Unmanned aerial vehicle cluster collaborative route planning method based on graph neural network

Abstract

The invention relates to the technical field of unmanned aerial vehicle path planning, in particular to an unmanned aerial vehicle cluster collaborative route planning method based on a graph neural network. According to the invention, the static uncontrollability and the dynamic uncontrollability are determined based on the change conditions of the static uncontrollable obstacle and the dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period relative to the previous analysis period; based on the course sudden change degree of the dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period in the previous analysis period, the dynamic correction uncontrollability is determined, and then the surrounding obstacle degree of the unmanned aerial vehicle in the analysis period is determined; and based on the difference condition of the surrounding obstacle degrees of each unmanned aerial vehicle and the surrounding unmanned aerial vehicles in the analysis period, updating the cooperative relation matrix, determining a correction relation matrix of each analysis period, and performing route planning on the unmanned aerial vehicle cluster. The cooperative relation matrix is updated through the environmental risk of the obstacles around the unmanned aerial vehicle, and the performance of unmanned aerial vehicle cluster route planning is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle path planning, in particular to a method for cooperative path planning of unmanned aerial vehicle cluster based on graph neural network. BACKGROUND

[0002] The cooperative operation of unmanned aerial vehicle cluster is increasingly widely used in the fields of reconnaissance and surveillance, emergency rescue, and material delivery. The core challenge is to achieve efficient cooperative path planning in a dynamic and uncertain environment. Traditional cooperative planning methods are mostly based on pre-defined fixed communication topology or formation structure. In the face of complex scenarios such as real-time updates of task targets, sudden appearance of obstacles, and unstable communication links caused by electromagnetic interference, these methods lack flexibility and adaptability, making it difficult to ensure the continuous and safe cooperation of the cluster.

[0003] Graph Neural Network (GNN) has strong graph structure data processing capability and can effectively mine the correlation information between nodes. By abstracting unmanned aerial vehicles as nodes in the graph and the cooperative relationships between unmanned aerial vehicles such as communication connection, perception interaction, and obstacle avoidance coupling as edges, the CNN model can effectively model the cooperative interaction in the cluster, thereby generating a dynamically adaptive cluster path planning scheme.

[0004] However, directly applying existing graph neural networks to dynamic cooperative planning of unmanned aerial vehicle cluster still faces fundamental challenges. The core problem is that existing CNN methods usually assume that the graph structure, i.e., the connection relationship between nodes, is static or pre-defined. In actual flight processes, due to task allocation leading to dispersion of formation, changes in communication radius due to interference, or temporary maneuvers to avoid obstacles, the cooperative relationship between unmanned aerial vehicles is dynamically evolving in real time. This dynamic nature leads to a mismatch between the pre-defined fixed graph structure and the real-time physical coupling relationship, causing the CNN model based on fixed graph structure to lose its timeliness and even fail, seriously affecting the overall cooperative performance and task reliability in the cluster, resulting in inaccurate unmanned aerial vehicle path planning. SUMMARY

[0005] To solve the technical problem that the graph neural network based on fixed graph structure cannot adapt to the real-time dynamic changes of the cooperative relationship of the unmanned aerial vehicle cluster, leading to a decline in cooperative path planning performance, the purpose of the present application is to provide a method for cooperative path planning of unmanned aerial vehicle cluster based on graph neural network. The technical solution adopted is as follows: One embodiment of the present application provides a method for cooperative path planning of unmanned aerial vehicle cluster based on graph neural network, which comprises: Obtaining the cooperative relationship matrix of the unmanned aerial vehicle cluster in each analysis period, and the static obstacles and dynamic obstacles of each unmanned aerial vehicle in each analysis period; the static obstacles include static uncontrollable obstacles, and the dynamic obstacles include dynamic uncontrollable obstacles; determine the static uncontrollability and the dynamic uncontrollability of the UAV in each analysis period based on the change of the static uncontrollable obstacle and the dynamic uncontrollable obstacle of the UAV in each analysis period relative to the previous analysis period; determine the dynamic modified uncontrollability of the UAV in each analysis period based on the dynamic uncontrollable obstacle of the UAV in each analysis period in the heading mutation degree of the previous analysis period, and determine the surrounding obstacle degree of the UAV in each analysis period based on the static uncontrollability and the dynamic modified uncontrollability; update the coordination relationship matrix of each analysis period based on the difference of the surrounding obstacle degree of each UAV and the surrounding UAV in each analysis period, determine the modified relationship matrix of the UAV cluster in each analysis period, and plan the route of the UAV cluster.

[0006] Further, the method for obtaining the static uncontrollability comprises: record the static uncontrollable obstacle and the dynamic uncontrollable obstacle as uncontrollable obstacles, and obtain the candidate labels of each uncontrollable obstacle of the UAV in each analysis period; set the state uncontrollability of each uncontrollable obstacle of the UAV in the first analysis period as a constant 1; for the analysis period after the first analysis period, determine whether the set composed of the candidate labels of each uncontrollable obstacle of the UAV in each analysis period is a subset of the set composed of the candidate labels of the same uncontrollable obstacle of the previous analysis period, if yes, the state uncontrollability of each uncontrollable obstacle of the UAV in each analysis period is equal to the state uncontrollability of the same uncontrollable obstacle of the previous analysis period; if not, select a new obstacle from each uncontrollable obstacle of the UAV in each analysis period, and the candidate label of the new obstacle does not belong to the elements in the set composed of the candidate labels of the same uncontrollable obstacle of the previous analysis period of the UAV in each analysis period; adjust the state uncontrollability of each uncontrollable obstacle of the UAV in each analysis period of the previous analysis period based on the number of the new obstacles, and determine the state uncontrollability of each uncontrollable obstacle of the UAV in each analysis period; record the state uncontrollability of the static uncontrollable obstacle and the dynamic uncontrollable obstacle of the UAV in each analysis period as the static uncontrollability and the dynamic uncontrollability of the UAV in each analysis period, respectively.

[0007] Further, the method for determining the state uncontrollability of each uncontrollable obstacle of the UAV in each analysis period comprises: The number ratio of the new obstacles of the unmanned aerial vehicle in each analysis period in each uncontrollable obstacle is taken as the obstacle complexity of each uncontrollable obstacle; The sum of the constant 1 and the obstacle complexity of the unmanned aerial vehicle in each analysis period in each uncontrollable obstacle is calculated, and the product of the state uncontrollability of the same uncontrollable obstacle of the unmanned aerial vehicle in the previous analysis period and the sum is normalized to obtain the state uncontrollability of each uncontrollable obstacle of the unmanned aerial vehicle in each analysis period.

[0008] Further, the determination of the dynamic correction uncontrollability of the unmanned aerial vehicle in each analysis period comprises: The running trajectory of each dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period in the corresponding period and the heading angle of each trajectory point on the running trajectory are obtained. The absolute difference between the heading angle of each trajectory point and the adjacent next trajectory point on the running trajectory of each dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period is taken as the heading change degree of each trajectory point, and the trajectory point with the heading change degree greater than a preset change threshold is selected from the running trajectory and is recorded as a trajectory inflection point. The number ratio of the trajectory inflection point in the trajectory point on the running trajectory of the dynamic uncontrollable obstacle is taken as the inflection point ratio, the mean value of the heading change degree of all trajectory inflection points on the running trajectory of the dynamic uncontrollable obstacle is taken as the trajectory change degree, and the motion unpredictability of each dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period is obtained according to the inflection point ratio and the trajectory change degree. The mean value of the motion unpredictability of the same dynamic uncontrollable obstacle in all previous analysis periods corresponding to the same dynamic uncontrollable obstacle of the unmanned aerial vehicle in each analysis period is taken as the comprehensive trajectory complexity of each uncontrollable obstacle of the unmanned aerial vehicle in each analysis period. The mean value of the comprehensive trajectory complexity of all dynamic uncontrollable obstacles of the unmanned aerial vehicle in each analysis period is used to weight the dynamic uncontrollability of the unmanned aerial vehicle in the corresponding period to obtain the dynamic correction uncontrollability of the unmanned aerial vehicle in each analysis period.

[0009] Further, the determination of the surrounding obstacle degree of the unmanned aerial vehicle in each analysis period comprises: The number ratio of the static uncontrollable obstacle in the static obstacle of the unmanned aerial vehicle in each analysis period and the number ratio of the dynamic uncontrollable obstacle in the dynamic obstacle are sequentially recorded as a static weight and a dynamic weight. The static uncontrollability and the dynamic correction uncontrollability of the unmanned aerial vehicle in each analysis period are weighted and fused based on the static weight and the dynamic weight to obtain the surrounding obstacle degree of the unmanned aerial vehicle in each analysis period.

[0010] Further, the determination of the correction relationship matrix of the UAV cluster in each analysis period comprises: Optionally, one of the UAVs is denoted as an example UAV, and one of the analysis periods is denoted as an example period; an adjacent vector corresponding to the example UAV is extracted from a cooperative relationship matrix of a previous analysis period of the example period; and the UAVs corresponding to the elements with a value of 1 in the adjacent vector are taken as adjacent UAVs of the example UAV. A target UAV is selected from the adjacent UAVs, and the peripheral obstacle degree of the example UAV in the example period is greater than the peripheral obstacle degree of the target UAV in the example period; It is determined whether the number of the target UAVs is greater than or equal to a preset number threshold, and if yes, elements corresponding to all the adjacent UAVs in the adjacent vector in the cooperative relationship matrix of the example UAV in the example period are set to 1, to obtain the correction relationship matrix of the UAV cluster in the example period; If not, elements corresponding to the non-target UAVs in the adjacent vector in the cooperative relationship matrix of the example UAV in the example period are set to 1, to obtain the correction relationship matrix of the UAV cluster in the example period.

[0011] Further, the inflection point ratio and the trajectory change degree are positively correlated with the motion unpredictability.

[0012] Further, the static obstacles and the dynamic obstacles of the UAV in each analysis period are static obstacles and dynamic obstacles in a horizontal circular region with the trajectory point of each UAV at the start time of each analysis period as the center and a preset length as the radius.

[0013] Further, the preset change threshold is 30 degrees.

[0014] Further, the cooperative relationship matrix is a symmetric matrix.

[0015] The present application has the following advantages: First aspect: In view of the problem that the existing graph neural network has a performance decline when the real-time maneuvering of the UAV and the change of the communication change, the present application can sensitively capture the evolution trend of the static risk and the dynamic risk of the local environment of the UAV by analyzing the change of the static uncontrollable obstacles and the dynamic uncontrollable obstacles of the UAV in each analysis period relative to the previous analysis period, and fuse the static risk and the dynamic risk into a peripheral obstacle degree to more accurately represent the overall threat situation of the local area where the UAV is located, and dynamically update the cooperative relationship matrix based on this, so that the correction relationship matrix can reflect the cooperative demand evolved due to the change of the environmental threat distribution of the UAV cluster in real time, fundamentally guarantee the consistency of the coupling relationship between the graph model input and the real world, and significantly improve the accuracy of the flight path planning and the immediate response ability to dynamic scenes.

[0016] The second aspect: considering the motion characteristics of large dynamic obstacles will introduce additional uncertainty, introducing the analysis of the historical heading mutation degree of the dynamic uncontrollable obstacle to adjust the risk threat level presented by the dynamic uncontrollable degree, so that the dynamic modified uncontrollable degree more truly reflects the real risk of the dynamic obstacle in the time-varying environment, and provides key information support for the unmanned aerial vehicle to plan a safer and more robust flight path in a complex dynamic environment.

[0017] The third aspect: by comparing the risk difference between each unmanned aerial vehicle and the surrounding unmanned aerial vehicles, adjusting the connection relationship between the unmanned aerial vehicles, that is, updating the cooperation relationship matrix, so as to realize the adaptive optimization of the communication topology of the unmanned aerial vehicle cluster, and to balance the cooperative robustness and operation efficiency in a complex dynamic environment.

[0018] In summary, the scheme updates the graph structure driven by the environment, gives the graph neural network the ability to process real-time changing cooperative relationship, and finally realizes the high reliability and high cooperativity of the autonomous flight path planning of the unmanned aerial vehicle cluster in the dynamic uncontrollable environment. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0020] Figure 1 A step flow chart of a cooperative flight path planning method for an unmanned aerial vehicle cluster based on a graph neural network provided by an embodiment of the present application; Figure 2 A system structure diagram of a cooperative flight path planning system for an unmanned aerial vehicle cluster based on a graph neural network provided by an embodiment of the present application; Figure 3 A computer device schematic diagram of a cooperative flight path planning device for an unmanned aerial vehicle cluster based on a graph neural network provided by an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined invention purpose, the following combines the drawings and preferred embodiments to specifically describe the cooperative flight path planning method for an unmanned aerial vehicle cluster based on a graph neural network according to the present application, its specific implementation, structure, features and effects, which are described in detail as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0023] The specific scheme of the method for cooperative path planning of a UAV cluster based on a graph neural network provided by the present application is described below in detail with reference to the accompanying drawings.

[0024] Embodiment 1 The present application provides a method for cooperative path planning of a UAV cluster based on a graph neural network, please refer to Figure 1 , which shows a step flow chart of a method for cooperative path planning of a UAV cluster based on a graph neural network provided by an embodiment of the present application, which comprises: Step S1: Obtain the cooperative relationship matrix of the UAV cluster in each analysis period, and the static obstacles and dynamic obstacles of each UAV in each analysis period; the static obstacles include static uncontrollable obstacles, and the dynamic obstacles include dynamic uncontrollable obstacles.

[0025] Through the multi-sensor fusion module carried by each UAV in the UAV cluster, the real-time synchronous acquisition of the self-state data of the UAV is realized, including the longitude, latitude and height position and the heading angle in the global coordinate system. The laser radar and the visual camera of each UAV work synchronously, and at the beginning of each analysis period, the obstacles in the preset neighborhood range of the UAV are detected to obtain the outline and the relative position in the sensor coordinate system, and these obstacles are recorded as the obstacles of the UAV in the corresponding analysis period. In order to realize the cluster-level cooperative perception, the local observation of each UAV needs to be unified to the global navigation coordinate system such as the station-centered coordinate system, and the specific process is as follows: through pre-calibration, the transformation relationship from the sensor coordinate system to the UAV body coordinate system is established, and then combined with the real-time pose of the UAV such as the position, attitude angle, etc., the rotation-translation matrix from the UAV body coordinate system to the global coordinate system is calculated; the relative coordinates of each obstacle are sequentially applied to the above transformation to obtain its global coordinates.

[0026] The displacement of the obstacles of the UAV in each analysis period in the period is calculated, and if the displacement is less than a preset distance threshold such as 0.5 meters, it is determined as a static obstacle, otherwise it is determined as a dynamic obstacle. The static obstacle outline obtained by the visual camera is matched with the UAV outline model predefined in the profile library pre-stored by the ground control center, and the matching method can adopt similarity calculation based on shape context descriptor or deep learning feature vector, if the static obstacle outline is successfully matched with any controllable model in the profile library, it is marked as a static controllable obstacle, otherwise it is marked as a static uncontrollable obstacle. The controllability discrimination of the dynamic obstacle adopts the same process. Among them, the profile library contains multi-view profile models of known UAV models, common aircrafts and other preset cooperative targets.

[0027] The basic parameters of the system mission are set by the ground control center during initialization. During the mission, the system can dynamically receive updates from the superior command system, enabling online updates of parameters such as mission objectives and priorities. At the initial moment of the entire collaborative mission, based on the initial spatial formation state of the UAV swarm, a unique node identifier is assigned to each UAV in the swarm. Based on the general principles of graph construction using graph neural networks, an attribute graph for each analysis cycle is constructed, with UAV nodes as vertices and the communication connections between all UAVs in the swarm at the beginning of each analysis cycle as edges. Specifically, according to a preset communication radius threshold, the Euclidean distance between any two UAVs at the beginning of each analysis cycle is calculated. If this distance is less than or equal to the communication radius, an undirected edge is established between the corresponding nodes, indicating that they can communicate directly; otherwise, no connection is established. After traversing all UAV node pairs, a binary symmetric adjacency matrix is ​​obtained, denoted as the collaborative relationship matrix for each analysis cycle. The collaborative relationship matrix represents the communication topology within the UAV cluster at the start of the analysis period. The number of rows and columns are the total number of UAVs in the cluster. An element value of 1 in the matrix indicates that two UAVs are connected, and 0 indicates that two UAVs are not connected.

[0028] In one implementation of this invention, since most collaborative decisions, such as obstacle avoidance and route planning, primarily consider two-dimensional planar projection relationships, a horizontal circular area with a radius of 100 meters, centered on the trajectory point of the UAV at the beginning of each analysis cycle, is defined as the preset neighborhood range of the UAV in the corresponding analysis cycle. If the collaborative task involves complex three-dimensional airspace, the preset neighborhood range is defined as a spherical space centered on the UAV to ensure perception of threats above and below.

[0029] In one implementation of this invention, the data acquisition frequency of all sensors in the multi-sensor fusion module carried by the UAV is set to 10 Hz.

[0030] In one implementation of this invention, the analysis period is set to 1 minute.

[0031] Step S2: Based on the changes in static and dynamic uncontrollable obstacles of the UAV in each analysis cycle relative to the previous analysis cycle, determine the static and dynamic uncontrollability of the UAV in each analysis cycle.

[0032] By analyzing the changes in static and dynamic uncontrollable obstacles for UAVs in each analysis cycle relative to the previous cycle, the evolution trends of static and dynamic risks in the UAV's local environment can be sensitively captured, yielding static and dynamic uncontrollability. These two indicators provide the core metrics driving graph structure updates. Fixed graph structures fail precisely because they cannot reflect the risk evolution of the UAV's local environment. These two metrics provide a basis for subsequent judgments on adjustments to the collaborative relationships between UAVs, solving the problem of not being able to perceive dynamic changes in node states in mismatch issues.

[0033] Step S3: Based on the degree of abrupt change in the heading of the UAV in the previous analysis period due to the dynamic uncontrollable obstacles in each analysis period, adjust the dynamic uncontrollability and determine the dynamic corrected uncontrollability of the UAV in each analysis period; based on the static uncontrollability and the dynamic corrected uncontrollability, determine the surrounding obstacle degree of the UAV in each analysis period.

[0034] The threat posed by dynamic obstacles lies not only in their state of existence but also in their unpredictable movement patterns, which introduce additional uncertainties. The degree of abrupt changes in the course of dynamic uncontrollable obstacles in each analysis cycle compared to previous analysis cycles measures the inherent uncertainty of the obstacle's movement pattern. Adjusting the dynamic uncontrollability, determined solely by changes in the state of the dynamic uncontrollable obstacle, allows the dynamic correction of uncontrollability to more accurately reflect the true risk of dynamic obstacles in a time-varying environment. In actual flight, UAVs face threats from both static and dynamic obstacles, each with different uncontrollability characteristics. Integrating static and dynamic threats—i.e., static uncontrollability and dynamic correction of uncontrollability—into a surrounding obstacle degree provides a more accurate characterization of the overall threat situation in the local area where the UAV is located. Ensuring that subsequent signals used to update the graph structure are high-quality and information-rich leads to more precise dynamic adjustments to the graph structure, directly improving the quality and reliability of route planning based on this dynamic graph.

[0035] Step S4: Based on the difference in the degree of surrounding obstacles between each UAV and the surrounding UAVs in each analysis period, update the cooperative relationship matrix for each analysis period, determine the corrected relationship matrix of the UAV cluster in each analysis period, and perform route planning for the UAV cluster.

[0036] The surrounding obstacle degree measures the threat level of the local environment in which the UAV is located during the analysis period. By comparing the risk differences between each UAV and its surrounding UAVs, the connection relationships between UAVs are adjusted, i.e., the cooperative relationship matrix is ​​updated, thereby achieving adaptive optimization of the UAV swarm communication topology. This can balance cooperative robustness and operational efficiency in complex dynamic environments. The corrected relationship matrix is ​​a dynamic graph that matches the current threat distribution in the environment. Using it for route planning can completely solve the mismatch problem between the fixed graph structure and dynamic physical coupling, thus ensuring the accuracy and timeliness of cooperative route planning.

[0037] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining static uncontrollability and dynamic uncontrollability includes: denoting static uncontrollable obstacles and dynamic uncontrollable obstacles as uncontrollable obstacles; obtaining candidate labels for each uncontrollable obstacle of the UAV in each analysis cycle; setting the state uncontrollability of each uncontrollable obstacle of the UAV in the first analysis cycle to a constant 1; for analysis cycles after the first analysis cycle, determining whether the set of candidate labels for each uncontrollable obstacle of the UAV in each analysis cycle is a subset of the set of candidate labels for the same type of uncontrollable obstacle in the previous analysis cycle; if so, then the state uncontrollability of each uncontrollable obstacle of the UAV in each analysis cycle is equal to the set of candidate labels for the same type of uncontrollable obstacle in the previous analysis cycle. The state uncontrollability of the same type of uncontrollable obstacle in each analysis cycle is determined; if not, a new obstacle is selected from each type of uncontrollable obstacle in each analysis cycle. The candidate labels of the new obstacle are not elements in the set of candidate labels of the same type of uncontrollable obstacle in the previous analysis cycle of the UAV in each analysis cycle. The number of new obstacles is used to adjust the state uncontrollability of each type of uncontrollable obstacle in the previous analysis cycle of the UAV in each analysis cycle, and the state uncontrollability of each type of uncontrollable obstacle in each analysis cycle of the UAV is determined. The state uncontrollability of the static uncontrollable obstacle and the dynamic uncontrollable obstacle in each analysis cycle of the UAV are recorded as the static uncontrollability and the dynamic uncontrollability of the UAV in each analysis cycle, respectively.

[0038] In this embodiment of the invention, the method for obtaining candidate labels for uncontrollable obstacles includes: a pre-stored contour library in the ground control center contains predefined obstacle contour templates of various categories; the contour of the uncontrollable obstacle is matched with all predefined obstacle contour templates in the contour library, and the similarity between the uncontrollable obstacle and each template is calculated; the category corresponding to the template with the highest similarity is selected as the candidate label for the uncontrollable obstacle. The shape matching algorithm used to calculate the similarity includes at least one of the following: Hu moment algorithm, Fourier descriptor algorithm, or feature matching algorithm based on deep learning; the predefined obstacle contour templates of various categories include at least one of the following: bird contours of different species and postures, drone or aircraft contours, building structure contours, and natural obstacle contours.

[0039] It should be noted that the UAV has no prior knowledge of the environment in the first analysis cycle, and all obstacle types are unknown. Therefore, the state uncontrollability is set to the maximum value of 1. For subsequent analysis cycles, if the set of candidate labels for each type of uncontrollable obstacle in each analysis cycle is a subset of the set of candidate labels for the same type of uncontrollable obstacle in the previous analysis cycle, it indicates that the UAV has not discovered any new obstacle types, and the UAV faces a relatively stable environmental risk. No adjustment to the state uncontrollability is needed; that is, the state uncontrollability of the UAV in each analysis cycle is equal to that of the same type of uncontrollable obstacle in the previous analysis cycle. Conversely, it indicates that the UAV has discovered new obstacle types, meaning that the UAV faces new unknown threats, i.e., the surrounding environment is more complex. The number of newly added obstacles quantifies the scale of the new threat; a larger number indicates a more drastic environmental change. Based on historical estimates, the information of newly added obstacles is used for correction, so that the state uncontrollability of the UAV in each analysis cycle both inherits the historical state and responds to the latest obstacle changes. The specific calculation method is as follows: In this embodiment of the invention, the proportion of newly added obstacles among each type of uncontrollable obstacle in each analysis cycle of the UAV is used as the obstacle complexity of each type of uncontrollable obstacle. The sum of a constant 1 and the obstacle complexity of each type of uncontrollable obstacle of the UAV in each analysis cycle is calculated. The product of the state uncontrollability of the same type of uncontrollable obstacle in the previous analysis cycle of the UAV in each analysis cycle and the sum is normalized to obtain the state uncontrollability of each type of uncontrollable obstacle of the UAV in each analysis cycle. A higher state uncontrollability indicates a significant increase in the UAV's cognitive uncertainty regarding the types of obstacles in its local environment, i.e., a more complex environment.

[0040] In this embodiment of the invention, based on the state uncontrollability of each uncontrollable obstacle in each analysis cycle for all UAVs before each analysis cycle, the state uncontrollability of the same type of uncontrollable obstacle for each UAV in each analysis cycle is normalized using the Z-score normalization method; the Sigmoid function can also be used for normalization.

[0041] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the dynamic correction of uncontrollability includes: obtaining the running trajectory of each dynamic uncontrollable obstacle of the UAV in each analysis cycle, and the heading angle of each trajectory point on the running trajectory; taking the absolute difference between the heading angle of each trajectory point on the running trajectory of each dynamic uncontrollable obstacle of the UAV in each analysis cycle and the next adjacent trajectory point as the heading change degree of each trajectory point; selecting trajectory points on the running trajectory whose heading change degree is greater than a preset change threshold, and recording them as trajectory inflection points; taking the proportion of trajectory inflection points among the trajectory points on the running trajectory of the dynamic uncontrollable obstacle as the inflection point ratio; and taking the running trajectory of the dynamic uncontrollable obstacle as the inflection point ratio. The mean of the heading change at all trajectory inflection points is taken as the trajectory change degree. Based on the inflection point ratio and the trajectory change degree, the motion unpredictability of each dynamic uncontrollable obstacle of the UAV in each analysis period is obtained. The mean of the motion unpredictability of each dynamic uncontrollable obstacle of the UAV in each analysis period and the corresponding dynamic uncontrollable obstacle in all previous analysis periods is taken as the comprehensive trajectory complexity of each uncontrollable obstacle of the UAV in each analysis period. Using the mean of the comprehensive trajectory complexity of all dynamic uncontrollable obstacles of the UAV in each analysis period, the dynamic uncontrollability of the UAV in the corresponding period is weighted to obtain the dynamic corrected uncontrollability of the UAV in each analysis period.

[0042] In this embodiment of the invention, the method for obtaining the running trajectory is as follows: The three-dimensional position coordinates of the dynamic uncontrollable obstacle in the global coordinate system at each sampling moment within the analysis period are used. The sequence of all three-dimensional position coordinates of the dynamic uncontrollable obstacle arranged chronologically within the analysis period is taken as the running trajectory of the obstacle during the analysis period. When the dynamic obstacle experiences partial monitoring loss during continuous monitoring, the system analyzes only the trajectory point sequence obtained within each continuous monitoring period. The continuous monitoring period refers to the time interval during which the obstacle is continuously tracked without any loss of tracking. For obstacles where monitoring is interrupted, their running trajectory consists of trajectory segments obtained within multiple continuous monitoring periods. The three-dimensional position coordinates of the dynamic uncontrollable obstacle in the global coordinate system at each sampling moment are denoted as trajectory points.

[0043] It should be noted that the heading angle quantifies the obstacle's direction of motion at each moment, while the heading change measure the magnitude of the direction change between adjacent moments. A larger heading change indicates sharp turns, evasive maneuvers, or sudden, irregular changes in direction, reflecting the non-smoothness and unpredictability of the trajectory. Trajectory inflection points are points on the trajectory where the direction of motion changes significantly, concretely demonstrating the complexity of the motion and the unpredictability of its intent, such as birds randomly turning or out-of-control drones tumbling irregularly. The inflection point ratio characterizes the irregularity of the motion from a frequency perspective; a larger value means the obstacle frequently performs unpredictable maneuvers. The trajectory change measure characterizes the intensity of the maneuver from an intensity perspective; a larger value means the obstacle's direction change is more abrupt, and the greater the possibility of abrupt changes in the motion state, i.e., the stronger the unpredictability of the motion state. Therefore, both the inflection point ratio and the trajectory change measure are positively correlated with motion unpredictability. In this embodiment of the invention, the product of the inflection point ratio and the trajectory change measure for each dynamic uncontrollable obstacle of the drone in each analysis period is normalized to obtain the motion unpredictability. A larger motion unpredictability means that the dynamic uncontrollable obstacle operates more complexly and unpredictably within the analysis period.

[0044] In this embodiment of the invention, based on the product of the inflection point ratio and trajectory change of all dynamic uncontrollable obstacles of all UAVs in the same analysis period, the product of the inflection point ratio and trajectory change of each UAV in the corresponding analysis period is normalized using the minimax normalization method.

[0045] The unpredictability of motion in a single analysis cycle may fluctuate drastically due to noise or accidental maneuvers. By calculating the mean of the unpredictability of motion in previous analysis cycles, the inherent behavioral patterns of obstacles are reflected, making the assessment results more reliable. The mean of the comprehensive trajectory complexity of all dynamically uncontrollable obstacles for the UAV in each analysis cycle reflects the intensity of the threat posed by the UAV from the motion behavior of all dynamic obstacles in its current local airspace. Adjusting the dynamic uncontrollability using this mean allows the UAV to more realistically reflect its real-time risks in a time-varying environment. In this embodiment of the invention, the mean of the comprehensive trajectory complexity of all dynamically uncontrollable obstacles for the UAV in each analysis cycle is calculated. The product of this mean and the dynamic uncontrollability of the UAV in the corresponding cycle is used as the dynamically corrected uncontrollability of the UAV in each analysis cycle.

[0046] In one implementation of this invention, the preset change threshold is set to 30 degrees, which can be set by the implementer according to specific circumstances.

[0047] It should be noted that the fusion contour feature vector of dynamic uncontrollable obstacles extracted from LiDAR point clouds and visual images may include elements such as principal component analysis shape features based on point clouds, 3D bounding box size, and CNN depth features or traditional contour moments extracted from visual images. Calculate the cosine similarity of the fused contour feature vectors of each dynamic uncontrollable obstacle of the UAV in each analysis cycle to the dynamic uncontrollable obstacles in all previous analysis cycles. Select the dynamic uncontrollable obstacle with the highest cosine similarity from each previous analysis cycle as the same dynamic uncontrollable obstacle corresponding to each dynamic uncontrollable obstacle of the UAV in each analysis cycle in all previous analysis cycles. If a certain cycle does not exist, then that analysis cycle is not included in the analysis.

[0048] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the surrounding obstacle degree includes: recording the proportion of static uncontrollable obstacles in the static obstacles and the proportion of dynamic uncontrollable obstacles in the dynamic obstacles of the UAV in each analysis cycle as static weight and dynamic weight, respectively; and weighting and fusing the static uncontrollability and dynamic corrected uncontrollability of the UAV in each analysis cycle based on the static weight and dynamic weight to obtain the surrounding obstacle degree of the UAV in each analysis cycle.

[0049] It should be noted that static weights measure the proportion of unknown or difficult-to-handle obstacles in a static environment, while dynamic weights measure the proportion of unpredictable obstacles in a dynamic environment, i.e., the degree of unpredictability in a dynamic environment. Static uncontrollability assesses the novelty and cognitive uncertainty of static obstacles in the environment surrounding the drone, while dynamic corrected uncontrollability assesses the unpredictability and complexity of the movement patterns of moving obstacles in the environment surrounding the drone. The weighted fusion of these two factors yields a more accurate characterization of the overall threat situation in the local area where the drone is located. The higher the peripheral obstacle degree, the higher the threat level of the local environment in which the drone is located during the analysis period.

[0050] In one specific implementation of this invention, the surrounding obstacle degree is expressed by the formula: ; In the formula, Z represents the degree of surrounding obstacles for each UAV in each analysis cycle; The number of static, uncontrollable obstacles for each drone in each analysis cycle; The number of static obstacles for each drone in each analysis cycle; Static weights for each drone in each analysis period; The static uncontrollability of each drone in the current analysis period; The number of dynamically uncontrollable obstacles for each drone in each analysis cycle; The number of dynamic obstacles for each drone in each analysis cycle; Dynamic weights for each drone in each analysis cycle; The uncontrollability of each drone is dynamically corrected in each analysis period. It is important to note that the surrounding obstacle degree is obtained using the same method as described above for all drones in the drone swarm, calculating the surrounding obstacle degree of each drone in the swarm in each analysis period.

[0051] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the modified relation matrix includes: arbitrarily selecting one drone as an example drone and one analysis period as an example period; extracting the adjacency vector corresponding to the example drone from the collaborative relation matrix of the previous analysis period of the example period; taking the drones corresponding to the non-zero elements in the adjacency vector as the adjacency drones of the example drone; selecting a target drone from the adjacency drones, wherein the surrounding obstacle degree of the example drone in the example period is greater than the surrounding obstacle degree of the target drone in the example period; determining whether the number of target drones is greater than or equal to a preset number threshold; if so, setting all elements corresponding to the adjacency vectors of the example drone in the collaborative relation matrix of the example period to 1 to obtain the modified relation matrix of the drone cluster in the example period; if not, setting the elements corresponding to the non-target drones in the adjacency vectors of the example drone in the collaborative relation matrix of the example period to 1 to obtain the modified relation matrix of the drone cluster in the example period.

[0052] It should be noted that the surrounding obstacle level measures the threat level of the local environment in which the drone is located during the analysis period. Target drones are relatively safe and have more stable routes in the local environment during the example period, and their route decisions are likely to be more reliable and less affected by environmental interference. Non-target drones may undergo violent path maneuvers due to emergency obstacle avoidance in the local environment during the example period. If the number of target drones is greater than or equal to a preset threshold, it indicates that the example drones are more dangerous than most of their neighbors. In this case, a "safety-first" strategy should be adopted, maintaining the connection between the example drones and neighboring drones to ensure that the state changes of high-risk example drones can be perceived by all neighbors in real time, thereby enabling collaborative avoidance and reducing collision risk. Conversely, if the risk of example drones is not significantly higher than that of most neighbors during the example period, and the environment in which the target drones are located is favorable and less affected by the uncertainty of the example drones, connecting the example drones with the target drones may bring unnecessary communication overhead and computational burden. In this case, an "efficiency-first" strategy should be adopted, maintaining connection only with non-target drones with higher local environmental threat levels, ensuring necessary information flow while reducing redundant communication and optimizing network load and decision-making efficiency. By responding in real time to the differences in environmental risks faced by individual drones and dynamically adjusting the connection strategy based on the modified relationship matrix, the drone swarm communication network can proactively adapt to external obstacle interference, thereby improving the overall survivability and mission continuity of the system in complex airspace.

[0053] In one implementation of this invention, the preset quantity threshold is set as the product of the total number of neighboring UAVs of the example UAV and the ratio optimization value. The specific method for optimizing the ratio is as follows: A cluster task simulation model incorporating an adjacency update strategy is established. Under various typical task scenarios such as dense obstacle environments and dynamic target tracking, the overall task success rate, average flight path length, communication overhead, and collision avoidance rate are used as comprehensive performance indicators. A ratio parameter value is obtained by scanning parameters from 0.5 to 0.9 in steps of 0.05. The ratio parameter value that optimizes the comprehensive performance indicators is selected as the ratio optimization value.

[0054] As an example, suppose the collaboration relationship matrix G1 of the drone swarm in the first analysis period and the collaboration relationship matrix G2 in the second analysis period are as follows: , In both matrices, the i-th row or column represents the connection of the i-th drone in the drone swarm, where i is an integer between 1 and 4. If the example drone is the 3rd drone, its adjacency vector in the first analysis period is the 3rd row of matrix G1. With column 3 In the two adjacency vectors, the second and fourth elements are 1, corresponding to the 2nd and 4th drones respectively. These two drones are adjacent to the 3rd drone. The adjacency vector corresponding to the 3rd drone in matrix G2 is in the 3rd row. With column 3 The adjacent drones, i.e., the 2nd and 4th drones, correspond to the 2nd and 4th elements in the adjacency vector. Assuming the 4th drone is the target drone, if the number of target drones is greater than or equal to a preset threshold, since the 2nd element in the adjacency vector is 1, only the 4th element in the adjacency vector needs to be set to 1. Then, the correction relation matrix for the second analysis period... If the number of target drones is less than a preset threshold, the non-target drones are the other adjacent drones (i.e., the second drone). The second element in the adjacency vector is 1, so the adjacency vector does not need to be changed. Therefore, the correction relation matrix for the second analysis period... .

[0055] It should be noted that the collaborative relationship matrix of the UAV cluster in the first analysis period can be used as the corrected relationship matrix; the method for obtaining the corrected relationship matrix of the UAV cluster in subsequent analysis periods is the same as the method for obtaining the corrected relationship matrix of the UAV cluster in the example period.

[0056] The system uses the aforementioned method to obtain the corrected relation matrix of the UAV swarm in real time for each analysis cycle. This corrected relation matrix, along with the state attributes of all UAVs in that cycle (position, speed, remaining battery power, or current task status), is used to construct the input data graph for a graph neural network. Nodes in the graph represent individual UAVs, their feature vectors contain their state attributes, and the corrected relation matrix precisely defines the connections between nodes. Subsequently, a pre-defined graph neural network model, such as a path planner based on graph convolutional networks or graph attention networks, is used to learn and infer from the input data graph. The model encodes global task constraints (such as formation, overall target point, and task priority) and environmental constraints (such as the complexity of surrounding obstacle distribution indirectly implied by the relation matrix) into the computation process. Through the propagation and aggregation of multi-layer graph neural networks, each UAV node can obtain cooperative state information within its neighborhood and even the entire swarm. Finally, the network output layer generates an optimal flight path for each UAV that balances individual optimality, collision avoidance, swarm cooperation, and dynamic adaptability. This path planning process is executed automatically in each update cycle, allowing the path to dynamically adjust according to the latest swarm topology and risk situation reflected in the relation matrix.

[0057] This invention is now complete.

[0058] Example 2: This invention proposes a collaborative route planning system for UAV swarms based on graph neural networks. Please refer to [link / reference]. Figure 2 The diagram illustrates a system architecture of a UAV swarm cooperative route planning system based on a graph neural network, according to an embodiment of the present invention. The system includes: The data acquisition module 510 is used to acquire the collaborative relationship matrix of the UAV cluster in each analysis period, as well as the static and dynamic obstacles of each UAV in each analysis period; static obstacles include static uncontrollable obstacles, and dynamic obstacles include dynamic uncontrollable obstacles. The obstacle uncontrollability analysis module 520 is used to determine the static uncontrollability and dynamic uncontrollability of the UAV in each analysis cycle based on the changes in static and dynamic uncontrollable obstacles of the UAV in each analysis cycle relative to the previous analysis cycle. The UAV obstacle analysis module 530 is used to adjust the dynamic uncontrollability based on the degree of abrupt change in the heading of the UAV in the previous analysis cycle of the dynamic uncontrollable obstacles in each analysis cycle, and to determine the dynamic corrected uncontrollability of the UAV in each analysis cycle; and to determine the surrounding obstacle degree of the UAV in each analysis cycle based on the static uncontrollability and the dynamic corrected uncontrollability. The route planning module 540 is used to update the collaborative relationship matrix for each analysis period based on the difference in the surrounding obstacle degree between each UAV and the surrounding UAVs in each analysis period, determine the corrected relationship matrix of the UAV cluster in each analysis period, and perform route planning for the UAV cluster.

[0059] It should be noted that the devices provided in the above embodiments are only illustrative examples of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the UAV swarm cooperative route planning system based on graph neural networks and the UAV swarm cooperative route planning method based on graph neural networks provided in the above embodiments belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0060] Example 3: Figure 3 This is a schematic diagram of a computer device for a collaborative route planning system for unmanned aerial vehicle (UAV) swarms based on a graph neural network, provided as an embodiment of the present invention. For example,... Figure 3 As shown, the computer device includes: a memory 601, a processor 602, and a computer program 603 stored in the memory 601 and running on the processor 602. When the processor 602 executes the computer program 603, the computer device can execute any of the aforementioned graph neural network-based UAV swarm cooperative route planning methods.

[0061] Furthermore, embodiments of this application also protect an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute a UAV swarm cooperative route planning method based on graph neural networks provided in embodiments of this application.

[0062] This embodiment can divide the device into functional modules based on the above method example. For example, each module can correspond to a separate function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0063] It should be understood that the apparatus provided in this embodiment is used to execute the above-described method for collaborative route planning of unmanned aerial vehicle (UAV) swarms based on graph neural networks, and therefore can achieve the same effect as the above-described implementation method.

[0064] When using integrated units, the device may include a processing module and a storage module. When applied to a workpiece, the processing module can be used to control and manage the workpiece's operations. The storage module can be used to support the execution of program code by the workpiece.

[0065] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits contained in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0066] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0067] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for collaborative route planning of unmanned aerial vehicle (UAV) swarms based on graph neural networks, characterized in that, The method includes: Obtain the collaborative relationship matrix of the drone swarm in each analysis period, as well as the static and dynamic obstacles of each drone in each analysis period; static obstacles include static uncontrollable obstacles, and dynamic obstacles include dynamic uncontrollable obstacles; Based on the changes in static and dynamic uncontrollable obstacles of the UAV in each analysis cycle relative to the previous analysis cycle, the static and dynamic uncontrollability of the UAV in each analysis cycle is determined. Based on the degree of abrupt change in the heading of the UAV in each analysis cycle due to dynamic uncontrollable obstacles in the previous analysis cycle, the dynamic uncontrollability is adjusted to determine the dynamic corrected uncontrollability of the UAV in each analysis cycle; based on the static uncontrollability and the dynamic corrected uncontrollability, the surrounding obstacle degree of the UAV in each analysis cycle is determined. Based on the differences in the surrounding obstacle degree between each UAV and its surrounding UAVs in each analysis period, the collaborative relationship matrix for each analysis period is updated to determine the corrected relationship matrix of the UAV cluster in each analysis period, and the flight path planning for the UAV cluster is performed.

2. The UAV swarm cooperative route planning method based on graph neural networks according to claim 1, characterized in that, The method for obtaining the static uncontrollability includes: Static and dynamic uncontrollable obstacles are categorized as uncontrollable obstacles; candidate labels for each uncontrollable obstacle of the UAV are obtained in each analysis cycle; Set the uncontrollability of the state of each uncontrollable obstacle of the UAV to a constant of 1 during the first analysis cycle; For each analysis cycle after the first analysis cycle, determine whether the set of candidate labels for each type of uncontrollable obstacle of the UAV in each analysis cycle is a subset of the set of candidate labels for the same type of uncontrollable obstacle in the previous analysis cycle. If so, the state uncontrollability of each type of uncontrollable obstacle of the UAV in each analysis cycle is equal to the state uncontrollability of the same type of uncontrollable obstacle in the previous analysis cycle. If not, then new obstacles are selected from each type of uncontrollable obstacle in each analysis cycle by the UAV, and the candidate labels of the new obstacles are not elements in the set of candidate labels of the same type of uncontrollable obstacle in the previous analysis cycle of each analysis cycle by the UAV. Using the number of newly added obstacles, adjust the uncontrollability of the state of each uncontrollable obstacle in the previous analysis cycle of each analysis cycle of the UAV, and determine the uncontrollability of the state of each uncontrollable obstacle of the UAV in each analysis cycle. The uncontrollability of the static and dynamic uncontrollable obstacles of the UAV in each analysis cycle is denoted as the static uncontrollability and dynamic uncontrollability of the UAV in each analysis cycle, respectively.

3. The UAV swarm cooperative route planning method based on graph neural networks according to claim 2, characterized in that, The determination of the uncontrollability of the state of each uncontrollable obstacle for the UAV in each analysis cycle includes: The percentage of newly added obstacles in each type of uncontrollable obstacle in each analysis cycle of the UAV is used as the obstacle complexity of each type of uncontrollable obstacle; The constant 1 is calculated as the sum of the obstacle complexity of the UAV for each type of uncontrollable obstacle in each analysis cycle. The product of the state uncontrollability of the UAV for the same type of uncontrollable obstacle in the previous analysis cycle and the sum is normalized to obtain the state uncontrollability of the UAV for each type of uncontrollable obstacle in each analysis cycle.

4. The UAV swarm cooperative route planning method based on graph neural networks according to claim 1, characterized in that, The determination of the uncontrollability of dynamic correction of the UAV in each analysis cycle includes: The UAV acquires the trajectory of each dynamic uncontrollable obstacle in each analysis cycle, as well as the heading angle of each trajectory point on the trajectory. The absolute difference between the heading angle of each trajectory point on the trajectory of each dynamic uncontrollable obstacle in each analysis cycle and the next adjacent trajectory point is taken as the heading change degree of each trajectory point; trajectory points with heading change degree greater than a preset change threshold are selected from the trajectory and recorded as trajectory inflection points; The proportion of the number of trajectory inflection points among the trajectory points on the running trajectory of the dynamic uncontrollable obstacle is taken as the inflection point ratio; the average of the heading change of all trajectory inflection points on the running trajectory of the dynamic uncontrollable obstacle is taken as the trajectory change degree; based on the inflection point ratio and the trajectory change degree, the motion unpredictability of each dynamic uncontrollable obstacle of the UAV in each analysis cycle is obtained. The average motion unpredictability of each dynamic uncontrollable obstacle of the UAV in each analysis cycle is taken as the comprehensive trajectory complexity of each uncontrollable obstacle of the UAV in each analysis cycle, corresponding to the same dynamic uncontrollable obstacle in all previous analysis cycles. By using the average of the comprehensive trajectory complexity of all dynamic uncontrollable obstacles of the UAV in each analysis period, the dynamic uncontrollability of the UAV in the corresponding period is weighted to obtain the dynamic corrected uncontrollability of the UAV in each analysis period.

5. The UAV swarm cooperative route planning method based on graph neural networks according to claim 1, characterized in that, Determining the surrounding obstacle degree of the UAV in each analysis cycle includes: The proportion of static uncontrollable obstacles in the static obstacles and the proportion of dynamic uncontrollable obstacles in the dynamic obstacles of the drone in each analysis period are respectively denoted as static weight and dynamic weight. Based on the static weights and the dynamic weights, the static uncontrollability and the dynamic corrected uncontrollability of the UAV in each analysis cycle are weighted and fused to obtain the surrounding obstacle degree of the UAV in each analysis cycle.

6. The UAV swarm cooperative route planning method based on graph neural networks according to claim 1, characterized in that, The determination of the modified relationship matrix of the drone cluster in each analysis period includes: Select one drone as the example drone and one analysis period as the example period; extract the adjacency vector corresponding to the example drone from the collaboration relationship matrix of the previous analysis period of the example period; take the drone corresponding to the element with a value of 1 in the adjacency vector as the neighboring drone of the example drone; Select a target drone from the adjacent drones, wherein the surrounding obstacle degree of the example drone in the example period is greater than the surrounding obstacle degree of the target drone in the example period; Determine whether the number of the target drones is greater than or equal to a preset number threshold. If so, set all the elements corresponding to the neighboring drones in the adjacency vector of the example drone in the collaborative relationship matrix of the example period to 1, and obtain the corrected relationship matrix of the drone cluster in the example period. If not, then set the element corresponding to the non-target drone in the adjacency vector of the example drone in the collaborative relationship matrix of the example period to 1 to obtain the corrected relationship matrix of the drone cluster in the example period.

7. The UAV swarm cooperative route planning method based on graph neural networks according to claim 4, characterized in that, The inflection point ratio and the trajectory change degree are both positively correlated with the motion unpredictability.

8. The UAV swarm cooperative route planning method based on graph neural networks according to claim 4, characterized in that, The static and dynamic obstacles of the UAV in each analysis cycle are static and dynamic obstacles within a horizontal circular area with a preset length as the center, centered on the trajectory point of each UAV at the beginning of each analysis cycle.

9. The UAV swarm cooperative route planning method based on graph neural networks according to claim 4, characterized in that, The preset change threshold is 30 degrees.

10. The UAV swarm cooperative route planning method based on graph neural networks according to claim 1, characterized in that, The cooperative relationship matrix is ​​a symmetric matrix.

Images