of the target UAV in the N sub-regions on the initial flight path are estimated. The time and energy consumption of the N sub-regions are summed to obtain the path time and path energy consumption of the target UAV on the initial flight path.

[0145] Based on preset weights, the path time and energy consumption values ​​are normalized and then weighted and summed to obtain the path cost of the target UAV model when executing the initial flight path.

[0146] In this embodiment of the application, in order to reduce the overall operating cost of UAV flight and avoid the problem of excessive time or energy consumption due to ignoring the influence of airflow, the operating cost of the initial flight path needs to be converted into a quantifiable path cost to provide an accurate evaluation basis for subsequent path iteration optimization.

[0147] Specifically, firstly, based on the constructed fast airflow simulator, micro airflow data of N sub-regions passed through on the initial flight path of the UAV are obtained.

[0148] The initial flight path clearly defines the flight trajectory of the UAV from the starting point to the end point. This trajectory will traverse multiple sub-regions within the target flight area. The value of N needs to be determined based on the precision of the sub-region division. For example, each 50-meter trajectory corresponds to one sub-region, and a 1000-meter initial path corresponds to 20 (N=20) sub-regions.

[0149] Meanwhile, the fast airflow simulator will output the specific wind speed and direction for each sub-region based on the current macro airflow data. For example, "Sub-region 1: wind direction southeast, wind speed 1.5m / s; Sub-region 2: wind direction northeast, wind speed 2.2m / s; Sub-region 3: wind direction southeast-southeast, wind speed 1.8m / s". This data can accurately reflect the actual airflow conditions faced by the UAV when flying in each sub-region, providing real-time and accurate airflow basis for subsequent time and energy consumption prediction.

[0150] Furthermore, based on the performance parameters of the target UAV model and the micro airflow data of N sub-regions, the estimated time and energy consumption of the UAV in each sub-region are predicted.

[0151] The method provided in this application, which "estimates the expected time and energy consumption of a target UAV in N sub-regions along its initial flight path based on the UAV model and micro-airflow data of N sub-regions", includes:

[0152] Based on the model of the target UAV, obtain the performance parameters of the target UAV, wherein the performance parameters include climb rate, climb energy consumption, level flight speed and level flight energy consumption;

[0153] Based on the initial flight path of the target UAV, the climb altitude, level flight distance, and flight direction of the target UAV in N sub-regions, as well as the micro airflow data of the N sub-regions, are obtained as the flight condition information of the target UAV on the initial flight path.

[0154] Based on the flight condition information of the target UAV and its performance parameters, the estimated time and energy consumption of the target UAV in N sub-regions are calculated.

[0155] In this embodiment of the application, in order to accurately quantify the actual flight cost of the UAV under the influence of low-altitude airflow in the city, it is necessary to first clarify the performance benchmark of the UAV and the flight conditions of each sub-region, and then adjust the calculation results in combination with real-time micro airflow data to obtain the expected time and expected energy consumption that are in line with reality.

[0156] First, based on the target UAV model, its performance parameters are obtained, specifically including climb rate, climb energy consumption, level flight speed, and level flight energy consumption. Different UAV models have different power systems, payload capacities, and aerodynamic designs, which directly determine their flight speed and energy consumption levels; therefore, it is necessary to accurately match the performance parameters corresponding to the specific model.

[0157] For example, if the target drone model is "XX-200", its performance parameters can be obtained by consulting the technical manual of this model: climb speed of 4m / s, climb energy consumption of 80mAh / m, level flight speed of 12m / s, and level flight energy consumption of 15mAh / s. These performance parameters are the basic benchmarks for calculating the flight time and energy consumption of the drone in a windless environment, ensuring that subsequent calculations have a clear quantitative basis.

[0158] Furthermore, based on the initial flight path of the target UAV, flight condition information for N sub-regions is obtained.

[0159] The initial flight path clearly defines the complete trajectory of the UAV from the starting point to the end point. Combined with the 3D building map and path division rules, such as dividing the trajectory into 40-meter sub-regions, the 1200-meter initial path corresponds to 30 (N=30) sub-regions, and the flight status of the UAV in each sub-region can be determined one by one.

[0160] Specifically, for sub-region 1, analysis of the 3D building map shows that the drone needs to climb from an altitude of 490m to 500m, then fly horizontally for 30m, with the flight direction being due south; the micro-airflow data of this sub-region obtained by the fast airflow simulator is "wind direction due south, wind speed 1.2m / s".

[0161] Similarly, in sub-region 2, the drone maintained an altitude of 500m and flew horizontally for 40m, with the flight direction being southeast, and the micro-airflow data being "northwest wind direction, wind speed 1.8m / s"; in sub-region 3, the drone climbed from an altitude of 500m to 508m, flew horizontally for 25m, with the flight direction being due east, and the micro-airflow data being "due east wind direction, wind speed 0.9m / s".

[0162] Through the above steps, complete flight condition information including "climb altitude - level flight distance - flight direction - micro airflow data (wind speed, wind direction)" is constructed for each sub-region to ensure that the impact of airflow on flight can be accurately correlated to specific regions.

[0163] Furthermore, based on flight condition information and performance parameters, the estimated time and energy consumption for each sub-region are calculated.

[0164] Specifically, the calculation needs to be carried out in two steps: "calculation of basic values ​​in windless environment" and "calculation combined with airflow adjustment". First, the flight cost in windless state is determined based on the performance parameters of the UAV. Then, the estimated flight time and energy consumption are adjusted according to the relative relationship between the micro airflow data of the sub-region and the flight direction to ensure that the results are consistent with the airflow impact of the city's low altitude.

[0165] Taking the target UAV model "XY-300" as an example, its performance parameters are: climb speed 3.5m / s, climb energy consumption 75mAh / m, level flight speed 14m / s, and level flight energy consumption 18mAh / s. Taking sub-region 5 on the initial flight path as an example, the flight conditions of this sub-region are: climb altitude 6m, level flight distance 35m, flight direction due north, and micro-airflow data of "wind direction due north, wind speed 1.5m / s" (tailwind).

[0166] First, calculate the basic time and energy consumption in a windless environment: Climb time = Climb altitude / Climb speed = 6m / 3.5m / s ≈ 1.71s; Climb energy consumption = Climb altitude × Climb energy consumption = 6m × 75mAh / m = 450mAh; Level flight time = Level flight distance / Level flight speed = 35m / 14m / s = 2.5s; Level flight energy consumption = Level flight time × Level flight energy consumption = 2.5s × 18mAh / s = 45mAh; The total time for this sub-region in a windless state is ≈ 1.71s + 2.5s = 4.21s, and the total energy consumption is 450mAh + 45mAh = 495mAh.

[0167] Secondly, considering the tailwind adjustment calculations, since the climb process is less affected by airflow, the time and energy consumption are still calculated based on the windless value. During level flight, the tailwind will increase the actual flight speed. Calculated using "actual level flight speed = level flight speed + wind speed × tailwind gain coefficient (taken as 0.65 based on the aerodynamic characteristics of the UAV)," the actual level flight speed = 14m / s + 1.5m / s × 0.65 = 14.975m / s; the actual level flight time = level flight distance / actual level flight speed = 35m / 14.975m / s ≈ 2.34s; the energy consumption during level flight decreases proportionally with the decrease in time, and the actual level flight energy consumption = 2.34s × 18mAh / s ≈ 42.12mAh; finally, the estimated time for sub-region 5 is ≈ 1.71s + 2.34s = 4.05s, and the estimated energy consumption is 450mAh + 42.12mAh ≈ 492.12mAh.

[0168] Through the calculation process described above, the corresponding estimated time and energy consumption can be accurately calculated for the specific flight conditions and micro-airflow data of each sub-region, ensuring that the flight cost of each sub-region can truly reflect the matching between airflow impact and UAV performance.

[0169] Furthermore, after completing the estimated time and energy consumption calculations for all N sub-regions, the estimated time for all sub-regions is summed to obtain the total time (i.e., path time) for the target UAV to execute the initial flight path.

[0170] At the same time, the estimated energy consumption of all sub-regions is summed to obtain the total energy consumption of the initial flight path (i.e., path energy consumption).

[0171] For example, the initial flight path passes through 30 (N=30) sub-regions. The total estimated time for each sub-region is 480 seconds, and the total estimated energy consumption for each sub-region is 6200mAh.

[0172] Furthermore, the path time and energy consumption values ​​are normalized based on preset weights and then weighted and summed to obtain the path cost.

[0173] Specifically, the path time and energy consumption values ​​are first normalized by selecting historical data of similar UAVs flying within the target flight area to determine a reasonable range of values.

[0174] For example, if the reasonable range for path time is 300-600 seconds and the reasonable range for path energy consumption is 4000-8000mAh, the normalized value is calculated using the formula "normalized value = (actual value - minimum value of the range) / (maximum value of the range - minimum value of the range)". If the current path time is 480 seconds, then the normalized time = (480-300) / (600-300) = 0.6; if the current path energy consumption is 6200mAh, then the normalized energy consumption = (6200-4000) / (8000-4000) = 0.55.

[0175] The preset weights need to be set according to the task requirements. For example, logistics and delivery tasks focus more on timeliness, so the time consumption weight is set to 0.6 and the energy consumption weight is set to 0.4; emergency reconnaissance tasks focus more on endurance and stability, so the time consumption weight is set to 0.3 and the energy consumption weight is set to 0.7.

[0176] For example, in the scenario of logistics and delivery tasks, the path cost = 0.6 × 0.6 + 0.4 × 0.55 = 0.58. This value transforms the costs of two different dimensions, path time and path energy consumption, into a single quantifiable and comparable indicator, providing a clear evaluation basis for the iterative optimization of the initial flight path.

[0177] S140: Using the 3D building map of the target flight area and the target UAV model as constraints, and with the goal of minimizing path cost, iteratively optimize the initial flight path to obtain the optimal flight path.

[0178] In this embodiment of the application, in order to enable the UAV flight path to avoid building obstacles and adapt to its own performance and actual airflow conditions, and achieve the flight goal of the lowest path cost, it is necessary to use the three-dimensional building map and UAV performance as constraints, generate and screen candidate flight paths through multiple rounds of iteration, and finally obtain the optimal flight path that takes into account safety, efficiency and economy.

[0179] Specifically, based on a 3D building map of the target flight area, the initial flight path is first divided into several flight nodes. Each flight node corresponds to specific coordinates in the 3D building map.

[0180] Furthermore, using the 3D building map as a constraint, within a preset flight node disturbance threshold, multiple flight nodes in the original flight path are locally disturbed to generate Q alternative flight paths.

[0181] Furthermore, based on the performance parameters of the target UAV model, the path costs of Q alternative flight paths are calculated by first acquiring the micro airflow data of the sub-regions through which the path passes, then estimating the time and energy consumption of each sub-region, and finally calculating the path cost of T alternative flight paths through normalized weighted summation, while retaining the T alternative flight paths with the lowest path cost (Q≥3T, T≥2).

[0182] Furthermore, the path cost is calculated for each of the T candidate flight paths, and the loop of "path cost-based screening - local perturbation - path cost-based screening" is continued until the preset convergence threshold is reached, thus completing the iterative optimization of the UAV flight path.

[0183] Finally, the alternative flight path with the lowest path cost generated during the iterative optimization process is selected as the optimal flight path for actual drone flight.

[0184] Step S140 in the method provided in this application embodiment includes:

[0185] Based on the 3D building map of the target flight area, the initial flight path is divided into several flight nodes;

[0186] Using the three-dimensional building map as a constraint, within a preset flight node disturbance threshold, multiple flight nodes in the original flight path are locally disturbed to generate Q alternative flight paths.

[0187] Based on the performance parameters of the target UAV model, calculate the path cost of Q alternative flight paths, and retain the T alternative flight paths with the lowest path cost, where Q≥3T and T≥2;

[0188] The path cost is calculated for each of the T candidate flight paths, and the process of filtering based on path cost, applying local perturbations, and filtering based on path cost is repeated until the preset convergence threshold is reached, thus completing the iterative optimization of the UAV flight path.

[0189] The alternative flight path with the lowest path cost generated during the iterative optimization process is selected as the optimal flight path for flight.

[0190] In this embodiment of the application, in order to accurately optimize the flight path of the UAV in the complex urban low-altitude building environment, it is necessary to use the three-dimensional building map and the performance of the UAV as constraints, generate and screen alternative flight paths through multiple rounds of iteration, so as to gradually approach the optimal solution and provide the UAV with a safe, efficient and economical optimal flight route.

[0191] First, based on the three-dimensional building map of the target flight area, the initial flight path is divided into several flight nodes.

[0192] Specifically, when splitting the road, the node spacing should be set according to the principle of "uniform coverage and reinforcement of key locations". For example, for road sections with sparse building distribution, the node spacing can be set to 100 meters to ensure that the overall path is controllable; for road sections with dense buildings and sudden changes in airflow, the node spacing should be reduced to 50 meters, or even additional nodes should be added at building corners and height changes to accurately capture key control points of the path.

[0193] For example, an initial flight path of 1200 meters is divided into 15 flight nodes (including the start and end points). Each node corresponds to a unique coordinate in three-dimensional space, such as "Node 3: 104.07°E, 30.66°N, 510m altitude" and "Node 8: 104.08°E, 30.67°N, 508m altitude".

[0194] Furthermore, using the 3D building map as a constraint, within a preset flight node disturbance threshold, multiple flight nodes in the original flight path are locally disturbed to generate Q alternative flight paths.

[0195] The constraints of the 3D building map are applied throughout the entire disturbance process. Before the disturbance, the safe disturbance range of each flight node must be determined: in the horizontal direction, the distance between the flight node and the surrounding buildings must always be greater than 10 meters (to avoid the path crossing the building safety zone); in the vertical direction, the altitude of the flight node must be more than 5 meters higher than the tallest surrounding building (to prevent the risk of low-altitude collision). This safe range is the preset flight node disturbance threshold.

[0196] For example, node 5, located between two 20-story buildings (500-560m above sea level), has a horizontal disturbance threshold of 5-8 meters (ensuring a distance of no less than 10 meters from the buildings on both sides) and a vertical disturbance threshold of 3-5 meters (maintaining an altitude between 565-570m).

[0197] When performing the perturbation operation, 60% of the non-starting / ending nodes in the original path are randomly selected, and the horizontal coordinates (latitude and longitude) and vertical coordinates (altitude) of each selected node are independently fine-tuned. For example, the coordinates of node 5 are adjusted from "104.072°E, 30.663°N, altitude 567m" to "104.075°E, 30.665°N, altitude 569m", and the coordinates of node 9 are adjusted from "104.081°E, 30.672°N, altitude 566m" to "104.079°E, 30.674°N, altitude 568m".

[0198] Furthermore, by perturbing different combinations of nodes, Q alternative flight paths with different structures are finally generated, and all alternative flight paths are tested for collisions on the 3D building map to ensure that there is no risk of building collisions.

[0199] Furthermore, based on the performance parameters of the target UAV model obtained, the path cost of Q alternative flight paths is calculated, and the T alternative flight paths with the lowest path cost are retained (Q≥3T, T≥2).

[0200] For each candidate flight path, micro-airflow data of N sub-regions along the path is first obtained through a fast airflow simulator. Then, the estimated time and energy consumption of each sub-region are estimated by combining the UAV performance parameters and the flight conditions of the sub-regions. The total time and energy consumption of the path are summed to obtain the total time and energy consumption of the path. Finally, the total time and energy consumption are normalized based on preset weights and then weighted and summed to obtain the path cost of the candidate path.

[0201] For example, among the 30 candidate flight paths (Q=30), the top 6 paths with the lowest cost (T=6, satisfying 30≥3×6) are retained, while the remaining 24 are eliminated due to excessive cost. This step can quickly screen out candidate paths with optimization potential, significantly reducing the computational load of subsequent iterations.

[0202] Furthermore, the path cost is calculated for each of the T candidate flight paths, and the process of "filtering based on path cost - performing local perturbation - filtering based on path cost again" continues until the preset convergence threshold is reached, thus completing the iterative optimization of the UAV flight path.

[0203] Each round of the cycle is based on the T currently retained paths, repeating the process of "node perturbation - generating new alternative paths - cost calculation - screening and retention". For example, the 6 paths retained in the first round generate 5 new perturbation paths for each path (a total of 30 new paths). After calculating the cost of each new path, the 6 paths with the lowest cost are still retained.

[0204] This process is iterated continuously, while the change in the minimum path cost in each round is monitored in real time. If the difference in the minimum path cost is less than 0.015 (normalized cost unit) in 6 consecutive iterations, the preset convergence threshold is reached.

[0205] For example, the lowest path cost is 0.52 in the initial iteration, drops to 0.48 in the second round, 0.45 in the third round, 0.43 in the fourth round, 0.42 in the fifth round, 0.418 in the sixth round, 0.417 in the seventh round, and 0.416 in the eighth round. At this point, the cost difference for six consecutive rounds (rounds 3-8) is less than 0.015, indicating that the path optimization has approached the optimal state, and it is difficult to significantly reduce the cost by continuing the iteration, so the iteration process can be stopped.

[0206] Finally, the alternative flight path with the lowest path cost generated during the iterative optimization process is selected as the optimal flight path for actual UAV flight control.

[0207] The optimal flight path obtained through the above steps not only meets the obstacle avoidance requirements of the 3D building map (maintaining a safe distance from all buildings), but also adapts to the performance parameter limitations of the target UAV model. At the same time, due to its lowest path cost, it enables the UAV to fly efficiently and with low energy consumption under the current airflow conditions.

[0208] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0209] This application proposes a method for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs). First, a surveying grid flight path is planned, and a surveying UAV scans the target area's buildings from multiple perspectives. A 3D building map is obtained by integrating a triangular mesh model with coordinate information. After marking the start and end points, an initial flight path with the shortest distance is generated, constrained by avoiding buildings. Next, based on the 3D building map and a fluid dynamics model, an initial airflow simulator is built. Sample airflow data is collected using a wind-measuring UAV to correct the simulator. Then, a neural network is trained using a sample prediction dataset to obtain a fast airflow simulator capable of quickly outputting micro-airflow data for sub-regions. Real-time micro-airflow data can be obtained by inputting macro-airflow data. Subsequently, based on the target UAV's performance parameters, the estimated time and energy consumption for each sub-region are accumulated and normalized with a weighted sum to obtain the path cost. Finally, the initial path is broken down into flight nodes, and alternative flight paths are generated through local perturbations. The flight cost is calculated, and the process is repeated through screening and iteration until convergence. The lowest-cost path is selected as the optimal flight path for controlling UAV flight, adapting to the needs of urban low-altitude flight.

[0210] The method provided in this application, through the technical solution of "constructing a 3D building map - constructing a fast airflow simulator - calculating path cost - iterative optimization", solves the problem that traditional path planning ignores the influence of urban buildings and airflow, resulting in short endurance or low safety. It enables efficient, safe and low-consumption flight of urban low-altitude UAVs, and provides reliable path planning support for the actual application scenarios of urban UAVs.

[0211] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of the urban low-altitude UAV flight path optimization method provided in Embodiment 1, this application also provides an urban low-altitude UAV flight path optimization system, specifically including:

[0212] The initial path acquisition module 01 is used to acquire a 3D building map of the target flight area. Based on the 3D building map, the flight start point and end point are marked. With the start point, end point, and 3D building map as constraints, the shortest flight path is acquired as the initial flight path.

[0213] The airflow data acquisition module 02 is used to obtain a fast airflow simulator of the target flight area based on a three-dimensional building map of the target flight area, and to obtain micro airflow data of multiple sub-regions in the target flight area based on the fast airflow simulator and macro airflow data. Both the macro airflow data and the micro airflow data include wind speed and wind direction.

[0214] The path cost calculation module 03 is used to calculate the path cost of the target UAV executing the initial flight path based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, combined with the performance parameters of the target UAV model.

[0215] The path iteration optimization module 04 is used to iteratively optimize the initial flight path with the constraints of the three-dimensional building map of the target flight area and the target type of UAV, and with the goal of minimizing the path cost, to obtain the optimal flight path.

[0216] In one embodiment, the initial path acquisition module 01 is further configured to:

[0217] Obtain the target flight area of ​​the UAV; plan a mapping grid route based on the flight planning software and the range and shape of the target flight area; use the mapping UAV to perform multi-view overlapping scans of the buildings in the target flight area based on the mapping grid route to obtain the triangular mesh model and coordinate information of each building in the target flight area; integrate the triangular mesh model and coordinate information of each building in the target flight area to obtain a three-dimensional building map of the target flight area.

[0218] In one embodiment, the airflow data acquisition module 02 is further configured to:

[0219] Extract macroscopic airflow data of the target flight area; train a fast airflow simulator of the target flight area based on a 3D building map of the target flight area; input the macroscopic airflow data of the target flight area into the fast airflow simulator to obtain microscopic airflow data of multiple sub-regions in the target flight area.

[0220] Furthermore, the airflow data acquisition module 02 also includes:

[0221] Based on a 3D building map of the target flight area and combined with a fluid dynamics model, an original airflow simulator is established. Using a wind-measuring UAV equipped with airflow sensors, actual airflow data from multiple sub-regions of the target flight area are measured as sample micro-airflow data, and macro-airflow data of the target flight area is acquired in real time as sample macro-airflow data. The simulation results of the original airflow simulator are corrected using the sample macro-airflow data and sample micro-airflow data to obtain an airflow simulator for the target flight area. The macro-airflow data of the target flight area is input into the airflow simulator to obtain predicted airflow data for multiple sub-regions of the target flight area. Macro-airflow data from multiple moments in the target flight area, as well as the micro-airflow data for multiple sub-regions predicted by the airflow simulator corresponding to the macro-airflow data at multiple moments, are collected as a sample predicted airflow dataset. Using the sample predicted airflow dataset, the fast airflow simulator is trained to obtain micro-airflow data for multiple sub-regions of the target flight area.

[0222] Furthermore, the airflow data acquisition module 02 also includes:

[0223] A rapid airflow simulator is built based on a neural network. Based on the 3D building map and the sample predicted airflow dataset, the rapid airflow simulator is trained to learn the mapping relationship between macroscopic airflow data and microscopic airflow data of multiple sub-regions until the rapid airflow simulator converges. Macroscopic airflow data is input into the rapid airflow simulator to predict and output microscopic airflow data of multiple sub-regions of the target flight area.

[0224] In one embodiment, the path cost calculation module 03 is further used for:

[0225] Based on a fast airflow simulator, micro-airflow data of N sub-regions along the initial flight path of a UAV are obtained. Based on the UAV model and the micro-airflow data of the N sub-regions, the estimated time and energy consumption of the target UAV model in the N sub-regions of the initial flight path are estimated. The time and energy consumption of the N sub-regions are summed to obtain the path time and path energy consumption of the target UAV model on the initial flight path. Based on preset weights, the path time and path energy consumption values ​​are normalized and then weighted and summed to obtain the path cost of the target UAV model when executing the initial flight path.

[0226] Furthermore, the path cost calculation module 03 also includes:

[0227] Based on the target UAV model, obtain the target UAV's performance parameters, including climb rate, climb energy consumption, level flight speed, and level flight energy consumption. Based on the target UAV's initial flight path, obtain the target UAV's climb altitude, level flight distance, flight direction, and micro-airflow data for N sub-regions, as well as the flight condition information of the target UAV on the initial flight path. Based on the target UAV's flight condition information and the UAV's performance parameters, calculate the expected time and expected energy consumption of the target UAV in the N sub-regions.

[0228] In one embodiment, the path iteration optimization module 04 is further configured to:

[0229] Based on a 3D building map of the target flight area, the initial flight path is divided into several flight nodes. Using the 3D building map as a constraint, within a preset flight node perturbation threshold, multiple flight nodes in the original flight path are locally perturbed to generate Q candidate flight paths. Based on the performance parameters of the target UAV model, the path cost of the Q candidate flight paths is calculated, and the T candidate flight paths with the lowest path cost are retained, where Q ≥ 3T and T ≥ 2. The path cost of each of the T candidate flight paths is calculated, and a cycle of path cost-based selection, local perturbation, and path cost-based selection continues until a preset convergence threshold is reached, completing the iterative optimization of the UAV flight path. The candidate flight path with the lowest path cost generated during the iterative optimization process is taken as the optimal flight path and used for flight.

[0230] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0231] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application. This specification and drawings are merely illustrative examples of this application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application. Thus, if these modifications and variations of this application fall within the scope of this application and its equivalents, this application intends to include these modifications and variations.

Claims

1. A method for optimizing the flight path of urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, include: Obtain a 3D building map of the target flight area. Based on the 3D building map, mark the start and end points of the flight. Using the start and end points and the 3D building map as constraints, obtain the shortest flight path as the initial flight path. Based on a 3D building map of the target flight area, a fast airflow simulator of the target flight area is obtained. Based on the fast airflow simulator and macro airflow data, micro airflow data of multiple sub-regions in the target flight area are obtained. Both the macro airflow data and the micro airflow data include wind speed and wind direction. Based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, combined with the performance parameters of the target UAV, the path cost of the target UAV executing the initial flight path is calculated. Using a 3D building map of the target flight area and the target UAV model as constraints, and with the goal of minimizing path cost, the initial flight path is iteratively optimized to obtain the optimal flight path; Based on a 3D building map of the target flight area, a fast airflow simulator for the target flight area is obtained. Based on the fast airflow simulator and macroscopic airflow data, microscopic airflow data for multiple sub-regions within the target flight area are obtained, including: Extract macroscopic airflow data of the target flight area; A fast airflow simulator for the target flight area is trained based on a 3D building map of the target flight area. The macroscopic airflow data of the target flight area is input into the fast airflow simulator to obtain the microscopic airflow data of multiple sub-regions in the target flight area; Based on a 3D building map of the target flight area, a fast airflow simulator for the target flight area is trained, including: Based on a 3D building map of the target flight area, combined with a fluid dynamics model, an original airflow simulator was established. Based on the wind measurement drone, equipped with airflow sensing equipment, the actual airflow data of multiple sub-regions of the target flight area is measured as sample micro airflow data, and the macro airflow data of the target flight area is acquired in real time as sample macro airflow data; Using sample macroscopic airflow data and sample microscopic airflow data, the simulation results of the original airflow simulator are corrected to obtain an airflow simulator for the target flight area. Macroscopic airflow data of the target flight area is input into the airflow simulator to obtain predicted airflow data for multiple sub-regions of the target flight area; Collect macroscopic airflow data of the target flight area at multiple times, as well as microscopic airflow data of multiple sub-regions predicted by the airflow simulator corresponding to the macroscopic airflow data at multiple times, as a sample predicted airflow dataset; Using the sample predicted airflow dataset, the fast airflow simulator is trained to obtain micro airflow data for multiple sub-regions in the target flight area; Based on the micro-airflow data of multiple sub-regions within the target flight area at the current moment, and combined with the performance parameters of the target UAV model, the path cost of the target UAV executing the initial flight path is calculated, including: Based on a fast airflow simulator, micro airflow data of N sub-regions along the initial flight path of the UAV are obtained; Based on the UAV model and micro airflow data of N sub-regions, the estimated time and energy consumption of the target UAV in the N sub-regions on the initial flight path are estimated. The time and energy consumption of the N sub-regions are summed to obtain the path time and path energy consumption of the target UAV on the initial flight path. Based on preset weights, the path time and energy consumption values ​​are normalized and then weighted and summed to obtain the path cost of the target UAV when executing the initial flight path. Based on the UAV model and micro-airflow data for N sub-regions, the estimated time and energy consumption of the UAV in the N sub-regions along its initial flight path are predicted, including: Based on the model of the target UAV, obtain the performance parameters of the target UAV, wherein the performance parameters include climb rate, climb energy consumption, level flight speed and level flight energy consumption; Based on the initial flight path of the target UAV, the climb altitude, level flight distance, and flight direction of the target UAV in N sub-regions, as well as the micro airflow data of the N sub-regions, are obtained as the flight condition information of the target UAV on the initial flight path. Based on the flight condition information of the target UAV and its performance parameters, the estimated time and energy consumption of the target UAV in N sub-regions are calculated.

2. The method for optimizing the flight path of a low-altitude unmanned aerial vehicle (UAV) in urban areas as described in claim 1, characterized in that, Obtain a 3D building map of the target flight area, including: Obtain the target flight area of ​​the drone; Based on flight planning software and the extent and shape of the target flight area, plan and map the grid flight path; Using a surveying drone, based on the surveying grid flight path, the buildings in the target flight area are scanned from multiple perspectives with overlapping data to obtain the triangular mesh model and coordinate information of each building in the target flight area. Based on the triangular mesh model and coordinate information of each building in the target flight area, a three-dimensional building map of the target flight area is obtained.

3. The method for optimizing the flight path of urban low-altitude unmanned aerial vehicles as described in claim 1, characterized in that, Using the sample predicted airflow dataset, the fast airflow simulator is trained to obtain micro-airflow data for multiple sub-regions within the target flight area, including: The rapid airflow simulator was built based on a neural network. Based on the 3D building map and the sample predicted airflow dataset, the fast airflow simulator is trained to learn the mapping relationship between macroscopic airflow data and microscopic airflow data of multiple sub-regions until the fast airflow simulator converges. Macroscopic airflow data is input into a fast airflow simulator, which predicts and outputs microscopic airflow data for multiple sub-regions of the target flight area.

4. The method for optimizing the flight path of urban low-altitude unmanned aerial vehicles as described in claim 1, characterized in that, Using a 3D building map of the target flight area and the target UAV model as constraints, and aiming at minimizing path cost, the initial flight path is iteratively optimized to obtain the optimal flight path, including: Based on the 3D building map of the target flight area, the initial flight path is divided into several flight nodes; Using the three-dimensional building map as a constraint, within a preset flight node disturbance threshold, multiple flight nodes in the original flight path are locally disturbed to generate Q alternative flight paths. Based on the performance parameters of the target UAV model, calculate the path cost of Q alternative flight paths, and retain the T alternative flight paths with the lowest path cost, where Q≥3T and T≥2; The path cost is calculated for each of the T candidate flight paths, and the process of filtering based on path cost, applying local perturbations, and filtering based on path cost is repeated until the preset convergence threshold is reached, thus completing the iterative optimization of the UAV flight path. The alternative flight path with the lowest path cost generated during the iterative optimization process is selected as the optimal flight path for flight.

5. A flight path optimization system for urban low-altitude unmanned aerial vehicles (UAVs), characterized in that, The system is used to execute the urban low-altitude unmanned aerial vehicle flight path optimization method according to any one of claims 1-4, the system comprising: The initial path acquisition module is used to acquire a 3D building map of the target flight area. Based on the 3D building map, the flight start and end points are marked. Using the start and end points and the 3D building map as constraints, the shortest flight path is acquired as the initial flight path. The airflow data acquisition module is used to obtain a fast airflow simulator of the target flight area based on a three-dimensional building map of the target flight area, and to obtain micro airflow data of multiple sub-regions in the target flight area based on the fast airflow simulator and macro airflow data. Both the macro airflow data and the micro airflow data include wind speed and wind direction. The path cost calculation module is used to calculate the path cost of the target UAV executing the initial flight path based on the micro airflow data of multiple sub-regions in the target flight area at the current moment, combined with the performance parameters of the target UAV model. The path iteration optimization module is used to iteratively optimize the initial flight path with the constraints of a 3D building map of the target flight area and the target UAV model, and with the goal of minimizing the path cost, to obtain the optimal flight path.

Images