Path planning method and related device for cellular access type unmanned aerial vehicle in urban environment

Abstract

The application provides a path planning method for a cellular access type unmanned aerial vehicle in an urban environment and a related device, and relates to the field of unmanned aerial vehicles. By constructing a flight risk degree model, the optimal path planning problem is modeled as a minimum flight distance model, and then the minimum flight distance model is decomposed into a global cost function for global planning and a local cost function for local planning. Therefore, before the unmanned aerial vehicle performs a flight task, the global cost function can be solved based on graph theory to obtain a base station sequence of the unmanned aerial vehicle through global planning, and then when the unmanned aerial vehicle starts to perform the flight task, the local cost function can be solved based on deep reinforcement learning to plan a flight path of the unmanned aerial vehicle in the signal coverage range of the current access base station in real time. In this way, the optimal flight path of the unmanned aerial vehicle in a complex urban environment is solved, and the unmanned aerial vehicle is guided to complete the flight task in the urban area more safely and efficiently.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicles (UAVs), and more specifically, to a path planning method and related apparatus for cellular access UAVs in urban environments. Background Technology

[0002] With the advancement of urbanization, the urban population continues to increase, and urban traffic congestion is gradually intensifying. Against this backdrop, drones, with their multi-functional operation capabilities, high efficiency, and low operating costs, can play a significant role in the construction of smart city systems. For example, well-known drone applications include goods transportation, intelligent monitoring, traffic surveys, air buses, and auxiliary communications.

[0003] Unmanned aerial vehicle (UAV) operations typically involve data communication with ground control. Traditional urban UAVs mostly use WiFi (Wireless Fidelity) for data transmission, but this is limited by communication distance, resulting in low data transmission rates and severe interference. In recent years, a method based on urban cellular network architecture to construct a UAV paradigm has been applied to address this problem and has become a research hotspot. This method, based on a high-density distribution of ground cellular base stations, provides economical communication links for UAV operation while also offering ultra-long coverage control range. Furthermore, cellular base station signals can improve positioning accuracy and effectively avoid communication interruptions caused by unstable GPS (Global Positioning System) signals, adverse weather, or sudden obstacles. In the UAV field, existing research has achieved good results in solving the shortest path problem for cellular-access UAVs under communication constraints. However, designing path planning algorithms in complex obstacle environments is still in its early stages, especially in complex low-altitude urban environments. How to efficiently solve for the optimal flight path of cellular-access UAVs and address the mixed constraints of communication reliability and environmental complexity remains a key and challenging point for the practical application of urban cellular-access UAVs.

[0004] Therefore, existing technologies lack path planning solutions for cellular access drones that consider communication conditions and obstacle avoidance in complex urban environments. Summary of the Invention

[0005] The purpose of this invention is to provide a path planning method and related apparatus for cellular-accessible unmanned aerial vehicles (UAVs) in urban environments, so as to improve the problems existing in the prior art.

[0006] The embodiments of the present invention can be implemented as follows:

[0007] In a first aspect, the present invention provides a path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, comprising:

[0008] A three-dimensional mesh model of the urban space where the flight area is located is established, and the three-dimensional mesh model includes several meshes;

[0009] A communication interruption probability model and an environmental complexity model are constructed for the UAV within the flight area. The communication interruption probability model is used to measure the communication reliability between the UAV and each base station within the flight area. The environmental complexity model is used to evaluate the safety of the UAV within the flight area.

[0010] Based on the communication interruption probability model and the environmental complexity model, a flight risk model is established;

[0011] Based on the flight risk model, a weighted directed graph is generated within the flight area. The weighted directed graph is used to characterize the interval distance and flight cost information between adjacent base stations within the flight area.

[0012] Based on the weighted directed graph, a sequence of base stations the UAV passes through in the flight area is generated; the sequence of base stations passing through the UAV includes multiple base stations from the starting point to the destination.

[0013] The DQN algorithm is used to plan the flight path of the UAV in real time within the signal coverage area of ​​each of the base stations it passes through.

[0014] Secondly, the present invention also provides a path planning device for a cellular-accessible unmanned aerial vehicle in an urban environment, comprising:

[0015] The segmentation module is used to create a three-dimensional mesh model of the urban space where the flight area is located. The three-dimensional mesh model includes several meshes.

[0016] A module is established for: constructing a communication interruption probability model and an environmental complexity model for the UAV within the flight area; the communication interruption probability model is used to measure the communication reliability between the UAV and each base station within the flight area; the environmental complexity model is used to evaluate the safety of the UAV within the flight area; a weighted sum of the communication interruption probability model and the environmental complexity model is performed to obtain a flight risk model; based on the flight risk model, a weighted directed graph is generated within the flight area, the weighted directed graph being used to characterize the distance between adjacent base stations and flight cost information within the flight area;

[0017] The processing module is configured to: generate a sequence of base stations the UAV passes through in the flight area based on the weighted directed graph; the sequence of base stations includes multiple base stations the UAV passes through from the starting point to the ending point; and use the DQN algorithm to plan the flight path of the UAV within the signal coverage area of ​​each of the base stations in real time.

[0018] Thirdly, the present invention also provides an electronic device, comprising: a memory and a processor, wherein the memory stores a software program, and when the electronic device is running, the processor executes the software program to implement the path planning method for a cellular-accessible unmanned aerial vehicle in an urban environment as described in the first aspect.

[0019] Compared with existing technologies, this invention provides a path planning method and related apparatus for cellular-accessible unmanned aerial vehicles (UAVs) in urban environments. By constructing a flight risk model, the optimal path planning problem is modeled as a minimum flight distance model. This minimum flight distance model is then decomposed into a global cost function for global planning and a local cost function for local planning. Therefore, before the UAV executes its flight mission, global planning based on graph theory can be used to solve the global cost function to obtain the sequence of base stations the UAV will pass through. Then, when the UAV begins its flight mission, deep reinforcement learning can be used to solve the local cost function, thereby planning the UAV's flight path within the signal coverage area of ​​the currently accessed base stations in real time. This enables the determination of the optimal flight path for UAVs in complex urban environments, guiding UAVs to complete flight missions within urban areas more safely and efficiently. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating a path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, as provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the position of a drone in a three-dimensional mesh model, provided as an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of an urban cellular network formed by the distribution of various base stations in a flight area, provided as an embodiment of the present invention.

[0024] Figure 4This is a schematic diagram of a weighted directed graph provided in an embodiment of the present invention.

[0025] Figure 5 This is a schematic diagram of a decision network architecture provided in an embodiment of the present invention.

[0026] Figure 6 This is the second flowchart illustrating a path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, as provided in an embodiment of the present invention.

[0027] Figure 7 This is the third flowchart illustrating a path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, as provided in an embodiment of the present invention.

[0028] Figure 8 This is a schematic diagram of the flight area provided in an embodiment of the present invention.

[0029] Figure 9 This is a side view of the planning effect in a three-dimensional mesh model during verification, provided as an embodiment of the present invention.

[0030] Figure 10 This is a top view of the planning effect in a three-dimensional mesh model during verification, provided as an embodiment of the present invention.

[0031] Figure 11 This is a schematic diagram of a path planning device for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, provided as an embodiment of the present invention.

[0032] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0033] 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.

[0034] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0035] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0036] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0037] As mentioned in the background section, existing technologies lack path planning solutions for cellular access drones that consider both communication conditions and obstacle avoidance in complex urban environments.

[0038] Meanwhile, existing technologies lack pathfinding algorithms for UAVs in highly dynamic environments. This is because current path planning algorithms used in engineering require real-time modeling of the environment and lack generalization learning capabilities, which leads to a sharp increase in the difficulty of solving the problem as the number of obstacles increases.

[0039] To address this, the DRL (Deep Reinforcement Learning) algorithm offers a novel approach to path planning in complex dynamic environments. Among these, path planning algorithms based on Deep Q-Networks (DQNs) exhibit strong generalization ability and robustness, and do not require repeated modeling of the dynamic environment, effectively optimizing the algorithm's solution time. However, in practical planning, existing DQN algorithms only satisfy path planning in static environments. When dealing with sudden obstacles, they still employ a phased approach, leading to local optima and poor performance on large scales.

[0040] Based on the discovery of the aforementioned technical problems, the inventors, through creative labor, proposed the following technical solutions to solve or improve these problems. It should be noted that the deficiencies in the solutions of the prior art are all results derived by the inventors after practical experience and careful research. Therefore, the discovery process of the aforementioned problems and the solutions proposed in the embodiments of this application below should be considered contributions made by the inventors to this application during the inventive process, and should not be construed as technical content known to those skilled in the art.

[0041] In view of this, embodiments of the present invention provide a path planning method for cellular-accessible unmanned aerial vehicles (UAVs) in urban environments. This method constructs a flight risk model, models the optimal path planning problem as a minimum flight distance model, and then decomposes the minimum flight distance model into a global cost function for global planning and a local cost function for local planning. Thus, before the UAV executes its flight mission, global planning based on graph theory can be used to solve the global cost function to obtain the sequence of base stations the UAV will pass through. Then, when the UAV begins its flight mission, deep reinforcement learning can be used to solve the local cost function to plan the UAV's flight path within the signal coverage area of ​​the currently accessed base stations in real time. In this way, the optimal flight path for UAVs can be found in complex urban environments, guiding UAVs to complete flight missions within urban areas more safely and efficiently.

[0042] The path planning for cellular-accessible drones in urban environments provided in this invention can be applied to electronic devices, such as various urban drones, including logistics drones, delivery drones, aerial photography drones, and surveying drones. The following detailed description, through embodiments and in conjunction with the accompanying drawings, provides further elaboration.

[0043] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, as provided in an embodiment of the present invention. The method includes the following steps: S110 to S160.

[0044] S110. Establish a three-dimensional mesh model of the urban space where the flight area is located.

[0045] In this embodiment, a side with length Δ d Using a cube as the dividing unit, the urban space where the flight area is located can be divided into several cubes according to the dividing unit, resulting in a three-dimensional mesh model. In the three-dimensional mesh model, the vertex of a cube is a mesh, that is, the three-dimensional mesh model can include several meshes.

[0046] S120. Construct a communication interruption probability model and an environmental complexity model for UAVs within the flight area.

[0047] In this embodiment, there are multiple base stations that the drone can access within the flight area. These base stations can be terrestrial cellular base stations. The communication interruption probability model can be used to measure the communication reliability between the drone and each base station within the flight area, while the environmental complexity model can be used to assess the safety of the drone within the flight area.

[0048] S130. Based on the communication interruption probability model and the environmental complexity model, a flight risk model is established.

[0049] In this embodiment, the flight risk model can be used to calculate the flight risk at any grid point in the three-dimensional mesh model. This flight risk takes into account both the impact of the communication environment and the impact of obstacles.

[0050] S140. Based on the flight risk model, generate a weighted directed graph within the flight area.

[0051] In this embodiment, the weighted directed graph can be used to characterize the spacing distance and flight cost information between adjacent base stations within the flight area.

[0052] S150. Based on the weighted directed graph, generate the sequence of base stations that the UAV passes through in the flight area.

[0053] In this embodiment, a global planning process can be performed based on a weighted directed graph to generate a sequence of base stations along the flight path of the UAV in the flight area. This sequence includes multiple base stations along the UAV's path from the starting point to the destination.

[0054] The S160 uses the DQN algorithm to plan the flight path of the UAV in real time within the signal coverage area of ​​each base station it passes through.

[0055] In this embodiment, during the flight of the UAV, the DQN algorithm can be used to perform local planning to plan the flight path of the UAV within the signal coverage area of ​​each base station it passes through in real time.

[0056] The path planning method for cellular-accessible unmanned aerial vehicles (UAVs) in urban environments provided by this invention can establish a flight risk model using a communication interruption probability model and an environmental complexity model within the flight area. Based on this flight risk model, a weighted directed graph within the flight area is generated. Then, based on the weighted directed graph, global planning is first performed to generate the sequence of base stations the UAV will pass through in the flight area. During the UAV's flight, the Directed QN algorithm is used for local planning to plan the UAV's flight path within the signal coverage area of ​​each passing base station in real time. This method, which considers communication conditions and obstacle avoidance in complex environments and performs path planning hierarchically, effectively solves the optimal flight path for UAVs in complex urban environments, guiding UAVs to complete flight missions within urban areas more safely and efficiently.

[0057] In this invention, the flight area of ​​the UAV is a three-dimensional dense urban area, with various static obstacles (such as static buildings) randomly distributed within the area, as well as multiple base stations. Occasionally, unexpected obstacles (such as other UAVs) may also appear. It is assumed that the three-dimensional mesh model comprises L1×L2×H meshes, where L1 and L2 represent two horizontal lengths of the flight area, and L1 and L2 may be equal or unequal. H represents the vertical height of the flight area.

[0058] When the current position of the drone in the 3D mesh model is q(n) = (x n y n h n ), where n represents the grid number where the UAV is located, x n ∈[0, L1], y n ∈[0, L2] represent the x and y coordinates of the nth grid, respectively, h n ∈[h min h max [] indicates the altitude where no one is located, h min h max These are the lower and upper limits of the drone's flight altitude, respectively.

[0059] Please see Figure 2 The three-dimensional motion space of a drone is determined by the adjacent grids of its location. Figure 2 As can be seen, there are 26 adjacent grids around the location of the drone, meaning the drone can have a maximum of 26 directions of movement. In the 3D mesh model, the shortest distance between adjacent grids is Δ. d The longest distance is

[0060] In this invention, let the positions of the starting point s and the ending point f of the UAV in the three-dimensional mesh model be respectively: q(s) = (x s y s h s ), q(f)=(x f y f h f The drone flies at a constant speed during its flight, making... This represents all the grids (including the start and end points) that the drone needs to pass through on its flight path from the start point to the end point, i.e., all waypoints, where N represents the number of grids that need to be passed.

[0061] The process of establishing a flight risk model will be described in detail below.

[0062] In the optional implementation, the actual communication interruption probability between the drone and the base station it is connected to can be expressed as:

[0063] P out (q(n), I(n))=Pr{γ(q(n), I(n))<γ th} (1)

[0064] Where I(n) represents the position of the base station connected to the UAV at q(n) in the 3D mesh model, and P out(q(n), I(n)) represents the actual communication interruption probability between the UAV and its connected base station; γ(q(n), I(n)) represents the signal-to-noise ratio of the communication link, γ th denoted by , where represents the minimum signal-to-noise ratio threshold for channel communication, and Pr{·} represents the probability of an event occurring.

[0065] However, it is difficult to accurately measure the actual communication interruption probability at any flight position during path planning. Therefore, it is possible to measure the signal-to-noise ratio of the communication link multiple times in a short period of time (for example, by querying the reference signal received power (RSRP) and reference signal received quality (RSRQ)). Then, based on the multiple measured communication link signal-to-noise ratios, the empirical communication interruption probability can be calculated, and thus the actual communication interruption probability can be predicted.

[0066] Therefore, in this invention, the expression for the communication interruption probability model can be:

[0067]

[0068]

[0069] In expression (2), γ(q(n), I(n)) represents the measured signal-to-noise ratio of the communication link; c(q(n), I(n)) is the interruption judgment function, and J is the number of measurements; Let be the probability of communication interruption between the UAV at its current location q(n) and the connected base station I(n). This communication interruption probability is essentially an empirical communication interruption probability. Because, based on the law of large numbers, when the number of measurements J is sufficiently large, the empirical interruption probability can be equivalent to the actual communication interruption probability, as shown in the following equation:

[0070]

[0071] In the optional implementation, the higher the number of obstacles and the probability of sudden obstacles occurring in the flight area, the higher the environmental complexity. For a single static obstacle, the higher its height, the higher the environmental complexity. Therefore, a spatial region with q(n) as the center and r as the radius can be used as the evaluation area. Then, the environmental complexity of the region can be quantified by comprehensively considering the size, height, and number of static obstacles in the evaluation area as well as the probability of sudden obstacles occurring.

[0072] Therefore, in this invention, static buildings can be regarded as static obstacles, and the expression of the environmental complexity model can be:

[0073]

[0074] In expression (5), r represents the radius of the area being evaluated in the flight area, B is the number of static obstacles in the area being evaluated, and S... ih represents the area occupied by the i-th building within the assessed area. i Let E be the height of the i-th building within the assessed area. s E represents the probability of a sudden obstacle occurring at point q(n); q(n) This represents the environmental complexity of the drone at its current position q(n).

[0075] In an optional implementation, the communication interruption probability model and the environmental complexity model can be weighted and summed to obtain a flight risk model, the expression of which can be:

[0076]

[0077] In expression (6), δ1 and δ2 are both weighting coefficients; α q(n) This represents the flight risk level at the current position q(n).

[0078] The above describes the process of establishing the flight risk model. The following section will introduce the modeling process involved in the optimal path planning problem.

[0079] It is understandable that the optimization goal of the optimal path planning problem is to control the flight trajectory of the drone from the starting point to the destination to minimize the flight distance while ensuring the safe flight of the drone.

[0080] Therefore, based on the three-dimensional mesh model and the aforementioned flight risk model, the flight risk at each mesh location can be calculated, and the optimal path planning problem can be modeled as a minimization of flight distance model, the expression of which is:

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] In expression (7a), Represents all path points along the drone's flight path from start to finish. η represents the position of each path point in the three-dimensional mesh model; η(i) represents the set of all adjacent meshes around the i-th path point on the flight trajectory; st represents: expressions (7b) to (7f) are all constraints that minimize the flight distance model (7a);

[0088] In expressions (7b) to (7f), χ q(i)，q(j) , χ q(j)，q(i) Both are Boolean variables, representing whether the drone moves from the i-th path point on the flight path to the j-th adjacent grid around the i-th path point and whether the drone moves from the j-th adjacent grid around the i-th path point on the flight path to the i-th path point, respectively. The i-th path point on the flight trajectory belongs to the set. But excluding the starting point s and the ending point f; This represents the distance between any two adjacent meshes in a 3D mesh model that does not exceed [a certain value].

[0089] Among them, expressions (7b) and (7c) are used to constrain the uninterrupted flight trajectory of the UAV and the existence of only a single path between adjacent path points; expression (7d) is used to constrain the flight risk degree of any grid to be no greater than 1, and the first path point and the last path point of the flight trajectory are the start point and the end point of the UAV, respectively; expression (7e) is used to constrain the single movement distance of the UAV to not exceed Expression (7f) is used to constrain the location of the UAV to not exceed the range of the three-dimensional mesh model.

[0090] Since the UAV used in this invention is a cellular access UAV, this invention can decompose the complex minimum flight distance model into a global cost function and a local cost function based on dynamic programming theory. The global cost function can be based on globally connected perception planning to first determine all the base stations the UAV needs to connect to from the starting point to the destination; during the UAV's flight, the local cost function can be based on collision-free local planning to calculate the UAV's flight path within the signal coverage area of ​​its currently connected base station in real time.

[0091] Therefore, the path planning of this invention belongs to hierarchical planning. The first level is the global planning level that can be performed in advance, and the second level is the local planning level that can be advanced in real time.

[0092] First, at the global planning level, this invention addresses the characteristics of cellular access UAVs, using base stations as the core for macro-level planning. To further improve solution efficiency, the signal coverage area of ​​each base station is used as the planning unit, allowing for the assessment of the average flight risk within the signal coverage area of ​​each base station. Therefore, the expression for the global cost function can be:

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] In expression (8a), N represents the sequence of base stations traversed. G This represents the number of base stations along the route. Represents the location of each of the aforementioned transit base stations in the three-dimensional mesh model; η G (i) represents the set of all adjacent base stations of the i-th passing base station; M represents the set of all grids within the signal coverage area of ​​the i-th passing base station; i This represents the number of grid cells within the signal coverage area of ​​the i-th passing base station;

[0099] Expressions (8b) to (8e) are the constraints of expression (8a), where χ i，j , χ j，i Both are Boolean variables, representing whether the UAV switches from the i-th passing base station to the j-th adjacent base station of the i-th passing base station and whether the UAV switches from the j-th adjacent base station of the i-th passing base station to the i-th passing base station; q(j) and q(i) represent the positions of the j-th adjacent base station and the i-th passing base station in the three-dimensional mesh model, respectively; ||q(j)-q(i)|| represents the distance between the j-th adjacent base station and the i-th passing base station.

[0100] Secondly, at the local planning level, based on the sequence of passing base stations obtained by solving the global cost function, the flight path within the signal coverage area of ​​each passing base station can be calculated in real time during the UAV's flight. Therefore, the expression for the local cost function can be:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107] In expression (9a), N represents the flight path of the UAV within the signal coverage area of ​​the i-th base station it passes through. UThis indicates the number of waypoints on the flight path. η represents the position of each path point within the signal coverage area of ​​the i-th passing base station in the three-dimensional mesh model; u (j) represents the set of adjacent grids consisting of all adjacent grids of the j-th path point on the flight path;

[0108] Expressions (9b) to (8f) are the constraints of expression (9a), where χ q(j)，q(k) , χ q(k)，q(j) Both are Boolean variables, representing whether the UAV moves from the j-th path point to the k-th adjacent grid of the j-th path point and whether the UAV moves from the k-th adjacent grid of the j-th path point to the j-th path point; q(k) and q(j) represent the positions of the k-th adjacent grid of the j-th path point and the j-th path point in the 3D mesh model, respectively; ||q(k)-q(j)|| represents the distance between the k-th adjacent grid of the i-th path point and the j-th path point. This represents the set to which the j-th path point belongs. However, this excludes the starting point s and the ending point f; q(s) and q(f) represent the positions of the starting point s and the ending point f in the three-dimensional mesh model, respectively.

[0109] In the optional implementation, at the global planning level, assuming the UAV's flight altitude is constant, the overall operating environment of the flight area can be transformed into a weighted directed graph. Then, based on the global cost function and the weighted directed graph, the Dijkstra algorithm is used to generate a macroscopic sequence of passing base stations.

[0110] Therefore, in an optional implementation, the sub-steps of step S140 above may include S141 to S146.

[0111] S141. Obtain the position of each base station in the three-dimensional mesh model within the flight area.

[0112] S142. Calculate the distance information between each pair of adjacent base stations within the flight area.

[0113] In this embodiment, the distance information between each pair of adjacent base stations can be calculated based on the position of each base station in the three-dimensional mesh model. This distance information can be a relative distance in the three-dimensional mesh model, rather than the actual physical distance.

[0114] In the optional example, it is assumed that there are 7 base stations (base stations q(1) to q(7)) in the flight area. Please refer to [link to relevant documentation]. Figure 3 Base station q(7) is located at the center, and base stations q(1) to q(6) are distributed around base station q(7). Therefore, there are a total of 12 pairs of adjacent base stations, and 12 sets of distance information need to be calculated (i.e., Figure 3 D in i，jIn this example, both i and j belong to [1, 7]. It should be noted that this example is only an illustration, and the number and distribution of base stations in the flight area are subject to actual application conditions. This invention does not impose any restrictions on this.

[0115] S143. For each base station, calculate the flight risk level at each grid point within the signal coverage area of ​​the base station based on the flight risk level model.

[0116] S144. Based on the flight risk level at each grid within the signal coverage area of ​​the base station, calculate the average flight risk level of the drone within the signal coverage area of ​​the base station.

[0117] In this embodiment, steps S143 and S144 are executed for each base station in the flight area to obtain the average flight risk of the UAV within the signal coverage area of ​​each base station.

[0118] S145. Based on the distance information between each pair of adjacent base stations and the average flight risk of the UAV within the signal coverage area of ​​each base station, determine two risk weight coefficients between each pair of adjacent base stations.

[0119] Optionally, the two base stations in a pair of adjacent base stations are a first base station and a second base station. The risk weight coefficient from the first base station to the second base station can be the product of the distance information between the two base stations and the average flight risk level corresponding to the second base station. Correspondingly, the risk weight coefficient from the second base station to the first base station can be the product of the distance information between the two base stations and the average flight risk level corresponding to the first base station.

[0120] S146. Based on the position of each base station in the three-dimensional mesh model and the two risk weight coefficients between each pair of adjacent base stations, construct a weighted directed graph within the flight area.

[0121] In this embodiment, the weighted directed graph may include multiple nodes, with each node representing a base station. There are two directed edges with opposite directions between a pair of adjacent nodes (adjacent base stations), and the weight of the directed edge is the risk weight coefficient.

[0122] That is, a pair of adjacent nodes includes a first node and a second node (corresponding to the first base station and the second base station, respectively). The weight of the directed edge starting from the first node and ending at the second node is the risk weight coefficient from the first base station to the second base station; the weight of the directed edge starting from the second node and ending at the first node is the risk weight coefficient from the second base station to the first base station.

[0123] Optionally, a weighted directed graph can be represented as

[0124] N′ represents the set of all base stations within the flight area. GLet E be the number of base stations in the flight area; let E be the set of directed edges for each pair of adjacent base stations in the flight area; and let W be the set of two risk weight coefficients for each pair of adjacent base stations in the flight area.

[0125] The risk weight coefficients from the i-th base station to the j-th base station can be:

[0126]

[0127] In the formula, M represents the set of all grids within the signal coverage area of ​​the j-th base station. j This represents the number of grid cells within the signal coverage area of ​​the j-th base station in the flight area; D represents the average flight risk within the signal coverage area of ​​the j-th base station. i，j This represents the distance information between the i-th base station and the j-th base station.

[0128] In optional examples, Figure 3 Based on this, please refer to Figure 4 A weighted directed graph can be like this: Figure 4 As shown, the directed edge from the i-th base station to the j-th base station is q(i)→q(j), and the risk weight coefficient on the directed edge from the i-th base station to the j-th base station is ω. i，j .

[0129] It should be noted that the values ​​in the weighted directed graph are for illustrative purposes only, and should be used in accordance with actual applications. This invention does not impose any limitations on these values.

[0130] Furthermore, the sub-steps of step S150 above may include S151 to S152.

[0131] S151. Obtain the starting point and ending point of the drone;

[0132] S152. Based on the global cost function, use Dijkstra's algorithm to search for the sequence of base stations along the route from the starting point to the ending point in a weighted directed graph.

[0133] It is understandable that, considering the meaning of the two risk weight coefficients between a pair of adjacent base stations mentioned above, the optimization objective in the global cost function (8a) is to find the optimal sequence of passing base stations while ensuring that the sum of the risk weight coefficients of each passing base station is minimized.

[0134] Since this invention transforms the overall operating environment of the flight area into a weighted directed graph, the Dijkstra algorithm can be directly used to search for the sequence of base stations passing through from the starting point to the ending point in the weighted directed graph, thus realizing the solution of the global cost function and obtaining the macro-optimal path (sequence of base stations passing through) for the whole data planning.

[0135] In the optional implementation, at the local planning level, based on the macro-optimal path, the flight path within the signal coverage area of ​​each path base station is planned in real time.

[0136] This invention proposes a "offline + online" collaborative approach based on the DQN algorithm for path planning within the signal coverage area of ​​passing base stations, thereby improving upon the drawbacks of conventional DQN algorithms that often get trapped in local optima. The decision network architecture of the DQN algorithm in this invention is as follows: Figure 5 As shown, it includes two DQN networks: a long-term decision network for calculating static obstacle data and autonomously learning flight strategies, and a short-term strategy network for calculating sudden obstacle data and guiding the UAV to avoid collisions.

[0137] Within a single time step, the UAV can detect sudden obstacles using sensors. Once an obstacle is detected, the short-term policy network is activated, outputting a short-term Q-value based on the current UAV state data (including sudden obstacle information). The long-term policy network outputs a short-term long-term Q-value based on the current UAV state data (including static obstacle information).

[0138] Therefore, the drone can begin its flight mission from the starting point and then fly according to the sequence of transit base stations. For each transit base station it connects to, it plans its flight path in real time until it reaches its destination. The drone's flight path corresponding to a transit base station includes multiple waypoints, each waypoint representing a grid within the signal coverage area of ​​that transit base station. Please see [link to relevant documentation]. Figure 6 The above step S160 may include the following sub-steps S161 to S165.

[0139] S161. Use the base stations along the route that the drone connects to as the current base station.

[0140] S162. Based on the local cost function, the DQN algorithm is used to plan the flight path of the UAV within the signal coverage area of ​​the current base station in real time.

[0141] It can be understood that in the global cost function (9a) mentioned above, the optimization objective is to find the optimal flight path within the signal coverage area of ​​the i-th passing base station while minimizing the sum of the products of flight risk and distance for each path point within the signal coverage area of ​​the i-th passing base station. Therefore, by using the decision network architecture of the DQN algorithm of this invention to plan the flight path of the UAV within the signal coverage area of ​​the i-th passing base station in real time, the global cost function (9a) is solved, and the locally planned flight path is obtained.

[0142] Optional, please see Figure 7The sub-steps of step S162 may include S1621 to S1629.

[0143] S1621. Use the grid position where the drone is located as the current path point.

[0144] It's understandable that when a drone begins its flight mission from its starting point, the first transit base station it connects to at that point becomes its current base station, and the starting point becomes its current waypoint. The drone needs to plan its flight path within the signal coverage area of ​​this current base station (the first transit base station) in real time, starting from the starting point. Alternatively, if the drone has just switched from the first transit base station to the second transit base station, then the second transit base station becomes its new current base station, and the drone's current grid position becomes its current waypoint. The drone needs to plan its flight path within the signal coverage area of ​​this current base station (the second transit base station) in real time, starting from this current waypoint.

[0145] S1622. Determine if there are any sudden obstacles at the current path point.

[0146] It is understandable that drones can use their onboard sensors (such as lidar) to detect the presence of sudden obstacles.

[0147] In this embodiment, if there is no sudden obstacle at the current path point, then the steps S1623 and S1624 are executed and then the step S1627 is executed; if there is a sudden obstacle at the current path point, then the steps S1623, S1625 and S1626 are executed and then the step S1627 is executed.

[0148] S1623. Input the current path point, destination, and static obstacle information within the flight area of ​​the UAV into the long-term decision network to output the long-term Q value under the mapping of the local cost function.

[0149] S1624. Use the adjacent grids of the current path point corresponding to the long-term Q value as the next path point.

[0150] S1625. Input the current path point, destination, and sudden obstacle information of the UAV into the short-term decision network to output the short-term Q value under the mapping of the local cost function.

[0151] It can be understood that, as an intelligent agent, the drone's state space, action space, and reward function are as follows:

[0152] With s t This represents the state data of the t-th path point of the UAV, which includes the current path point q(t) of the UAV, the destination q(f), and static obstacle information within the flight area (i.e., the current relative static building distance set). And the distance between the aircraft and the sudden obstacle (i.e., the current relative threat location distance l) t ).

[0153] by Representing the drone's action space, combined with Figure 2 It consists of 26 directions, and its set of actions is a. t When action a t In state s t If the following action is taken, a reward r will be obtained. t+1 The reward function in a long-term decision network can be set as expression (10):

[0154]

[0155] In the formula, d pre =||q(n)-q(f)|| and d cur =||q(n+1)-q(f)|| represents the relative distance between the drone and the destination before and the destination now, and μ1, μ2, and μ3 are all influencing factors.

[0156] In short-term decision networks, the reward function is set as expression (11):

[0157]

[0158] S1626. The neighboring grid of the current path point corresponding to the larger of the long-term Q value and the short-term Q value is taken as the next path point.

[0159] S1627. Determine whether the drone switches to connect to the next base station in the sequence of base stations it passes through.

[0160] In this embodiment, if the drone has switched to the next base station in the sequence of base stations it has passed through, then the flight path of the drone within the signal coverage area of ​​the current base station is planned and obtained. If the drone has not switched to the next base station in the sequence of base stations it has passed through, then after executing step S1628, the process returns to executing step S1622 until the drone has switched to the next base station in the sequence of base stations it has passed through, thus obtaining the flight path of the drone within the signal coverage area of ​​the current base station.

[0161] S1628. Set the next path point as the current path point.

[0162] S1629. Obtain the flight path of the drone within the signal coverage area of ​​the current base station.

[0163] In this embodiment, if the drone does not switch to the next base station in the sequence of base stations it has passed, the drone can fly from its own position to the next waypoint and use the next waypoint as the new current waypoint. Then it returns to step S1622 to calculate the new next waypoint. This process is repeated multiple times until the drone switches to the next base station in the sequence of base stations it has passed. Then all the waypoints calculated and flown through when connecting to the current base station are the optimal collision-free flight paths of the drone within the signal coverage area of ​​the current base station.

[0164] S163. Determine whether the current base station is the last base station in the sequence of base stations passed through.

[0165] In this embodiment, if the current base station is the last base station in the sequence of passing base stations, the overall path planning is completed, and the flight path of the UAV within the signal coverage area of ​​each passing base station is obtained; if the current base station is not the last base station in the sequence of passing base stations, the following step S164 is executed and then the execution of step S162 is returned until the current base station is the last base station in the sequence of passing base stations, and the flight path of the UAV within the signal coverage area of ​​each passing base station is obtained.

[0166] S164. Take the next base station in the sequence of base stations passed through as the current base station.

[0167] S165. Obtain the flight path of the drone within the signal coverage area of ​​each base station it passes through.

[0168] In one alternative example, in Figure 3 Based on this, please refer to Figure 8 Assuming the environment of the flight area is as follows: Figure 8 As shown, the sequence of base stations passed through includes base stations A2, A1, A3, and A4. Therefore, the flight trajectory from the starting point s to the ending point f includes the following four parts:

[0169] 1. When the drone connects to base station A2, the real-time planned flight path from starting point s within the signal coverage area of ​​A2 is as follows: Figure 8 The dashed curve segment between the midpoint s and S1;

[0170] 2. The UAV switches from base station A2 to A1 at point S1. Within the signal coverage area of ​​A1, the real-time planned flight path starts from S1 as follows: Figure 8 The dashed curve segment between S1 and S2;

[0171] 3. The UAV switches from base station A1 to A3 at point S2. Within the signal coverage area of ​​A3, the real-time planned flight path starts from S2 as follows: Figure 8 The dashed curve segment between S2 and S3;

[0172] IV. The drone switches from the transit base station A3 to A4 at point S3. Within the signal coverage area of ​​A4, the real-time planned flight path starts from S3 as follows: Figure 8 The dashed curve segment between S3 and the endpoint f.

[0173] It should be noted that this example is merely an illustration, and the environment of the flight area and the final flight trajectory of the drone shall be subject to the actual application situation. This embodiment of the invention does not limit these aspects.

[0174] The following describes the verification process of the path planning method for cellular-accessible drones in urban environments provided by this invention.

[0175] Assuming the 3D mesh model of the city space where the flight zone is located is 500×500×25, please refer to [link / reference]. Figure 9 The flight area contains a total of 123 static obstacles (gray bars), 17 sudden obstacles (black bars), and 25 base stations. The static obstacles are randomly generated, with lengths and widths ranging from 8 to 18 and heights from 5 to 22. The sudden obstacles have lengths and widths set to 10 and heights to 25. The base stations are evenly distributed, with a signal coverage radius of 70. It is assumed that the UAV's onboard sensor sensing radius is 5.

[0176] like Figure 9 , Figure 10 As shown, trajectory 1 is the flight trajectory solved using the method of this invention, and trajectory 2 is the flight trajectory solved using the conventional DQN algorithm. From Figure 9 The results from the side view show that the flight paths obtained by both methods gradually ascend, passing directly through low-rise buildings and detouring around high-rise buildings, effectively avoiding risks when encountering sudden restricted areas. From... Figure 10 The top-down view results show that the method of this invention achieves global optimization in path selection. Specifically, at the macro-planning level, suitable flight areas (i.e., the base stations passed through) are selected, and at the local planning level, the algorithm achieves the shortest flight path within the base station signal range. In contrast, the conventional DQN algorithm struggles to guarantee global optimization, and the flight environment in the areas it passes through is relatively complex, making it difficult to meet practical requirements.

[0177] It should be noted that the execution order of each step in the above method embodiments is not limited to that shown in the attached figures, and the execution order of each step shall be subject to the actual application situation.

[0178] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0179] This invention proposes a novel multi-layered path planning method for cellular-access UAVs that combines air-to-ground communication service assurance with collision avoidance in complex dynamic environments. This method consists of two layers: a global planning layer based on connectivity awareness and a collision-free local planning layer. Through layered planning, it effectively addresses the mixed constraints of communication reliability and environmental complexity, efficiently solving for the optimal flight path of UAVs and providing an effective solution for path planning of cellular-access UAVs in complex low-altitude urban environments.

[0180] Second, the flight risk index proposed in this invention integrates the probability of communication interruption between the UAV and the ground base station and the complexity of the flight environment. It integrates these two factors into the complex low-altitude urban scenario using grid theory, and thereby models the complex optimal path planning problem as a minimum flight distance model. By effectively constructing the optimization problem model, a more comprehensive and accurate description of the path planning problem for cellular-access UAVs in complex urban environments is achieved, laying the foundation for a more efficient solution strategy.

[0181] Third, this invention proposes a hierarchical method for solving optimal flight paths based on weighted directed graphs and the DRL algorithm. This method, based on dynamic programming theory, solves complex nonlinear programming problems hierarchically. Specifically, the proposed global planning layer based on weighted directed graphs ensures macroscopic optimization of path planning, and uses the coverage area of ​​ground base station signals as the smallest planning unit to improve solution efficiency. The proposed "offline + online" collaborative path planning method based on the DQN algorithm can effectively solve for locally optimal collision-free flight paths within the coverage area of ​​each base station signal, overcoming the shortcoming of existing DQN algorithms that get trapped in local optima. Therefore, this invention not only fills the gap in existing cellular access UAV path planning algorithms but also paves the way for the development of more complex and efficient UAV path planning systems.

[0182] In order to perform the corresponding steps in the above method embodiments and various possible implementations, the following is an implementation method of a path planning device for a cellular access UAV in an urban environment.

[0183] Please see Figure 11 , Figure 11 A schematic diagram of the path planning device for a cellular-access UAV in an urban environment provided by an embodiment of the present invention is shown. The path planning device 200 for a cellular-access UAV in an urban environment includes: a segmentation module 210, a creation module 220, and a processing module 230.

[0184] The segmentation module 210 is used to create a three-dimensional mesh model of the urban space where the flight area is located. The three-dimensional mesh model includes several meshes.

[0185] Module 220 is established for: constructing a communication interruption probability model and an environmental complexity model for the UAV within its flight area; the communication interruption probability model is used to measure the communication reliability between the UAV and each base station within the flight area; the environmental complexity model is used to assess the safety of the UAV within the flight area; the communication interruption probability model and the environmental complexity model are weighted and summed to obtain a flight risk model; based on the flight risk model, a weighted directed graph is generated within the flight area, which is used to represent the distance between adjacent base stations and flight cost information within the flight area;

[0186] The processing module 230 is used to: generate a sequence of base stations that the UAV passes through in the flight area based on a weighted directed graph; the sequence of base stations includes multiple base stations that the UAV passes through from the starting point to the ending point; and use the DQN algorithm to plan the flight path of the UAV in real time within the signal coverage area of ​​each base station.

[0187] Those skilled in the art will clearly understand that the establishment module 220 can be used to implement the above steps S120 to S140 and their sub-steps, and the processing module 230 can be used to implement the above steps S15 to S160 and their sub-steps. For the sake of convenience and brevity, the specific working process of the path planning device 200 for a cellular-accessible UAV in an urban environment described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0188] Please see Figure 12 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 300 includes a processor 310, a memory 320, and a bus 330, with the processor 310 connected to the memory 320 via the bus 330.

[0189] The memory 320 can be used to store software programs, such as the software program corresponding to the path planning device 200 for a cellular-access UAV in an urban environment provided in this embodiment of the invention. The processor 310 executes various functional applications and data processing by running the software program stored in the memory 320 to realize the path planning method for a cellular-access UAV in an urban environment provided in this embodiment of the invention.

[0190] The memory 320 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0191] The processor 310 can be an integrated circuit chip with signal processing capabilities. The processor 310 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0192] Understandable. Figure 12 The structure shown is for illustrative purposes only; the electronic device 300 may also include components that are more advanced than those shown. Figure 12 The more or fewer components shown, or having the same Figure 12 The different configurations shown. Figure 12 The components shown can be implemented using hardware, software, or a combination thereof.

[0193] In summary, this invention provides a path planning method and related apparatus for cellular-access UAVs in urban environments. First, a communication interruption probability model is used to measure the communication reliability between the UAV and various ground base stations. An environmental complexity model is constructed based on the proportion of static obstacles and the probability of sudden obstacle occurrence within the flight area to assess the safety of the flight area. Based on this, a flight risk model is constructed using a weighted summation method. Further, based on the established three-dimensional mesh model, the communication interruption probability and environmental complexity of each mesh within the flight area are calculated to obtain the flight risk level, modeling the optimal path planning problem as a minimum flight distance model. Next, based on dynamic programming theory, the global solution problem is decomposed into globally connected perception planning and local collision-free planning for hierarchical solution. In the global planning, considering the characteristics of cellular-access UAVs, the signal coverage area of ​​ground base stations is used as the planning unit. The flight area is transformed into a directed graph. The weights of the directed graph are obtained by comprehensively evaluating the flight risk level of each base station's signal coverage area, and the optimal macroscopic path (passing through a sequence of base stations) is solved based on Dijkstra's algorithm. Next, based on the optimal macroscopic path, a "offline + online" collaborative path planning method based on the DQN algorithm is proposed to solve the local collision-free flight path within the signal coverage area of ​​each path base station in real time. The decision network architecture of the DQN algorithm adopts a dual DQN network structure, consisting of a long-term decision network responsible for calculating static obstacle data and a short-term decision network responsible for calculating sudden obstacle data. The overall decision network architecture selects the best strategy under the current state and generates the optimal collision-free flight path of the UAV through continuous iteration.

[0194] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A path planning method for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, characterized in that, include: A three-dimensional mesh model of the urban space where the flight area is located is established, and the three-dimensional mesh model includes several meshes; A communication interruption probability model and an environmental complexity model are constructed for the UAV within the flight area. The communication interruption probability model is used to measure the communication reliability between the UAV and each base station within the flight area. The environmental complexity model is used to evaluate the safety of the UAV within the flight area. Based on the communication interruption probability model and the environmental complexity model, a flight risk model is established; Based on the flight risk model, a weighted directed graph is generated within the flight area. The weighted directed graph is used to characterize the interval distance and flight cost information between adjacent base stations within the flight area. The origin and destination of the UAV are obtained, and based on the global cost function, Dijkstra's algorithm is used to search for a sequence of transit base stations from the origin to the destination in the weighted directed graph; the sequence of transit base stations includes multiple transit base stations of the UAV from the origin to the destination. The DQN algorithm and local cost function are used to plan the flight path of the UAV in real time within the signal coverage area of ​​each of the base stations it passes through; The expression for the communication interruption probability model is as follows: in, This represents the current position of the UAV in the 3D mesh model. This indicates the location of the base station currently connected to the drone in the three-dimensional mesh model. This represents the minimum signal-to-noise ratio threshold for channel communication. This indicates the measured signal-to-noise ratio of the communication link; This is an interrupt detection function. For the number of measurements; For the drone at its current location Location and base station The probability of communication interruption between them; The expression for the environment complexity model is: in, This represents the radius of the area being evaluated within the stated flight area. This represents the upper limit of the flight altitude of the drone. The number of static obstacles within the evaluated area. For the first region under evaluation The area occupied by each building For the first region under evaluation The height of each building, for The probability of encountering sudden obstacles; This indicates that the drone is at its current location. The complexity of the environment; The expression for the flight risk model is: in, and All are weighted coefficients; This indicates that the drone is at its current location. Flight risk level at the location.

2. The method according to claim 1, characterized in that, The step of generating a weighted directed graph within the flight area based on the flight risk model includes: Obtain the position of each base station within the flight area in the three-dimensional mesh model; Calculate the distance information between each pair of adjacent base stations within the flight area; For each base station, the flight risk level at each grid point within the signal coverage area of ​​the base station is calculated according to the flight risk level model; Based on the flight risk level at each grid point within the signal coverage area of ​​the base station, calculate the average flight risk level of the UAV within the signal coverage area of ​​the base station; Based on the distance information between each pair of adjacent base stations and the average flight risk of the UAV within the signal coverage area of ​​each base station, two risk weighting coefficients are determined between each pair of adjacent base stations. Based on the position of each base station in the three-dimensional mesh model and the two risk weight coefficients between each pair of adjacent base stations, a weighted directed graph is constructed within the flight area.

3. The method according to claim 1, characterized in that, The flight path includes multiple path points within the signal coverage area of ​​the base stations along the route, and each path point represents a grid. The step of using the DQN algorithm and local cost function to plan the flight path of the UAV in real time within the signal coverage area of ​​each of the passing base stations includes: The base stations along the route connected to the drone are designated as the current base stations. Based on the local cost function, the DQN algorithm is used to plan the flight path of the UAV in real time within the signal coverage area of ​​the current base station; Determine whether the current base station is the last base station in the sequence of base stations passed through; If the current base station is the last base station in the sequence of passing base stations, then the flight path of the UAV within the signal coverage area of ​​each of the passing base stations is obtained; If the current base station is not the last base station in the sequence of passed base stations, then the next base station in the sequence of passed base stations will be taken as the current base station. Return to the step of planning the flight path of the UAV within the signal coverage area of ​​the current base station using the DQN algorithm based on the local cost function, until the current base station is the last base station in the sequence of passing base stations, and obtain the flight path of the UAV within the signal coverage area of ​​each of the passing base stations.

4. The method according to claim 3, characterized in that, The decision network architecture of the DQN algorithm includes a long-term decision network and a short-term decision network; The step of planning the flight path of the UAV within the signal coverage area of ​​the current base station in real time based on the local cost function and using the DQN algorithm includes: The grid position where the drone is located is taken as the current path point; Determine whether there is a sudden obstacle at the current path point; If there is no sudden obstacle at the current path point, the current path point, the destination, and the static obstacle information in the flight area of ​​the UAV are input into the long-term decision network to output a long-term Q value under the mapping of the local cost function; and the adjacent grid of the current path point corresponding to the long-term Q value is taken as the next path point. If the sudden obstacle exists at the current path point, the current path point, destination, and static obstacle information within the flight area of ​​the UAV are input into the long-term decision network to output the long-term Q value under the mapping of the local cost function; at the same time, the current path point, destination, and sudden obstacle information of the UAV are input into the short-term decision network to output the short-term Q value under the mapping of the local cost function. The neighboring grid of the current path point corresponding to the larger of the long-term Q value and the short-term Q value is taken as the next path point; Determine whether the drone switches to connect to the next base station in the sequence of passed base stations; If the drone has switched to the next base station in the sequence of base stations it has passed through, then the flight path of the drone within the signal coverage area of ​​the current base station is obtained. If the drone does not switch to the next base station in the sequence of passed base stations, the next path point is taken as the current path point and the process returns to the step of determining whether there is a sudden obstacle at the current path point, until the drone has switched to the next base station in the sequence of passed base stations, and the flight path of the drone within the signal coverage area of ​​the current base station is obtained.

5. The method according to claim 1, characterized in that, The expression for the global cost function is: ； ； in, This indicates the sequence of base stations along the route. This represents the number of base stations along the route. This represents the location of each of the aforementioned transit base stations in the three-dimensional mesh model; Indicates the first The set consisting of all adjacent base stations of a passing base station; Indicates the first A set consisting of all grids within the signal coverage area of ​​a passing base station; Indicates the first The number of grid cells within the signal coverage area of ​​each passing base station; , Both are Boolean variables, representing whether the drone departs from the first... The first transit base station switches to the first The first one passing through the base station Whether the adjacent base stations and the drone are from the first The first one passing through the base station The neighboring base station switched to the first One passing base station; , They represent the first The first one passing through the base station The adjacent base stations, the first The positions of each transit base station in the three-dimensional mesh model; Indicates the first The first one passing through the base station The neighboring base station and the first The distance between each passing base station; The three-dimensional mesh model includes One grid, , The two horizontal lengths representing the flight area Represents the vertical height of the flight area; This represents the lower limit of the flight altitude of the drone.

6. The method according to claim 5, characterized in that, The expression for the local cost function is: ； ； ； in, The drone mentioned above is in the A flight path that passes through the signal coverage area of ​​a base station. This indicates the number of waypoints on the flight path. Representing the The positions of each path point within the signal coverage area of ​​a passing base station in the three-dimensional mesh model; Indicates the first The set of adjacent grids consisting of all adjacent grids of each path point; , Both are Boolean variables, representing whether the drone departs from the first... The path point moves to the first... The path point of the first The number of adjacent grids and whether the drone is from the first... The path point of the first The adjacent grid moves to the first... 1 path point; , Representing the first The path point of the first The adjacent grid, the first The location of each path point in the 3D mesh model Representing the The path point of the first The adjacent grid and the first The distance between each path point; Representing the Each path point belongs to the set. But excluding the starting point and the end point ; , Representing the starting point and the end point Their respective positions within the three-dimensional mesh model.

7. A path planning device for a cellular-accessible unmanned aerial vehicle (UAV) in an urban environment, characterized in that, include: The segmentation module is used to create a three-dimensional mesh model of the urban space where the flight area is located. The three-dimensional mesh model includes several meshes. Create a module for: A communication interruption probability model and an environmental complexity model are constructed for the UAV within the flight area. The communication interruption probability model is used to measure the communication reliability between the UAV and each base station within the flight area. The environmental complexity model is used to evaluate the safety of the UAV within the flight area. The flight risk model is obtained by weighted summation of the communication interruption probability model and the environmental complexity model. Based on the flight risk model, a weighted directed graph is generated within the flight area. The weighted directed graph is used to characterize the interval distance and flight cost information between adjacent base stations within the flight area. Processing module, used for: The origin and destination of the UAV are obtained, and based on the global cost function, Dijkstra's algorithm is used to search for a sequence of transit base stations from the origin to the destination in the weighted directed graph; the sequence of transit base stations includes multiple transit base stations from the origin to the destination of the UAV. The DQN algorithm and local cost function are used to plan the flight path of the UAV in real time within the signal coverage area of ​​each of the base stations it passes through; The expression for the communication interruption probability model is as follows: in, This represents the current position of the UAV in the 3D mesh model. This indicates the location of the base station currently connected to the drone in the three-dimensional mesh model. This represents the minimum signal-to-noise ratio threshold for channel communication. This indicates the measured signal-to-noise ratio of the communication link; This is an interrupt detection function. For the number of measurements; For the drone at its current location Location and base station The probability of communication interruption between them; The expression for the environment complexity model is: in, This represents the radius of the area being evaluated within the stated flight area. This represents the upper limit of the flight altitude of the drone. The number of static obstacles within the evaluated area. For the first region under evaluation The area occupied by each building For the first region under evaluation The height of each building, for The probability of encountering sudden obstacles; This indicates that the drone is at its current location. The complexity of the environment; The expression for the flight risk model is: in, and All are weighted coefficients; This indicates that the drone is at its current location. Flight risk level at the location.

8. An electronic device, characterized in that, include: The electronic device includes a memory and a processor, wherein the memory stores a software program, and the processor executes the software program when the electronic device is running to implement the path planning method for a cellular-accessible unmanned aerial vehicle in an urban environment as described in any one of claims 1-6.

Images