A multi-unmanned aerial vehicle path planning method and system based on a graph neural network

Abstract

This application discloses a multi-UAV path planning method and system based on graph neural networks. The method includes: initializing acquired basic information to obtain initialization information; calculating the probability of communication interruption at each spatial location within the planning area corresponding to the radio map based on the initialization information, and determining a communication interruption probability map; calculating a communication quality path cost matrix based on the initialization information; constructing graph structure data based on the communication quality path cost matrix; determining the allocation relationship between the task set and the UAVs based on the graph structure data; calculating the path planning result for each UAV based on the allocation relationship; and optimizing multiple UAV tasks and paths based on the path planning results for each UAV, determining the multiple UAV task allocation results and multiple UAV path planning results. This method can reduce the risk of communication interruption during UAV flight and improve the mission execution reliability and flight safety of multiple UAV systems.

Description

Technical Field

[0001] This application relates to the field of UAV path planning and wireless communication support technology, specifically to a multi-UAV path planning method and system based on graph neural networks. Background Technology

[0002] With the development of applications such as low-altitude economy, smart logistics, emergency rescue, and inspection and monitoring, multiple unmanned aerial vehicle (UAV) systems are gradually becoming an important development direction in the field of UAV applications due to their advantages such as wide mission coverage, high collaborative execution efficiency, and flexible deployment. When multiple UAVs perform delivery, inspection, or search tasks, path planning is typically required based on factors such as mission location, UAV take-off and landing points, flight distance, energy consumption, and safety constraints to ensure that multiple UAVs can collaboratively complete the mission.

[0003] In real-world low-altitude flight environments, UAVs typically rely on air-to-ground wireless links for status feedback, control command reception, and mission information exchange. Due to factors such as urban building obstruction, multipath propagation, terrain undulations, and small-scale fading, the quality of wireless communication varies significantly across different areas. If a UAV passes through areas with poor communication quality during flight, it may experience link quality degradation, unstable data transmission, or even communication interruptions, thus affecting the continuity and reliability of mission execution.

[0004] Existing UAV path planning methods typically prioritize minimizing flight distance, flight time, energy consumption, or mission completion efficiency. They focus on spatial location, task allocation, and flight costs during path search, neglecting the importance of communication quality assurance during UAV flight. In real-world low-altitude flight scenarios, UAVs rely continuously on wireless communication links for flight status feedback, control command reception, mission data transmission, and multi-UAV collaborative information exchange. If a UAV enters an area with poor communication quality, it can lead to increased link latency, data packet loss, abnormal telemetry information feedback, or delayed control command delivery. In severe cases, the UAV may experience short-term or continuous disconnection, making it difficult for ground control to monitor its flight position, attitude, battery level, and mission status, and hindering timely responses to sudden obstacles, weather changes, or mission adjustments. For multi-UAV scenarios, communication interruption of a single UAV can further impact swarm coordination, collision avoidance, and mission continuity, leaving the UAV in a near-blind flight state, increasing the risk of flight path deviation, mission failure, airspace conflicts, and even safety incidents. Summary of the Invention

[0005] This application aims to provide a multi-UAV path planning method and system based on graph neural networks, which can reduce the risk of communication interruption during UAV flight and improve the mission execution reliability and flight safety of multiple UAV systems.

[0006] The technical solution of this application is implemented as follows: In a first aspect, embodiments of this application provide a multi-UAV path planning method based on graph neural networks, the method comprising: Acquire various basic information for path planning of multiple drones; and initialize the basic information to obtain initialization information; wherein, the initialization information includes drone set, drone take-off and landing point location, task set, task pickup and delivery point location, task payload weight, drone flight performance parameters, radio map, communication quality threshold and path planning related parameters; Based on the radio map and the communication quality threshold, the probability of communication interruption at each spatial location within the planning area corresponding to the radio map is calculated, and a communication interruption probability map is determined. Based on the communication interruption probability map, the location of the UAV take-off and landing point, and the location of the task pickup and delivery point, calculate the communication quality path cost matrix between the UAV take-off and landing point, the task pickup point, and the task delivery point. Based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set, a graph structure data is constructed; and based on the graph structure data, the allocation relationship between each task and the UAV in the task set is determined. Based on the allocation relationship between each task and the UAV, the path planning result for each UAV is calculated; and based on the path planning result for each UAV, multiple UAV tasks and paths are optimized to determine multiple UAV task allocation results and multiple UAV path planning results.

[0007] In the above scheme, the step of calculating the communication quality path cost matrix between the UAV take-off and landing point, the UAV take-off and landing point location, and the task pickup and delivery point location based on the communication interruption probability map, the UAV take-off and landing point location, and the task pickup and delivery point location includes: Based on the location of the UAV take-off and landing point and the location of the task pickup and delivery point, calculate the spatial distance information; Based on the communication interruption probability map, the spatial distance information, the preset distance cost weight, and the preset communication interruption risk weight, calculate the single-step path cost of feasible paths between each pair of key nodes in the UAV take-off and landing point, the mission pickup point, and the mission delivery point; wherein, the key nodes are the nodes in the UAV take-off and landing point, the mission pickup point, and the mission delivery point. Based on the single-step path cost of the feasible paths between the two key nodes, the feasible path corresponding to the minimum single-step path cost is selected as the communication quality path between the two key nodes, and the single-step path cost corresponding to the communication quality path is determined as the optimal path cost between the two key nodes. The communication quality path cost matrix is ​​determined based on the optimal path cost between each pair of key nodes.

[0008] In the above scheme, the construction of graph structure data based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set includes: Based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set, determine the feature vector of each task node in the task set; Based on the feature vector of each task node in the task set, calculate the communication risk coupling cost between any two task nodes. Based on the communication risk coupling cost between each pair of task nodes, the neighbor nodes of each task node in the task set are determined; and based on the neighbor nodes of each task node, the neighbor set of each task node is determined. Based on the communication risk coupling cost and the preset temperature coefficient, the normalized edge weights are calculated; The graph structure data is determined based on the feature vector of each task node in the task set, the neighbor set of each task node, and the normalized edge weight.

[0009] In the above scheme, determining the graph structure data based on the feature vector of each task node in the task set, the neighbor set of each task node, and the normalized edge weights includes: By using a graph neural network, the feature vector of each task node in the task set is linearly mapped to obtain the initial embedding representation of the feature vector of each task node in the task set. By using the neighbor set of each task node and the normalized edge weight, the initial embedding representation is subjected to communication risk-weighted message aggregation to determine the neighbor messages of each task node in the task set. Based on the initial embedding representation and the neighbor messages, the node embedding is updated to determine the final embedding representation of each task node in the task set; The final embedded representation of each task node in the task set is determined as the graph structure data.

[0010] In the above scheme, determining the allocation relationship between each task and the UAV in the task set based on the graph structure data includes: Based on the graph structure data, the task embedding values ​​of each of the multiple drones are determined; and based on the task embedding values, the load state vectors of each of the multiple drones are determined. For any task to be assigned in the task set, calculate the communication matching features between any task to be assigned in the task set and each of the multiple UAVs; Based on the graph structure data, the load state vector, and the communication matching features, the scores for each of the multiple drones to be assigned a given task are calculated; and the scores are normalized to obtain the probability of assigning each of the multiple drones to the given task. The drone with the highest probability among the plurality of drones is selected as the drone to perform any of the assigned tasks. Based on the drone corresponding to any one of the tasks to be assigned, determine the allocation relationship between each task and the drone in the task set.

[0011] In the above scheme, calculating the path planning result for each UAV based on the allocation relationship between each task and the UAV includes: Based on the allocation relationship between each task and the UAV, determine the task subset for each UAV; Based on the communication quality path cost matrix, the task subset of each UAV, and the pickup and delivery order constraints, an initial path for each UAV is generated; and the initial path of each UAV is optimized to obtain a candidate path for each UAV. Calculate the comprehensive path cost of the candidate paths for each UAV, and select the candidate path whose comprehensive path cost satisfies the pickup and delivery order constraint, the preset load constraint, and the energy constraint as the latest path for each UAV. The latest path of each UAV is optimized for communication quality to determine the path planning result of each UAV.

[0012] In the above scheme, the path planning results for each UAV include the task access order, the sequence of pickup and delivery nodes, the specific flight path between nodes, the comprehensive path cost, the estimated flight time, and the estimated energy consumption. The step of optimizing multiple drone tasks and paths based on the path planning results of each drone, and determining multiple drone task allocation results and multiple drone path planning results, includes: Under the condition that the path planning result of each UAV meets the preset conditions, multiple UAV task transfers and task exchanges are performed to obtain the multiple UAV task allocation results. Based on the task allocation results of the multiple drones, the path planning results of each of the multiple drones are optimized to determine the path planning results of the multiple drones. The preset conditions include at least one of the following: For each of the multiple drones, the estimated flight time of each drone is greater than the estimated flight time of the remaining drones in the multiple drones. The estimated energy consumption of each drone is greater than the preset energy consumption threshold; The total path cost for each drone exceeds a preset path cost threshold. Each drone has been adjusted up to the preset maximum number of adjustments.

[0013] Secondly, embodiments of this application provide a multi-UAV path planning system based on graph neural networks. This system includes: an acquisition module, a calculation module, and a determination module; wherein... The acquisition module is used to acquire various basic information for path planning of multiple UAVs; and initialize the basic information to obtain initialization information; wherein, the initialization information includes UAV set, UAV take-off and landing point location, task set, task pickup and delivery point location, task payload weight, UAV flight performance parameters, radio map, communication quality threshold and path planning related parameters; The calculation module is used to calculate the probability of communication interruption at each spatial location within the planning area corresponding to the radio map based on the radio map and the communication quality threshold, and to determine the communication interruption probability map; based on the communication interruption probability map, the location of the UAV take-off and landing point and the location of the task pickup and delivery point, to calculate the communication quality path cost matrix between the UAV take-off and landing point, the task pickup point and the task delivery point. The determining module is used to construct graph structure data based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set; and based on the graph structure data, determine the allocation relationship between each task and the UAV in the task set; based on the allocation relationship between each task and the UAV, calculate the path planning result for each UAV; and based on the path planning result for each UAV, perform multiple UAV task and path optimization to determine multiple UAV task allocation results and multiple UAV path planning results.

[0014] Thirdly, embodiments of this application provide a multi-UAV path planning device based on graph neural networks, comprising: a processor and a memory; wherein, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in the first aspect.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing executable instructions for causing a processor to perform the method described in the first aspect.

[0016] This application provides a multi-UAV path planning method and system based on graph neural networks. The method includes: acquiring various basic information for path planning of multiple UAVs; initializing the basic information to obtain initialization information; wherein, the initialization information includes a set of UAVs, UAV take-off and landing point locations, a set of tasks, task pickup and delivery point locations, task payload weight, UAV flight performance parameters, a radio map, a communication quality threshold, and path planning related parameters; based on the radio map and the communication quality threshold, calculating the probability of communication interruption at each spatial location within the planning area corresponding to the radio map, and determining a communication interruption probability map; based on the communication interruption probability map and the UAVs... The communication quality path cost matrix between the UAV take-off and landing point, the task pickup and delivery point, and the task delivery point is calculated based on the take-off and landing point location and the task pickup and delivery point location. A graph structure data is constructed based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set. Based on the graph structure data, the allocation relationship between each task and the UAV in the task set is determined. Based on the allocation relationship between each task and the UAV, the path planning result for each UAV is calculated. Based on the path planning result for each UAV, multiple UAV tasks and paths are optimized to determine multiple UAV task allocation results and multiple UAV path planning results. In the above scheme, based on the radio map, a communication interruption probability graph is constructed by combining communication quality thresholds and impact factors. This graph is then combined with the flight distance to determine the communication quality path cost matrix. A graph neural network is used to learn the correlation between multiple UAV task nodes, path costs, and communication reliability, thereby generating a multi-UAV path planning scheme that balances task execution efficiency and communication quality assurance. This reduces the risk of communication interruption during UAV flight and improves the task execution reliability and flight safety of the multi-UAV system. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0018] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0019] Figure 1 This is an optional flowchart illustrating a multi-UAV path planning method based on graph neural networks, provided as an embodiment of this application. Figure 2 An optional communication interruption probability graph is provided for a multi-UAV path planning method based on graph neural networks in an embodiment of this application; Figure 3 This is a schematic diagram of an optional multi-UAV path planning result provided by an embodiment of the present application for a multi-UAV path planning method based on graph neural networks; Figure 4 A schematic diagram of a multi-UAV path planning system based on graph neural networks is provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of a multi-UAV path planning device based on a graph neural network, provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0021] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application is for the purpose of describing embodiments of this application only and is not intended to be limiting of this application.

[0022] In the following description, references to "some embodiments," "this embodiment," "this application embodiment," and examples, etc., describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.

[0023] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0024] This application provides a multi-UAV path planning method based on graph neural networks. Figure 1 This is an optional flowchart illustrating a multi-UAV path planning method based on graph neural networks provided in an embodiment of this application, which will combine... Figure 1 The steps shown are explained.

[0025] S101. Obtain various basic information for path planning of multiple UAVs; and initialize the basic information to obtain initialization information.

[0026] In some embodiments of this application, the initialization information includes a set of drones, drone take-off and landing point locations, a set of tasks, task pickup and delivery point locations, task payload weight, drone flight performance parameters, radio map, communication quality threshold, and path planning related parameters.

[0027] In some embodiments of this application, a multi-UAV path planning method based on graph neural networks is adapted to multi-UAV path planning scenarios.

[0028] In some embodiments of this application, a multi-UAV path planning method based on graph neural networks is adapted to a multi-UAV path planning system based on graph neural networks.

[0029] In some embodiments of this application, various basic information for path planning of multiple UAVs is obtained, and the basic information is initialized to obtain a set of UAVs, UAV take-off and landing point locations, task set, task pickup and delivery point locations, task payload weight, UAV flight performance parameters, radio map, communication quality threshold, and path planning related parameters.

[0030] For example, the basic information required for high-quality communication path planning for multiple UAVs is uniformly set to provide input data for subsequent construction of communication interruption probability graphs, generation of path cost matrices, assignment of graph neural network tasks, and path planning.

[0031] Suppose that the system contains the following set of drones:

[0032] Where U represents the number of drones, and assuming that each drone u corresponds to a take-off / landing point or hangar node, its location is represented as follows:

[0033] When multiple hangars exist in the system, different drones can correspond to different hangar nodes. Let the set of tasks to be executed be... ,in, This indicates the number of tasks. Each task i includes a pickup point. Delivery point and mission payload weight During the mission, the drone must first visit the corresponding pickup point and then the corresponding delivery point to meet the order constraint of pickup and delivery.

[0034] Simultaneously, initialize the UAV's flight performance parameters, including its cruise speed, maximum flight time, maximum payload capacity, and maximum energy constraint. Let the UAV u's flight speed be... Its speed satisfies:

[0035] And assume the maximum available energy of the drone is This is used to subsequently determine whether the drone's path meets the energy constraints.

[0036] In addition, the radio map corresponding to the planning area is read. The radio map is used to represent the distribution of wireless communication quality at different spatial locations within the planning area. Let the radio map be represented as... , Indicates the spatial location within the planning area. This indicates the average communication quality or received signal strength at that location. A further communication quality threshold is set as follows: This threshold is used to subsequently determine whether there is a risk of communication interruption for the drone in different locations.

[0037] During the initialization phase, path planning parameters also need to be set, including distance weight coefficients, communication quality weight coefficients, path search adjacency methods, communication interruption probability calculation parameters, and graph neural network training parameters. Let the distance weight coefficient be... The communication interruption risk weighting coefficient is The subsequent path cost will be modeled using a weighted average of the two factors.

[0038] The output initialization information includes: the drone set and its take-off and landing point locations, the mission set and its pickup and delivery point locations, mission payload weight, drone flight performance parameters, radio map, communication quality threshold, and path planning related parameters.

[0039] S102. Based on the radio map and communication quality threshold, calculate the probability of communication interruption at each spatial location within the planning area corresponding to the radio map, and determine the communication interruption probability map.

[0040] In some embodiments of this application, the probability of communication interruption at each spatial location within the planning area is calculated based on radio maps and communication quality thresholds, providing communication environment information for subsequent high-communication-quality path cost modeling.

[0041] For example, suppose the planning area is discretized into a two-dimensional raster map, and any raster position is represented as:

[0042] Radio maps can be represented as follows Used to indicate position The average communication quality parameters at a given location can be received signal strength, signal-to-noise ratio (SNR), signal-to-interference-plus-noise ratio (SINR), or other metrics that characterize the quality of the wireless link. In a real-world wireless propagation environment, the UAV at its location... The instantaneous communication quality at a given location is not only related to the average communication quality in the radio map, but also affected by small-scale fading. Let the small-scale fading random variable be... Then the drone is in position The instantaneous communication quality at a point can be expressed as:

[0043] in, This function represents the instantaneous communication quality calculated under the combined effects of average communication quality and small-scale fading. A communication quality threshold is set as... When the drone is in position When the instantaneous communication quality at a location falls below a certain threshold, the location is considered to be at risk of communication interruption. Therefore, the location... The probability of communication interruption at a certain point is defined as:

[0044] in, The range of values ​​for is:

[0045] when The closer the value is to 0, the lower the probability of communication interruption at that location, and the more stable the communication quality; when... The closer the probability is to 1, the higher the probability of communication interruption at that location, and the worse the communication reliability. Based on this method, the communication interruption probability can be calculated for each grid location within the planning area, resulting in a communication interruption probability map.

[0046] Here, G represents the set of grid locations within the planning area. This communication interruption probability map can reflect the distribution of communication interruption risk when UAVs fly in different spatial areas.

[0047] Output the communication interruption probability map corresponding to the planned area, such as... Figure 2 As shown, this communication interruption probability map will serve as input to the subsequent communication quality path cost matrix construction module. It will be used together with spatial distance to constitute the path search cost, enabling the UAV to prioritize avoiding areas with a high probability of communication interruption during path planning.

[0048] S103. Based on the communication interruption probability map, the location of the UAV take-off and landing point, and the location of the mission pickup and delivery point, calculate the communication quality path cost matrix between the UAV take-off and landing point, the mission pickup point, and the mission delivery point.

[0049] In some embodiments of this application, spatial distance information is calculated based on the location of the UAV take-off and landing point and the location of the task pickup and delivery point; based on the communication interruption probability map, spatial distance information, preset distance cost weight, and preset communication interruption risk weight, the single-step path cost of feasible paths between any two key nodes in the UAV take-off and landing point, task pickup point, and task delivery point is calculated; wherein, key nodes are nodes among the UAV take-off and landing point, task pickup point, and task delivery point; based on the single-step path cost of feasible paths between any two key nodes, the feasible path corresponding to the minimum single-step path cost is selected as the communication quality path between any two key nodes, and the single-step path cost corresponding to the communication quality path is determined as the optimal path cost between any two key nodes; based on the optimal path cost between any two key nodes, the communication quality path cost matrix is ​​determined.

[0050] For example, based on the communication interruption probability map and spatial distance information, the path cost between the UAV take-off and landing point, the task pickup point and the task delivery point is calculated, providing basic path data for subsequent graph neural network task allocation and single UAV path planning.

[0051] Let the set of key nodes involved in multi-UAV path planning be:

[0052] in, This represents the set of drone take-off and landing points. This represents the set of pickup points for the task. Represents the set of delivery points for a task, for any two key nodes It is necessary to calculate the path cost between the two under communication quality constraints.

[0053] In the raster map of the planning area, for the current raster node g, let its set of neighboring raster nodes be . Adjacent grid nodes can include their four neighboring nodes (top, bottom, left, and right) or diagonally adjacent nodes, thus forming an eight-neighbor search. For the process of moving from grid node g to adjacent node g′, the single-step path cost is defined as:

[0054] in, This represents the spatial distance between two adjacent grid nodes. Indicates adjacent nodes The probability of communication interruption at the location, and These represent the distance cost weight and the communication interruption risk weight, respectively.

[0055] when When the distance is large, path planning tends to choose the shorter path; when... When the value is large, path planning tends to avoid areas with a high probability of communication interruption. This can be achieved by adjusting... and This allows for a trade-off between flight range and communication quality. In real-world scenarios, this trade-off can be dynamically adjusted based on actual needs.

[0056] For any starting point vi and ending point vj, an improved path search algorithm is used to search on the communication interruption probability graph. During the search, the cumulative cost of each candidate path is expressed as:

[0057] in, This represents a raster path from the starting point vi to the ending point vj. The target of the path search is:

[0058] That is, among all feasible paths, the path with the lowest overall cost is selected as the communication quality path from the starting point vi to the ending point vj.

[0059] To improve search efficiency, heuristic estimation terms can be introduced during the search process;

[0060] in, Indicates the current grid node To the target node The Euclidean distance. Then the node search evaluation function can be expressed as:

[0061] in, This represents the cumulative path cost from the starting point to the current node. This represents the heuristically estimated cost from the current node to the destination. This method can improve path search efficiency while maintaining communication quality constraints.

[0062] After completing the path search between any two key nodes, the optimal path cost is written into the matrix:

[0063] in, This represents the communication quality path cost from the starting point vi to the ending point vj. After calculating the cost pairwise for all critical nodes, the communication quality path cost matrix can be obtained:

[0064] Where K represents the total number of critical nodes. When there is no feasible path between two nodes, or the path cost exceeds a preset limit, the corresponding matrix element is set to an unreachable flag value for subsequent task feasibility assessment.

[0065] The output is a communication quality path cost matrix and the corresponding inter-node path information. This matrix is ​​no longer a simple Euclidean distance matrix, but a comprehensive path cost matrix that integrates communication interruption probability and spatial distance. It will serve as the basic input for the subsequent graph neural network task allocation module and single UAV path planning module.

[0066] S104. Based on the communication quality path cost matrix, mission payload weight, UAV take-off and landing point location, mission pickup and delivery point location, and mission set, construct graph structure data; and based on the graph structure data, determine the allocation relationship between each mission and the UAV in the mission set.

[0067] In some embodiments of this application, the graph structure data can be constructed based on the communication quality path cost matrix, mission payload weight, UAV take-off and landing point location, mission pickup and delivery point location, and mission set, as follows: S1041. Based on the communication quality path cost matrix, task payload weight, UAV take-off and landing point location, task pickup and delivery point location, and task set, determine the feature vector of each task node in the task set.

[0068] S1042. Based on the feature vector of each task node in the task set, calculate the communication risk coupling cost between any two task nodes.

[0069] S1043. Based on the communication risk coupling cost between each pair of task nodes, determine the neighbor nodes of each task node in the task set; and determine the neighbor set of each task node according to the neighbor nodes of each task node.

[0070] S1044. Calculate the normalized edge weights based on the communication risk coupling cost and the preset temperature coefficient.

[0071] S1045. Based on the feature vector of each task node in the task set, the neighbor set of each task node, and the normalized edge weight, determine the graph structure data.

[0072] In some embodiments of this application, a graph neural network is used to linearly map the feature vector of each task node in the task set to obtain an initial embedding representation of the feature vector of each task node in the task set; the initial embedding representation is then aggregated with communication risk weighted messages using the neighbor set and normalized edge weights of each task node to determine the neighbor messages of each task node in the task set; based on the initial embedding representation and the neighbor messages, the node embedding is updated to determine the final embedding representation of each task node in the task set; and the final embedding representation of each task node in the task set is determined as graph structure data.

[0073] In some embodiments of this application, determining the allocation relationship between each task and the UAV in the task set based on graph structure data can be achieved through S1046-S1049, as follows: S1046. Based on graph structure data, determine the task embedding values ​​of each of the multiple UAVs; and based on the task embedding values, determine the load state vectors of each of the multiple UAVs.

[0074] S1047. For any task to be assigned in the task set, calculate the communication matching characteristics between any task to be assigned in the task set and each of the multiple UAVs.

[0075] S1048. Based on graph structure data, load state vector and communication matching features, calculate the score of each of the multiple UAVs for any task to be assigned; and normalize the score to obtain the probability of each of the multiple UAVs for any task to be assigned.

[0076] S1049. Select the drone with the highest probability among multiple drones to determine the drone corresponding to any assigned task; based on the drone corresponding to any assigned task, determine the allocation relationship between each task and drone in the task set.

[0077] For example, based on the communication quality path cost matrix, task payload information, communication risk coupling relationship between tasks, and the current load status of the UAV, the allocation relationship between each task and the UAV is determined, providing task input for subsequent high communication quality path planning for a single UAV.

[0078] After constructing the communication quality path cost matrix, each pickup and delivery task is considered as a task node in a graph structure. For task i, its corresponding pickup point is pi, delivery point is di, and task payload weight is wi. Let the set of UAV take-off and landing points be... Where U represents the number of drones, and su represents the take-off / landing point or hangar node corresponding to drone u. For any two critical nodes a and b, let the elements in the aforementioned communication quality path cost matrix be... , representing the communication quality path cost from node a to node b. This cost is weighted by the flight distance and the probability of communication interruption, and can simultaneously reflect the spatial travel cost and communication reliability cost of the path between nodes.

[0079] For task i, first determine its associated reference take-off and landing point. Methods for determining the reference take-off and landing point include, but are not limited to, selecting the take-off and landing point with the minimum communication quality path cost to the task's pickup point, such as:

[0080] Based on the communication quality path cost matrix, the feature vector of task node i based on communication-distance constraints can be constructed:

[0081] in, This represents the normalized quality of communication path cost from the pickup point to the delivery point. This represents the normalized path cost between the drone's take-off / landing point and the pickup point. This represents the communication quality path cost between the delivery point and the take-off / landing point. This is the normalized value of the mission payload. This represents the average probability of communication interruption on the task-related path. This represents the maximum probability of communication interruption on the task-related path. Let these be the feature weights of the task nodes. And they satisfy:

[0082] To highlight the impact of communication quality on task node representation, the communication risk-related weights satisfy the following: Therefore, the features of a task node not only include the cost of the pickup and delivery path and the weight of the load, but also statistical information on the risk of communication interruption along the task path, enabling the subsequent graph neural network to perceive the reliability of communication during task execution.

[0083] To describe the degree of correlation between different tasks in terms of spatial location, communication quality path cost, and task connection relationship, the communication risk coupling cost between task i and task j is defined as:

[0084] in, This represents the communication risk coupling cost between task i and task j; Represents the normalized communication quality path cost between the pickup point of task i and the pickup point of task j. Represents the normalized communication quality path cost between the delivery point of task i and the delivery point of task j; The normalized communication quality path cost represents the distance between the delivery point of task i and the pickup point of task j, and is used to characterize the connection cost when the UAV continuously executes tasks i and j. This indicates the load difference between the two tasks; This represents the average probability of communication interruption along the path from the delivery point of task i to the pickup point of task j. ,⋯, This refers to the inter-task coupling weight, used to adjust the influence of different coupling factors on the degree of task correlation, and satisfying the following:

[0085] Since the path cost between the delivery point of the previous task and the pickup point of the next task is the primary factor when a drone performs two consecutive tasks, the path weight for task connection satisfies the following:

[0086] when The smaller the value, the more suitable it is for the same UAV to execute tasks i and j consecutively in terms of communication quality path cost, task connection cost, and payload difference; when The larger the value, the weaker the coupling of communication risks between the two tasks, making them unsuitable for consecutive scheduling within the same UAV path.

[0087] Based on the communication risk coupling cost mentioned above Construct a task graph:

[0088] in, Represents a set of task nodes. Let $\mathbf{i}$ represent the set of edges between task nodes. For any task $i$, select the $K$ task nodes with the lowest communication risk coupling cost as neighbor nodes, forming the neighbor set:

[0089] That is, select The K smallest task nodes are selected as the neighbor nodes of task i. Furthermore, the normalized edge weights in the task graph are defined based on the coupling cost of inter-task communication risks:

[0090] in, This represents the message passing weight of task node j to task node i. This is a temperature coefficient used to adjust the smoothness of the edge weight distribution.

[0091] When the communication risk coupling cost between task i and task j When the edge weight is small, A larger value indicates that task j has a higher influence on the graph neural network information aggregation process of task i; when When the edge weight is large, A smaller value indicates that task j has a weaker impact on task i.

[0092] The task graph constructed in the above manner differs from that constructed solely based on Euclidean distance or simple node proximity relationships. Instead, it uses communication quality path cost, task connection cost, load difference, and communication interruption probability as the basis for generating edge weights between tasks.

[0093] Task node features The input graph neural network first obtains the initial embedding representation of the task nodes through linear mapping:

[0094] in, This represents the initial node embedding for task i. The input feature mapping weight matrix, This is the bias term. In the l-th layer of the graph neural network, task node i is biased based on its set of neighbor nodes. and border rights Perform risk-weighted message aggregation for communication purposes:

[0095] in, This represents the neighbor message received by task node i at layer l. Let represent the neighbor message transformation weight matrix at layer l.

[0096] Then, update the node embedding based on the task node's own representation and neighbor messages:

[0097] in, This represents the transformation weight matrix of the l-th layer's own nodes. This represents the bias term at level l. This represents a non-linear activation function. After L layers of risk-weighted message passing, the final embedded representation (i.e., graph structure data) of task node i is obtained:

[0098] This task embedding representation integrates the task's own pickup and delivery path cost, payload weight, and communication interruption risk statistics, as well as the communication quality path cost and task connection relationship between neighboring tasks. Compared to graph neural networks that perform equal-weighted average aggregation of neighboring task nodes, this invention achieves weighted aggregation for communication risk perception through edge weight aij, giving higher weights to neighboring tasks with lower communication quality path costs and stronger task connection relationships during the node embedding learning process.

[0099] During the task allocation process, let Tu be the set of tasks currently assigned to UAV u. Based on the embedding representations of the tasks assigned to UAV u, calculate the mean task embedding of UAV u:

[0100] When the drone has not yet been assigned a task, Set it to the zero vector. Further define the load state vector of the UAV u:

[0101] in, This represents the normalized value indicating the number of tasks currently assigned to drone u. This represents the normalized value of the total cost of the current path of drone u. This represents the normalized value of the estimated completion time of drone u. This represents the normalized value of the current estimated energy consumption of the drone u.

[0102] For the current task i to be assigned, construct the communication matching features between task i and drone u:

[0103] in, This represents the normalized communication quality path cost between the take-off and landing point of UAV u and the pickup point of mission i. The normalized communication quality path cost between the delivery point of task i and the take-off and landing point of drone u is represented. This represents the average probability of communication interruption along the path from the take-off and landing point of the drone u to the mission pickup point; This represents the average probability of communication interruption along the path from the delivery point back to the drone's take-off and landing point. This represents the normalized value of the task payload.

[0104] Embedded by task Drone load status Matching features between mission and drone communication Calculate the score assigned to drone u for task i:

[0105] in, The task embedding transformation weight matrix represents the weight matrix. This represents the weight matrix for the drone's load state transformation. Let q represent the weight matrix of the task-drone communication matching feature transformation, and let q represent the scoring vector. This indicates the rating bias.

[0106] The probability of task i being assigned to drone u is further obtained by using a normalization function:

[0107] During the training phase, the probability can be assigned. Sampling is performed to explore different task allocation schemes; during the actual reasoning phase, the drone with the highest allocation probability is selected as the drone to execute task i.

[0108] This leads to the following task allocation decision variables:

[0109] And ensure that each task is assigned to only one drone:

[0110] Using the task allocation method described above, whether task i is assigned to drone u depends not only on the communication risk embedding representation of the task itself, but also on the drone's current load state, current path cost, estimated completion time, estimated energy consumption, and the communication quality matching relationship between the task and the drone. Therefore, this module can simultaneously consider task relevance, communication quality path cost, drone load balancing, and path execution cost during the task allocation phase.

[0111] To train the aforementioned communication risk-weighted graph neural network and task allocation policy network, reinforcement learning can be used to optimize the task allocation policy. For a complete task allocation result, the training reward function is defined as:

[0112] in, Indicates the maximum mission completion time for a multi-UAV system; This represents the overall communication quality path cost of the UAV u. This represents the variance of the completion time for each drone mission; This represents the variance of the number of missions performed by each drone. Indicates the estimated energy consumption of the drone u; This indicates the number of idle drones without assigned tasks. The statistical value representing the communication interruption risk of multiple drone paths satisfies:

[0113] in, This represents the average probability of communication interruption along the path u of the drone. Indicates the maximum probability of communication interruption. This indicates the number of path segments where the probability of communication interruption exceeds a preset risk threshold. Weighting for communication risk statistics.

[0114] The training reward weights are determined by the following criteria: To emphasize task completion efficiency and communication quality assurance, the reward weights satisfy the following:

[0115] In other words, the maximum system completion time, the comprehensive cost of communication quality paths, and the risk of communication interruption have high weights in the training objectives, so that the task allocation strategy obtained from the training can prioritize the generation of multi-UAV task allocation results that take into account both task execution efficiency and communication reliability.

[0116] Upon completion of this module, the task set corresponding to each drone will be output:

[0117] This task set will serve as input to the subsequent high-quality communication path planning module for each UAV, which will be used to further generate the task access sequence and specific flight path for each UAV.

[0118] S105. Based on the allocation relationship between each task and the UAV, calculate the path planning result for each UAV; and based on the path planning result for each UAV, perform multiple UAV task and path optimization to determine multiple UAV task allocation results and multiple UAV path planning results.

[0119] In some embodiments of this application, the path planning results for each UAV include the task access order, the sequence of pickup and delivery nodes, the specific flight path between nodes, the comprehensive path cost, the estimated flight time, and the estimated energy consumption.

[0120] In some embodiments of this application, based on the allocation relationship between each task and the UAV, a task subset for each UAV is determined; based on the communication quality path cost matrix, the task subset for each UAV, and the pickup and delivery order constraints, an initial path for each UAV is generated; and the initial path for each UAV is optimized to obtain a candidate path for each UAV; the comprehensive path cost of the candidate path for each UAV is calculated, and the candidate path whose comprehensive path cost satisfies the pickup and delivery order constraints, the preset load constraints, and the energy constraints is selected as the latest path for each UAV; the latest path for each UAV is optimized for communication quality to determine the path planning result for each UAV.

[0121] In some embodiments of this application, under the condition that the path planning result of each UAV meets preset conditions, multiple UAV task transfers and task exchanges are performed to obtain multiple UAV task allocation results; based on the multiple UAV task allocation results, path optimization is performed on the path planning result of each UAV among the multiple UAVs to determine the multiple UAV path planning results; wherein, the preset conditions include at least one of the following: for each UAV among the multiple UAVs, the expected flight time of each UAV is greater than the expected flight time of the remaining UAVs among the multiple UAVs; the expected energy consumption of each UAV is greater than a preset energy consumption threshold; the comprehensive path cost of each UAV is greater than a preset path cost threshold; and the number of adjustments for each UAV reaches a preset maximum number of adjustments.

[0122] For example, after the task allocation results are determined, a task access path that meets the pickup and delivery order constraints, payload constraints, and communication quality requirements is generated for each UAV. Using the output task set as input, and combining it with the communication quality path cost matrix, the task execution order and corresponding flight path for each UAV are determined.

[0123] For any drone u, let its assigned task set be:

[0124] Task i includes a pickup point pi, a delivery point di, and a payload weight wi. The take-off and landing point of drone u is denoted as su. This module needs to determine the access order of drones within the task set Tu, so that the drones start from the take-off and landing points, complete the assigned pickup and delivery tasks sequentially, and finally return to the take-off and landing points.

[0125] Let the path node sequence of the drone u be:

[0126] in, Let be the k-th visited node in the drone's path. Node types include task pickup points and task delivery points. For any task i∈Tu, the path sequence must contain both a pickup point pi and a delivery point di, and the pickup point must be visited before the delivery point, i.e.:

[0127] Here, pos represents the access position of the node in the path sequence. This constraint is used to ensure the task execution logic of the drone picking up the goods first and then delivering them.

[0128] In the path evaluation process, the path cost is no longer calculated using simple Euclidean distance or flight time, but rather using elements from the communication quality path cost matrix. For a path sequence Ru, its comprehensive path cost can be expressed as:

[0129] in, Indicates from node To the node The path cost for communication quality between the two points. This cost already incorporates flight distance and communication interruption probability; therefore, the smaller the path cost, the better the overall performance of the path in terms of flight distance and communication reliability. Furthermore, the path cost or actual flight distance can be converted into flight time based on the UAV's flight speed. Let UAV u be in flight segment (…). , The flight distance on ) is If the flight speed is vu, then the flight time for the UAV to complete the path Ru can be expressed as:

[0130] Meanwhile, considering the impact of mission payload changes on UAV energy consumption, energy consumption can be estimated based on the UAV's payload status in different flight segments, and the path must meet the maximum energy constraint:

[0131] During path search, the order of mission access is optimized to reduce communication quality path cost, flight time, and energy consumption. The optimization objective can be expressed as:

[0132] in, , , These represent the weighting coefficients for communication quality path cost, flight time, and energy consumption, respectively. Depending on the actual needs, communication quality path cost can also be used as the primary objective, while flight time and energy consumption can be used as constraints.

[0133] To solve the aforementioned path planning problem, heuristic search, local search, tabu search, ant colony optimization, or other path optimization methods can be employed. Specifically, an initial path satisfying the pickup and delivery order constraint can be generated first based on the communication quality path cost matrix. Then, the path can be iteratively optimized by exchanging the task access order, adjusting the arrangement of local nodes, or reconnecting local paths. When the overall cost of a candidate path decreases and satisfies the pickup and delivery constraint, load constraint, and energy constraint, the candidate path is accepted as the new current path.

[0134] After determining the access order of key nodes, for any adjacent nodes in the path sequence ( , The system calls the optimal grid path between nodes stored in the aforementioned communication quality path cost matrix construction module, and sequentially concatenates the communication quality paths between adjacent nodes to obtain the complete flight path of the UAV u. The resulting flight path not only meets the mission access requirements, but also avoids areas with a high probability of communication interruption as much as possible during flight between nodes.

[0135] The system outputs high-quality communication path planning results for each UAV, including the mission access order, pickup and delivery node sequence, specific flight paths between nodes, overall path cost, estimated flight time, and estimated energy consumption. This result will serve as input to the subsequent multi-UAV path collaborative adjustment module, used to further determine whether collaborative correction is needed between the paths of multiple UAVs.

[0136] After each UAV generates a high-quality communication path individually, the task allocation and path planning results of multiple UAVs are evaluated and adjusted as a whole. This ensures that the multi-UAV system meets communication quality requirements while minimizing the overall task completion time and avoiding excessive workload or path costs for some UAVs.

[0137] After completing the high-communication-quality path planning for a single UAV, the path node sequence for each UAV can be obtained:

[0138] The corresponding total cost of the path, flight time, and energy consumption are expressed as follows: .in, This represents the communication quality path cost of the drone u. This indicates the estimated mission completion time for drone u. This indicates the expected energy consumption of the drone u.

[0139] To evaluate the overall planning results of a multi-UAV system, the maximum system completion time is defined as:

[0140] The total system cost is defined as:

[0141] in, This indicates a penalty for unbalanced task load. , , , This is the weighting factor. The task load imbalance penalty can be calculated based on the variance of the number of tasks or flight time of each drone, in order to avoid excessive concentration of tasks on a few drones.

[0142] When a drone's mission completion time is significantly longer than that of other drones, or its path cost, communication interruption risk, or energy consumption exceeds preset limits, the current mission allocation results will be adjusted collaboratively. Collaborative adjustment methods include two categories: mission transfer and mission exchange.

[0143] Task transfer refers to selecting a task from drones with high load or high path cost, transferring it to a drone with low load or low path cost, and recalculating the path planning results for the relevant drones. Let task i originally belong to drone u. After transferring it to drone v, the task set is updated as follows:

[0144]

[0145] Task swapping refers to exchanging parts of a task between two drones to reduce the system's maximum completion time or overall path cost. Let task i belong to drone u and task j belong to drone v. After the swap, the task set is updated as follows:

[0146]

[0147] For each candidate adjustment scheme, recalculate the high communication quality path for the affected UAVs and update the corresponding... If the adjusted total system cost J′ is less than the original total system cost J, and the delivery constraints, energy constraints, and communication quality requirements are met, then the adjustment plan is accepted; otherwise, the original planning result is retained.

[0148] The above-described collaborative adjustment process can be executed iteratively until any of the following conditions are met: 1. The maximum system completion time will no longer decrease; 2. The overall system path cost will no longer decrease; 3. The energy consumption of all drones meets the constraints; 4. Reach the preset maximum number of adjustments.

[0149] Output the multi-UAV task allocation results and multi-UAV path planning results after collaborative adjustment, such as... Figure 3 As shown in the figure. Compared with the path planning results of a single UAV, this result can further balance the task load among different UAVs at the system level, reduce the maximum completion time of the multi-UAV system, and reduce the problem of excessive communication quality path cost caused by unreasonable local task allocation.

[0150] Understandably, based on radio maps, a communication interruption probability map is constructed by combining communication quality thresholds and impacts. This probability map is then combined with flight distance to determine the communication quality path cost matrix. Graph neural networks are used to learn the correlation between multi-UAV mission nodes, path costs, and communication reliability, thereby generating a multi-UAV path planning scheme that balances mission execution efficiency and communication quality assurance. This reduces the risk of communication interruption during UAV flight and improves the mission execution reliability and flight safety of multi-UAV systems.

[0151] In some embodiments of this application, the planning area is discretized into a two-dimensional grid map. The system includes multiple drones, multiple drone take-off and landing points, and several pickup and delivery tasks. Each task includes a pickup point, a delivery point, and the weight of the goods. The drones depart from their corresponding take-off and landing points, complete their assigned tasks, and return to their respective take-off and landing points.

[0152] First, a radio map of the planned area is read, and a communication quality threshold is set. Considering the impact of small-scale fading, the probability that the communication quality at each grid location will fall below the threshold is calculated, resulting in a communication interruption probability map. The values ​​in this probability map are between 0 and 1, representing the risk of communication interruption when the UAV passes through the corresponding location.

[0153] Then, a path search cost is constructed based on the communication interruption probability map and the distance between grid cells. For adjacent grid nodes, their movement cost is set as a weighted sum of the distance cost and the communication interruption probability cost. An improved path search algorithm is used to calculate the optimal path cost between the UAV take-off and landing point, the task pickup point, and the task delivery point, and a communication quality path cost matrix is ​​generated. This matrix is ​​used to represent the comprehensive communication cost between key nodes under communication quality constraints.

[0154] Next, each pickup and delivery task is treated as a node in the task graph, and the task graph is constructed based on the task pickup and delivery cost, the path cost from the take-off and landing point to the task point, the weight of the task cargo, and the path cost relationship between tasks. This task graph is then input into a graph neural network to obtain the embedded representation of each task node. Subsequently, combining the number of tasks currently assigned to each UAV and the task embedding state, the allocation relationship between tasks and UAVs is determined through a policy network.

[0155] After task allocation, path planning is performed for each drone. For the set of tasks assigned to a drone, a task access order that satisfies the "pick up first, deliver later" constraint is generated, and the comprehensive cost of candidate paths is calculated based on the communication quality path cost matrix, thereby selecting the flight path with higher communication quality and lower comprehensive cost.

[0156] Finally, the path results for multiple drones are evaluated as a whole. When there are situations where drones are overloaded with tasks, path costs are too high, or the maximum system completion time is too long, the task allocation results are adjusted through task transfer or task exchange, and the relevant drone paths are recalculated. If the overall system cost is reduced after adjustment, the adjustment is accepted. The final output includes the task allocation results for each drone, the order of pickup and delivery, the specific flight path, the comprehensive path cost, the estimated flight time, energy consumption, and the communication interruption risk assessment results.

[0157] The above implementation method can comprehensively consider flight distance and communication interruption risk during the execution of multiple UAV missions, so that the planned path can avoid areas with poor communication quality as much as possible, thereby improving the communication reliability and system security of UAVs during flight while ensuring mission execution efficiency.

[0158] This application has the following beneficial effects: A communication outage probability map is constructed based on known radio maps. Considering the impact of small-scale fading, the risk of communication outage at different locations within the planning area is quantified into probability values ​​between 0 and 1. Compared to methods that only use average communication quality or binary judgments of communicability / non-communicability, this approach more meticulously reflects changes in communication quality under complex wireless environments, providing a more accurate basis for UAV path planning regarding communication risks.

[0159] By weighting the probability of communication interruption with flight distance to construct a communication quality path cost model, the UAV's path planning process considers not only path length but also the communication reliability of the areas traversed by the path. Therefore, the planned path can proactively avoid areas with a high probability of communication interruption, reducing the risks of link interruptions, abnormal status feedback, and loss of control commands during UAV flight, and improving the communication continuity and security of UAV mission execution.

[0160] By using a communication quality path cost matrix to uniformly represent the comprehensive travel cost between the UAV take-off and landing point, the mission pickup point, and the mission delivery point, subsequent mission allocation and path planning are based on the same communication quality constraints. Compared to traditional planning methods that rely solely on geometric distance matrices, this approach ensures that both mission allocation and path planning results are constrained by communication quality factors, avoiding the problem of reasonable mission allocation but poor communication quality along the actual flight path.

[0161] By using graph neural networks to learn features from task nodes, payload information, and the relationship between communication quality paths and costs between tasks, it is possible to uncover the correlations between multiple tasks in terms of spatial location, communication risk, and execution cost. Compared with manual rules or traditional heuristic allocation methods, this can improve the rationality of multi-UAV task allocation and make the task allocation results more conducive to the subsequent generation of high-communication-quality paths.

[0162] By coordinating and adjusting path planning results at the multi-drone level, the maximum system completion time and overall path cost can be reduced through task transfer or task exchange, thereby avoiding excessive concentration of tasks on a few drones and improving the efficiency of multiple drones.

[0163] In summary, this application can generate path planning schemes that balance flight efficiency, task load balancing, and communication quality assurance when multiple UAVs are performing tasks such as logistics delivery, inspection and monitoring, and emergency transportation. This effectively reduces the risk of UAVs losing contact or flying blindly after entering areas with poor communication quality, and improves the reliability, collaborative stability, and flight safety of multi-UAV systems in complex wireless environments. Simultaneously, it enhances the system's task load balancing and overall execution efficiency.

[0164] Based on the above embodiments of the multi-UAV path planning method based on graph neural networks, this application also provides a multi-UAV path planning system based on graph neural networks, such as... Figure 4 As shown, Figure 4 This application provides a schematic diagram of the structure of a multi-UAV path planning system based on a graph neural network. The system includes: an acquisition module 401, a calculation module 402, and a determination module 403. The acquisition module 401 is used to acquire various basic information for path planning of multiple UAVs; and initialize the basic information to obtain initialization information; wherein, the initialization information includes UAV set, UAV take-off and landing point location, task set, task pickup and delivery point location, task payload weight, UAV flight performance parameters, radio map, communication quality threshold and path planning related parameters; The calculation module 402 is used to calculate the probability of communication interruption at each spatial location within the planning area corresponding to the radio map based on the radio map and the communication quality threshold, and to determine the communication interruption probability map; based on the communication interruption probability map, the location of the UAV take-off and landing point and the location of the task pickup and delivery point, to calculate the communication quality path cost matrix between the UAV take-off and landing point, the task pickup point and the task delivery point. The determining module 403 is used to construct graph structure data based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set; and based on the graph structure data, determine the allocation relationship between each task and the UAV in the task set; based on the allocation relationship between each task and the UAV, calculate the path planning result for each UAV; and based on the path planning result for each UAV, perform multiple UAV task and path optimization to determine multiple UAV task allocation results and multiple UAV path planning results.

[0165] Based on the above embodiments of the multi-UAV path planning method based on graph neural networks, this application also provides a multi-UAV path planning device based on graph neural networks, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of the structure of a multi-UAV path planning device based on a graph neural network, provided in an embodiment of this application. The multi-UAV path planning device 5 based on a graph neural network includes a processor 501 and a memory 502. The memory 502 is used to store a computer program; the processor 501 is used to call and run the computer program from the memory to execute a multi-UAV path planning method based on a graph neural network as described in the above embodiment.

[0166] In the embodiments of this application, the processor 501 described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above processor function can also be other types, and the embodiments of this application do not specifically limit it.

[0167] This application provides a computer-readable storage medium storing a computer program for implementing, when executed by a processor, a multi-UAV path planning method based on a graph neural network as described in any of the above embodiments.

[0168] For example, the program instructions corresponding to the multi-UAV path planning method based on graph neural networks in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to the multi-UAV path planning method based on graph neural networks in the storage media are read or executed by an electronic device, the multi-UAV path planning method based on graph neural networks as described in any of the above embodiments can be implemented.

[0169] Furthermore, in the embodiments of this application, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0170] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0171] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the embodiments in this application are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, these will not be repeated here.

[0172] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0173] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.

[0174] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0175] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0176] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0177] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0178] The above description is merely an embodiment of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A multi-UAV path planning method based on graph neural networks, characterized in that, include: Acquire various basic information for path planning of multiple drones; and initialize the basic information to obtain initialization information; wherein, the initialization information includes drone set, drone take-off and landing point location, task set, task pickup and delivery point location, task payload weight, drone flight performance parameters, radio map, communication quality threshold and path planning related parameters; Based on the radio map and the communication quality threshold, the probability of communication interruption at each spatial location within the planning area corresponding to the radio map is calculated, and a communication interruption probability map is determined. Based on the communication interruption probability map, the location of the UAV take-off and landing point, and the location of the task pickup and delivery point, calculate the communication quality path cost matrix between the UAV take-off and landing point, the task pickup point, and the task delivery point. Based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set, a graph structure data is constructed; and based on the graph structure data, the allocation relationship between each task and the UAV in the task set is determined. Based on the allocation relationship between each task and the UAV, the path planning result for each UAV is calculated; and based on the path planning result for each UAV, multiple UAV tasks and paths are optimized to determine multiple UAV task allocation results and multiple UAV path planning results.

2. The method according to claim 1, characterized in that, The step of calculating the communication quality path cost matrix between the UAV take-off and landing point, the UAV take-off and landing point, and the task pickup and delivery point based on the communication interruption probability map, the UAV take-off and landing point location, and the task pickup and delivery point includes: Based on the location of the UAV take-off and landing point and the location of the task pickup and delivery point, calculate the spatial distance information; Based on the communication interruption probability map, the spatial distance information, the preset distance cost weight, and the preset communication interruption risk weight, calculate the single-step path cost of feasible paths between each pair of key nodes in the UAV take-off and landing point, the mission pickup point, and the mission delivery point; wherein, the key nodes are the nodes in the UAV take-off and landing point, the mission pickup point, and the mission delivery point. Based on the single-step path cost of the feasible paths between the two key nodes, the feasible path corresponding to the minimum single-step path cost is selected as the communication quality path between the two key nodes, and the single-step path cost corresponding to the communication quality path is determined as the optimal path cost between the two key nodes. The communication quality path cost matrix is ​​determined based on the optimal path cost between each pair of key nodes.

3. The method according to claim 1, characterized in that, The graph structure data is constructed based on the communication quality path cost matrix, the mission payload weight, the UAV take-off and landing point location, the mission pickup and delivery point location, and the mission set, including: Based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set, determine the feature vector of each task node in the task set; Based on the feature vector of each task node in the task set, calculate the communication risk coupling cost between any two task nodes. Based on the communication risk coupling cost between each pair of task nodes, the neighbor nodes of each task node in the task set are determined; and based on the neighbor nodes of each task node, the neighbor set of each task node is determined. Based on the communication risk coupling cost and the preset temperature coefficient, the normalized edge weights are calculated; The graph structure data is determined based on the feature vector of each task node in the task set, the neighbor set of each task node, and the normalized edge weight.

4. The method according to claim 3, characterized in that, The process of determining the graph structure data based on the feature vector of each task node in the task set, the neighbor set of each task node, and the normalized edge weights includes: By using a graph neural network, the feature vector of each task node in the task set is linearly mapped to obtain the initial embedding representation of the feature vector of each task node in the task set. By using the neighbor set of each task node and the normalized edge weight, the initial embedding representation is subjected to communication risk-weighted message aggregation to determine the neighbor messages of each task node in the task set. Based on the initial embedding representation and the neighbor messages, the node embedding is updated to determine the final embedding representation of each task node in the task set; The final embedded representation of each task node in the task set is determined as the graph structure data.

5. The method according to claim 1, characterized in that, The step of determining the allocation relationship between each task and the UAV in the task set based on the graph structure data includes: Based on the graph structure data, the task embedding values ​​of each of the multiple drones are determined; and based on the task embedding values, the load state vectors of each of the multiple drones are determined. For any task to be assigned in the task set, calculate the communication matching features between any task to be assigned in the task set and each of the multiple UAVs; Based on the graph structure data, the load state vector, and the communication matching features, the scores for each of the multiple drones to be assigned a given task are calculated; and the scores are normalized to obtain the probability of assigning each of the multiple drones to the given task. The drone with the highest probability among the plurality of drones is selected as the drone to perform any of the assigned tasks. Based on the drone corresponding to any one of the tasks to be assigned, determine the allocation relationship between each task and the drone in the task set.

6. The method according to claim 1, characterized in that, The calculation of the path planning result for each UAV based on the allocation relationship between each task and the UAV includes: Based on the allocation relationship between each task and the UAV, determine the task subset for each UAV; Based on the communication quality path cost matrix, the task subset of each UAV, and the pickup and delivery order constraints, an initial path for each UAV is generated; and the initial path of each UAV is optimized to obtain a candidate path for each UAV. Calculate the comprehensive path cost of the candidate paths for each UAV, and select the candidate path whose comprehensive path cost satisfies the pickup and delivery order constraint, the preset load constraint, and the energy constraint as the latest path for each UAV. The latest path of each UAV is optimized for communication quality to determine the path planning result of each UAV.

7. The method according to claim 1, characterized in that, The path planning results for each UAV include the mission access order, pickup and delivery node sequence, specific flight path between nodes, comprehensive path cost, estimated flight time, and estimated energy consumption. The step of optimizing multiple drone tasks and paths based on the path planning results of each drone, and determining multiple drone task allocation results and multiple drone path planning results, includes: Under the condition that the path planning result of each UAV meets the preset conditions, multiple UAV task transfers and task exchanges are performed to obtain the multiple UAV task allocation results. Based on the task allocation results of the multiple drones, the path planning results of each of the multiple drones are optimized to determine the path planning results of the multiple drones. The preset conditions include at least one of the following: For each of the multiple drones, the estimated flight time of each drone is greater than the estimated flight time of the remaining drones in the multiple drones. The estimated energy consumption of each drone is greater than the preset energy consumption threshold; The total path cost for each drone exceeds a preset path cost threshold. Each drone has been adjusted up to the preset maximum number of adjustments.

8. A multi-UAV path planning system based on graph neural networks, characterized in that, include: The module consists of an acquisition module, a calculation module, and a determination module; among which, The acquisition module is used to acquire various basic information for path planning of multiple UAVs; and initialize the basic information to obtain initialization information; wherein, the initialization information includes UAV set, UAV take-off and landing point location, task set, task pickup and delivery point location, task payload weight, UAV flight performance parameters, radio map, communication quality threshold and path planning related parameters; The calculation module is used to calculate the probability of communication interruption at each spatial location within the planning area corresponding to the radio map based on the radio map and the communication quality threshold, and to determine the communication interruption probability map; based on the communication interruption probability map, the location of the UAV take-off and landing point and the location of the task pickup and delivery point, to calculate the communication quality path cost matrix between the UAV take-off and landing point, the task pickup point and the task delivery point. The determining module is used to construct graph structure data based on the communication quality path cost matrix, the task payload weight, the UAV take-off and landing point location, the task pickup and delivery point location, and the task set; and based on the graph structure data, determine the allocation relationship between each task and the UAV in the task set; based on the allocation relationship between each task and the UAV, calculate the path planning result for each UAV; and based on the path planning result for each UAV, perform multiple UAV task and path optimization to determine multiple UAV task allocation results and multiple UAV path planning results.

9. A multi-UAV path planning device based on graph neural networks, characterized in that, include: Processor and memory, of which, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions for causing a processor to execute, thereby implementing the method of any one of claims 1 to 7.

Images