A collaborative scheduling system, method, device, and storage medium for unmanned aerial vehicle (UAV) swarms.

Abstract

This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, specifically disclosing a UAV swarm collaborative scheduling system, method, device, and storage medium. By constructing a collaborative scheduling system for UAV swarms that integrates a perception layer, communication layer, scheduling decision layer, execution layer, and simulation verification layer, this invention solves the problems in existing technologies, such as the decreased convergence speed and increased iteration time of multi-agent reinforcement learning algorithms due to the exponential increase in input dimensions with swarm size; bandwidth limitations and susceptibility to electromagnetic interference in complex low-altitude environments leading to increased communication latency and data loss in radio frequency communication; and the lack of flexibility in fixed formation structures and the lack of dynamic adaptation in resource allocation, resulting in insufficient swarm performance. This invention improves the efficiency of meeting the real-time control requirements and collaborative response of large-scale UAV swarms.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, and in particular to a UAV swarm collaborative scheduling system, method, device and storage medium. Background Technology

[0002] With the deep integration of drone technology and artificial intelligence, the demand for low-altitude drone swarms in logistics delivery, emergency rescue, and environmental monitoring continues to grow rapidly. Swarm collaborative scheduling technology has become a key factor restricting its large-scale application. However, existing technologies still have significant bottlenecks in algorithm architecture, communication mechanisms, and formation management, making it difficult to support the efficient collaborative operation of large-scale swarms.

[0003] Currently, traditional multi-agent reinforcement learning algorithms are the mainstream technology for UAV swarm control. However, their critic network needs to integrate the state feature information of all agents. As the swarm size increases, the input dimension grows exponentially, directly leading to a significant decrease in the algorithm's learning convergence speed and a significant increase in iteration time, making it difficult to meet the real-time control requirements of large-scale swarms. At the same time, traditional algorithms do not fully incorporate the characteristics of the formation topology into the network design, resulting in high overall model complexity and posing significant challenges to practical engineering deployment.

[0004] In terms of swarm communication, existing technologies mostly employ radio frequency (RF) communication, which suffers from inherent drawbacks such as limited bandwidth and susceptibility to electromagnetic interference. In complex low-altitude environments, such as those with obstructed urban buildings and dense electromagnetic signals, communication latency increases significantly and data packet loss becomes a major issue. This makes it impossible to support real-time information exchange for large-scale UAV swarms, severely restricting the overall collaborative response efficiency and mission execution reliability of the swarm.

[0005] Furthermore, existing drone swarm structures lack flexibility, often employing fixed formations that cannot be freely expanded in terms of swarm level and size based on mission type. Simultaneously, resource allocation mechanisms lack dynamic adaptability, making them prone to resource waste or mission failures during diverse mission execution by heterogeneous drone swarms, thus hindering the full realization of overall swarm effectiveness.

[0006] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a drone swarm collaborative scheduling system, method, device, and storage medium.

[0008] Firstly, the present invention provides a collaborative scheduling system for unmanned aerial vehicle (UAV) swarms, the technical solution of which is as follows: It includes: perception layer, communication layer, scheduling decision layer, execution layer, and simulation verification layer; The perception layer is used to perform weighted fusion of the individual state data collected by the UAV onboard sensor group and the environmental state data collected by the environmental monitoring sensor group, and output the fused standardized state data. The communication layer is used to construct a two-level self-organizing communication architecture based on free space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to a distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer is used to transmit the standardized state data via the star topology and the fully connected communication link. The scheduling decision layer is used to construct a multi-level fractal formation structure with self-similar characteristics, receive the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generate collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure. The execution layer is used to control the UAV swarm to perform formation actions and mission actions according to the collaborative decision-making instructions; The simulation verification layer is used to perform scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

[0009] The beneficial effects of the UAV swarm collaborative scheduling system of the present invention are as follows: The system of this invention constructs a collaborative scheduling system for UAV swarms, which integrates a perception layer, a communication layer, a scheduling decision layer, an execution layer, and a simulation verification layer. This system solves the problems in existing technologies, such as the decrease in convergence speed and increase in iteration time caused by the exponential growth of the input dimension with the swarm size in multi-agent reinforcement learning algorithms, the bandwidth limitation of radio frequency communication in complex low-altitude environments and the susceptibility to electromagnetic interference causing increased communication latency and data packet loss, and the lack of flexibility in fixed formation structures and the lack of dynamic adaptation capabilities in resource allocation, which makes it difficult to fully realize the swarm efficiency. This system improves the real-time control requirements and collaborative response efficiency of large-scale UAV swarms.

[0010] Based on the above scheme, the UAV swarm collaborative scheduling system of the present invention can be further improved as follows.

[0011] In one alternative approach, the perception layer includes an onboard sensor array for the UAV, an environmental monitoring sensor array, and a state fusion unit. The UAV's onboard sensor group is used to collect the UAV's three-dimensional coordinates, flight speed, heading angle, pitch angle, remaining battery power, and onboard CPU computing power as the individual unit's status data; The environmental monitoring sensor group is used to collect the three-dimensional coordinates of obstacles, atmospheric visibility, real-time wind speed, and coordinates of the mission target area as the environmental status data. The state fusion unit is used to perform weighted fusion of the individual state data and the environmental state data using a weighted average algorithm to obtain the fused standardized state data.

[0012] The beneficial effects of adopting the above-mentioned optional approach are as follows: the UAV's three-dimensional coordinates, flight speed, heading angle, pitch angle, remaining battery power, and onboard CPU computing power are further collected by the UAV's onboard sensor group as individual state data, and the three-dimensional coordinates of obstacles, atmospheric visibility, real-time wind speed, and coordinates of the mission target area are collected by the environmental monitoring sensor group as environmental state data. The above data are then weighted and fused by the state fusion unit using a weighted average algorithm, which solves the problem of insufficient information dimensions from a single data source leading to one-sided state representation, and improves the comprehensiveness of standardized state data and the reliability of scheduling decision input.

[0013] In one alternative approach, the two-level self-organizing communication architecture includes an intra-group communication sublayer and an inter-group communication sublayer. The intra-group communication sublayer is used to divide communication groups according to a distance threshold and construct the star topology with the intra-group virtual center as the core. The inter-group communication sublayer is used to determine the node with the smallest Euclidean distance from the virtual center and the remaining power is higher than a preset percentage of the total power as the group leader, and the group leaders construct the fully connected communication link.

[0014] The beneficial effects of adopting the above optional method are as follows: by further dividing the communication groups according to the distance threshold through the intra-group communication sub-layer and constructing a star topology structure with the virtual center of the group as the core, the node with the smallest Euclidean distance from the virtual center and the remaining power higher than the preset percentage of the total power is determined as the group leader through the inter-group communication sub-layer, and a fully connected communication link is constructed between the group leaders. This solves the link redundancy and energy consumption imbalance caused by the rigidity of the large-scale cluster communication topology, and improves the self-organizing ability of the communication architecture and the stability of the group leader node operation.

[0015] In one alternative approach, the virtual center is determined based on the arithmetic mean of the three-dimensional coordinates of each UAV within the group.

[0016] The advantages of adopting the above optional method are as follows: the virtual center is further determined by the arithmetic mean of the three-dimensional coordinates of each UAV in the group, which solves the problems of group leader election deviation and topology reconstruction inaccuracy caused by the ambiguity of the definition of the spatial reference point in the group, and improves the consistency of virtual center calculation and the rationality of communication group division.

[0017] In one alternative approach, the multi-level fractal formation structure is a Sierpinski fractal structure. The first-level formation of the Sierpinski fractal structure includes one leader drone and three follower drones. The leader drone is located at the center of the formation, and the three follower drones are evenly distributed around the leader drone with adjacent followers having a horizontal angle of 120 degrees. The second-level formation of the Sierpinski fractal structure consists of three first-level formations, and the leader of each first-level formation forms an equilateral triangle.

[0018] The advantages of adopting the above-mentioned optional approach are as follows: the multi-level fractal formation structure is further limited to the Sierpinski fractal structure. The first-level formation includes one leader drone and three follower drones. The leader drone is located at the center of the formation, and the three follower drones are evenly distributed around the leader drone with the horizontal angle between adjacent followers being 120 degrees. The second-level formation consists of three first-level formations, and the leader of each first-level formation forms an equilateral triangle. This solves the problem of insufficient task adaptability caused by the fixed formation size not being scalable, and improves the self-similarity characteristics of the formation structure and the flexibility of cluster size expansion.

[0019] In one alternative, the multi-level fractal formation structure is a Menger sponge fractal structure. The first-level formation of the Menger sponge fractal structure includes one leader drone and four follower drones. The four follower drones are evenly distributed around the leader drone and the horizontal angle between adjacent followers is 90 degrees. The second-level formation of the Menger sponge fractal structure consists of four first-level formations.

[0020] The beneficial effects of adopting the above-mentioned optional method are as follows: the multi-level fractal formation structure is further limited to a Menger sponge fractal structure. The first-level formation includes one leader drone and four follower drones. The four follower drones are evenly distributed with the leader drone as the center and the horizontal angle between adjacent followers is 90 degrees. The second-level formation consists of four first-level formations. This solves the problem of the single fractal mode being limited in its applicable scenarios and thus the problem of the formation type being monotonous. It also improves the diversity of fractal formation structure and its adaptability to different mission scenarios.

[0021] In one alternative approach, the simulation verification layer includes a scene modeling submodule, a UAV control submodule, a communication simulation submodule, and a result visualization submodule; The scene modeling submodule is used to construct a three-dimensional terrain scene and set the three-dimensional coordinates of obstacles and the coordinates of target points. The UAV control submodule is used to control the take-off, cruise, turning and landing of the UAV. The communication simulation submodule is used to simulate the two-level self-organizing communication architecture. The result visualization submodule is used to output the formation position error rate, obstacle avoidance speed change curve and formation reconstruction delay data corresponding to the scheduling algorithm.

[0022] The advantages of adopting the above optional approach are as follows: a three-dimensional terrain scene is further constructed through the scene modeling submodule and the three-dimensional coordinates of obstacles and the coordinates of target points are set; the take-off, cruise, turning and landing of the UAV are controlled through the UAV control submodule; the two-level self-organizing communication architecture is simulated through the communication simulation submodule; and the formation position error rate, obstacle avoidance speed change curve and formation reconstruction delay data corresponding to the scheduling algorithm are output through the result visualization submodule. This solves the problem of uncontrollable deployment risks caused by the lack of algorithm verification and improves the data support and simulation verification completeness of the scheduling algorithm optimization iteration.

[0023] Secondly, this invention provides a method for collaborative scheduling of unmanned aerial vehicle (UAV) swarms, employing the UAV swarm collaborative scheduling system provided by this invention. The technical solution of this method is as follows: The perception layer performs weighted fusion of the individual state data collected by the UAV's onboard sensor group and the environmental state data collected by the environmental monitoring sensor group, and outputs the fused standardized state data. The communication layer constructs a two-level self-organizing communication architecture based on free-space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to a distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer is used to transmit the standardized state data via the star topology and the fully connected communication link. The scheduling decision layer constructs a multi-level fractal formation structure with self-similar characteristics, receives the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generates collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure. The execution layer controls the drone swarm to perform formation actions and mission actions according to the collaborative decision-making instructions; The simulation verification layer performs scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

[0024] The beneficial effects of the UAV swarm collaborative scheduling method of the present invention are as follows: The method of this invention solves the problems in the prior art where multi-agent reinforcement learning algorithms suffer from decreased convergence speed and increased iteration time due to the exponential growth of input dimension with cluster size, bandwidth limitations of radio frequency communication in complex low-altitude environments and susceptibility to electromagnetic interference causing increased communication latency and data packet loss, and the lack of flexibility of fixed formation structure and lack of dynamic adaptation capability of resource allocation, which makes it difficult to fully realize the cluster efficiency. It improves the real-time control requirements and collaborative response efficiency of large-scale UAV clusters.

[0025] Thirdly, the technical solution of an electronic device according to the present invention is as follows: It includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the UAV swarm cooperative scheduling method of the present invention.

[0026] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the UAV swarm cooperative scheduling method of the present invention.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0028] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an embodiment of the UAV swarm collaborative scheduling system of the present invention; Figure 2 This is a schematic diagram of the overall architecture; Figure 3 This is a schematic diagram of the overall process; Figure 4 This is a flowchart illustrating an embodiment of a drone swarm collaborative scheduling method according to the present invention. Figure 5 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0029] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0030] Figure 1 A schematic diagram of an embodiment of a drone swarm collaborative scheduling system 100 provided by the present invention is shown. Figure 1 As shown, the UAV swarm collaborative scheduling system 100 includes: a perception layer 101, a communication layer 102, a scheduling decision layer 103, an execution layer 104, and a simulation verification layer 105.

[0031] The perception layer 101 is used to perform weighted fusion of the individual state data collected by the UAV onboard sensor group and the environmental state data collected by the environmental monitoring sensor group, and output the fused standardized state data.

[0032] The UAV onboard sensor group refers to a collection of various sensors installed on the UAV body to collect information about the UAV's own operational status. For example, on the hexacopter UAV A performing an environmental monitoring mission on January 1, 2026, the UAV onboard sensor group includes a global navigation satellite system sensor, a three-axis velocity sensor, an attitude sensor, a battery monitoring chip, and a computing power monitoring chip, which respectively collect the UAV's three-dimensional coordinates (31.23°, 121.47°, 50m), flight speed of 10m / s, heading angle of 45°, pitch angle of 5°, remaining battery power of 85%, and onboard CPU computing power of 2.1 TFLOPS. Individual status data refers to data collected by the UAV onboard sensor group that reflects the operational status of a single UAV. For example, the specific individual status data collected by UAV A are three-dimensional coordinates (31.23°, 121.47°, 50m), flight speed of 10m / s, heading angle of 45°, pitch angle of 5°, remaining battery power of 85%, and onboard CPU computing power of 2.1 TFLOPS.

[0033] The environmental monitoring sensor group refers to a collection of various sensors used to perceive the low-altitude environment in which the UAV operates. For example, UAV A is equipped with a lidar, visibility meter, anemometer, and target positioning sensor, which collect the three-dimensional coordinates of obstacles (31.24°, 121.48°, 60m), atmospheric visibility of 8km, real-time wind speed of 6m / s, and the coordinates of the target area (31.25°, 121.49°, 30m), respectively. Environmental state data refers to data collected by the environmental monitoring sensor group that reflects the external low-altitude environmental state of the UAV. For example, the environmental state data collected by UAV A specifically includes the three-dimensional coordinates of obstacles (31.24°, 121.48°, 60m), atmospheric visibility of 8km, real-time wind speed of 6m / s, and the coordinates of the target area (31.25°, 121.49°, 30m).

[0034] Standardized state data refers to dimensionless state data with a uniform range obtained by weighted fusion of individual unit state data and environmental state data. For example, the individual unit state data and environmental state data of UAV A are input into the state fusion unit. The individual unit state fusion value is calculated using a weight of 0.3 for remaining battery power, 0.3 for CPU computing power, and 0.4 for flight status. The environmental state fusion value is calculated using a weight of 0.5 for obstacle distribution, 0.3 for visibility, and 0.2 for wind speed. The final output standardized state data is a set of values ​​in the range of 0 to 1.

[0035] The communication layer 102 is used to construct a two-level self-organizing communication architecture based on free space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to a distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer 102 is used to transmit the standardized state data via the star topology and the fully connected communication link.

[0036] Free-space optical communication refers to a communication method that uses the atmosphere as a transmission medium and laser or visible light as a carrier for wireless information transmission. For example, the cluster of UAVs A uses a free-space optical communication transceiver module with a wavelength of λ=550nm for intra- and inter-cluster information exchange, and the communication link is not affected by radio frequency electromagnetic interference. A two-level self-organizing communication architecture refers to an architecture that divides the UAV cluster communication network into two levels: intra-cluster communication and inter-cluster communication, with the network topology dynamically adaptive. For example, in the cluster of UAVs A, adjacent UAVs with a distance less than a certain threshold are grouped into the same communication group to form a star topology for intra-group communication, and the group leaders of each communication group then establish fully connected communication links for inter-group communication.

[0037] The distance threshold refers to the critical distance between drones used to divide communication groups. For example, if drone A has a maximum flight speed of v = 20 m / s, a communication delay tolerance threshold of t = 0.1 s, and a distance threshold r0 calculated using the formula r0 = 5v × t, which is 10 m, then drones with a distance of 10 m or less are grouped into the same communication group. A star topology refers to a network topology within a communication group where a virtual center within the group serves as the core node, and other drones within the group directly establish communication links with this core node. For example, five drones within a communication group may form a star topology with the virtual center as the core, and each drone sends its status data to the drone acting as the virtual center node.

[0038] In this context, the "group leader" refers to a drone node within the communication group that is designated according to preset rules to handle inter-group communication forwarding. For example, within a communication group, drone A has a distance of 0.8m from the group's virtual center and 75% remaining battery power, while drone B has a distance of 1.2m and 62% remaining battery power. The system designates drone A, which has the shortest distance and a remaining battery power exceeding 60% of its total battery power, as the group leader. A fully connected communication link refers to a topology where each group leader establishes a bidirectional communication connection with every other group leader. For instance, in a cluster with three communication groups, bidirectional free-space optical communication links are established between group leader A and group leader B, group leader A and group leader C, and group leader B and group leader C, enabling global information synchronization across the cluster.

[0039] The intra-group communication sublayer is also used to control the communication link based on the signal attenuation coefficient of free-space optical communication, and the signal attenuation coefficient satisfies the formula... in, This represents the signal attenuation coefficient for free-space optical communication within the group. Atmospheric visibility, For communication wavelength, This refers to the atmospheric attenuation coefficient; for example, when the atmospheric visibility V = 10 km, the atmospheric attenuation coefficient p = 1.3, and the wavelength λ = 550 nm, the signal attenuation coefficient can be calculated using the formula. .

[0040] The three-dimensional coordinates of the virtual center within the group satisfy the formula: in, The three-dimensional coordinates of the virtual center within the communication group. , , Let i be the three-dimensional coordinates of the i-th UAV in the group. The number of drones in the group; for example, the three-dimensional coordinates of the five drones in the group are (10m, 20m, 5m), (12m, 21m, 5m), (11m, 19m, 5m), (13m, 20m, 5m), and (12m, 22m, 5m), respectively. The virtual center coordinates of the group are calculated according to the formula as (11.6m, 20.4m, 5m).

[0041] The scheduling decision layer 103 is used to construct a multi-level fractal formation structure with self-similar characteristics, receive the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generate collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure.

[0042] Among them, the multi-level fractal formation structure refers to a fractal formation organization form that has self-similar properties and can be gradually combined from lower-level formation units to form higher-level formations. For example, the cluster where UAV A is located adopts the Sierpinski fractal structure. The first-level formation consists of a leader UAV L1 and three follower UAVs F1, F2, and F3, which are distributed in an equilateral triangle. After the three first-level formations are combined, each leader forms a larger equilateral triangle to constitute the second-level formation.

[0043] For the Sierpinski fractal structure, the radius of the expected location region satisfies the formula: ; The radius of the expected location area. For flight speed, The unit of time; the safe distance satisfies the formula: ; The safe distance between drones; for example, taking a flight speed v = 10 m / s and a unit time t = 1 s, then the radius of the expected location area. safe distance The standard distance between leaders and followers is .

[0044] In the case of the second-level formation of the Sierpinski fractal structure, the leaders of each first-level formation form an equilateral triangle, and the distance from the centroid of the triangle to each leader is... For a Menger sponge fractal structure, the standard distance between the leader and the follower is... .

[0045] The scheduling algorithm refers to the reinforcement learning algorithm used to generate cooperative control commands for UAV swarms. For example, for the Sierpinski fractal structure, the system runs the G-MIXDDPG algorithm, and the policy network outputs the speed and heading angle adjustments for each UAV based on standardized state data. Grouping information refers to the identification information of the formation level to which UAVs within the same formation unit belong in a fractal formation. For example, in a first-level formation, the leader UAV L1 and the follower UAVs F1, F2, and F3 are all marked as the same group, and the scheduling algorithm only uses the state characteristics of UAVs within the same group to calculate the action value.

[0046] The loss function of the G-MIXDDPG algorithm is: in, This represents the loss value of the policy network; These are the parameters of the policy network; This represents the total number of samples collected in the experience replay pool. The action value function output by the hybrid critic network with attention mechanism. As the current strategy, Let i be the state of the i-th agent. Let k be the group to which the i-th agent belongs, k be the group type, and att represent the introduction of an attention mechanism; The target action value is the true label value of the strategy.

[0047] The gradient update formula for the G-MIXDDPG algorithm is as follows: in, For policy network parameters The gradient is used for parameter iterative updates; This represents the policy network of the i-th agent with respect to the parameters. gradient, For the action of the i-th agent, For the observations of the i-th agent; The action value function with attention mechanism is expressed as a function of action. The gradient; This indicates that the action is generated by the current policy network based on observations.

[0048] The formula for the target action value of the G-MIXDDPG algorithm is as follows: in, This represents the immediate reward for the i-th agent; This represents the discount factor, used to weigh immediate rewards against future rewards; This represents the action-value function of the target critic network; This represents the state of the i-th agent at the next moment; Indicates the task type corresponding to the i-th agent. Grouping; This represents the target action of the j-th agent; Represents the target policy network; Let represent the observation value of the j-th agent.

[0049] The loss function of the MIX-MAPPO algorithm is: in, It represents the mathematical expectation at time step t, used to characterize the long-term average performance of the strategy; Indicates the total number of agents; Let represent the advantage function of the i-th agent, which characterizes the degree of advantage of the current action relative to the average policy; Indicates the strategy at time step t. Generate Actions The logarithmic probability, Let this be the state of the i-th agent at time t; This represents the regularization coefficient, used to balance the weights for policy optimization and weight distribution optimization; This represents the weight of the i-th agent; This represents the weight distribution strategy and characterizes the state. Generate weights The probability of.

[0050] The leader selection unit employs a fuzzy logic algorithm, and the remaining energy of the nodes in the input parameters satisfies the following formula: in, Let be the remaining energy of the m-th node; Let be the initial total energy of the m-th node; Let m be the energy consumption for transmitting data at the m-th node; Let m be the energy consumption for receiving data at the m-th node; The aggregated energy consumption of the model at the m-th node; The energy consumption for training the model at the m-th node.

[0051] The node centrality satisfies the formula: in, Let m be the node centrality of the m-th node; This represents the number of neighboring nodes of the m-th node. Let be the Euclidean distance between the m-th node and its neighboring node i.

[0052] The client selection unit uses a multi-armed gambling machine algorithm, and the client selection score satisfies the following formula: in, Score the upper bound of the confidence of the i-th node in the t-th round of training; c is the historical average reward; 'c' is the exploration factor. Let be the number of times the i-th node is selected.

[0053] The cruise speed adjustment satisfies the following formula: in, This is the adjusted cruise speed; The base cruise speed; p is the speed adjustment factor; For the magnitude of acceleration, T represents the decision-making cycle.

[0054] Among them, the collaborative decision-making instruction refers to the instruction generated by the scheduling decision-making layer 103 and issued to the execution layer 104 for controlling the collaborative action of the drone swarm; for example, the scheduling decision-making layer 103 generates a collaborative decision-making instruction for the current task, requiring the first-level formation leader drone L1 to fly at a heading angle of 60° and a speed of 8m / s, and the three follower drones to maintain a relative distance of 20m and a spacing of 120° from the leader L1.

[0055] The execution layer 104 is used to control the UAV cluster to perform formation actions and mission actions according to the collaborative decision-making instructions.

[0056] Among them, a drone swarm refers to a group of multiple drones that are organically organized through a collaborative scheduling system; for example, in this example, the drone swarm consists of 12 drones, which are divided into three primary formations and combined into one secondary formation.

[0057] Formation maneuvers refer to the actions performed by a drone swarm to form, maintain, or reconfigure a formation. For example, upon receiving a coordinated decision-making instruction, three follower drones, F1, F2, and F3, perform formation maneuvers, adjusting their flight speed and heading angle to maintain a predetermined relative position with the leader drone, L1. Mission maneuvers refer to the actions performed by a drone swarm to execute a specific mission objective. For example, during formation flight, the drone swarm performs mission maneuvers, acquiring visible light images and recording environmental parameters of the mission objective area.

[0058] The simulation verification layer 105 is used to perform scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

[0059] Scene modeling refers to the process of constructing a virtual three-dimensional flight environment using computer graphics and geographic information system (GIS) technologies. For example, simulation verification layer 105 constructs a three-dimensional terrain scene containing buildings, mountains, and no-fly zones, and sets the coordinates of obstacles and mission target points. Flight control simulation refers to the process of numerically simulating the flight kinematics and dynamics model of the UAV to verify the performance of the control algorithm. For example, simulation verification layer 105 performs flight control simulation on the UAV's takeoff, cruise, turning, and landing processes according to collaborative decision-making commands. The takeoff pitch angle is set to 30°, and the cruise speed is adjusted according to the formula v=v0+p×|α|×T, where the magnitude of acceleration α=5m / s². 2 Result visualization verification refers to the process of presenting simulation data in the form of graphs, curves, or charts for analysis and evaluation. For example, after the simulation, the result visualization submodule outputs a curve showing the change in formation position error rate over time, a curve showing the change in speed during obstacle avoidance, and a bar chart showing the formation reconstruction delay data.

[0060] The technical solution of this embodiment constructs a UAV swarm collaborative scheduling system that coordinates the operation of a perception layer 101, a communication layer 102, a scheduling decision layer 103, an execution layer 104, and a simulation verification layer 105. This solves the problems in the prior art, such as the decrease in convergence speed and increase in iteration time of multi-agent reinforcement learning algorithms due to the exponential increase in input dimension with the swarm size, the bandwidth limitation of radio frequency communication in complex low-altitude environments and the susceptibility to electromagnetic interference causing increased communication latency and data packet loss, and the lack of flexibility in fixed formation structures and the lack of dynamic adaptation capability in resource allocation, which makes it difficult to fully utilize the swarm efficiency. This improves the satisfaction of real-time control requirements and the efficiency of collaborative response of large-scale UAV swarms.

[0061] In one alternative embodiment, the perception layer 101 includes an unmanned aerial vehicle (UAV) onboard sensor array, an environmental monitoring sensor array, and a state fusion unit.

[0062] The UAV's onboard sensor group is used to collect the UAV's three-dimensional coordinates, flight speed, heading angle, pitch angle, remaining battery power, and onboard CPU computing power as the individual unit's status data.

[0063] The three-dimensional coordinates of the UAV refer to its longitude, latitude, and altitude in a Cartesian or geodetic coordinate system; for example, the three-dimensional coordinates of the Leader UAV L1 are (31.23°, 121.47°, 50m). Flight speed refers to the instantaneous rate of movement of the UAV relative to the air or ground; for example, the current flight speed of the Leader UAV L1 is 10m / s. Heading angle refers to the angle between the projection of the UAV's longitudinal axis onto the horizontal plane and true north; for example, the heading angle of the Leader UAV L1 is 45°. Pitch angle refers to the angle between the UAV's longitudinal axis and the horizontal plane; for example, the pitch angle of the Leader UAV L1 is 5°. Remaining battery power refers to the percentage of the UAV's onboard battery currently available; for example, the remaining battery power of the Leader UAV L1 is 85%. Onboard CPU computing power refers to the number of floating-point operations that the UAV's onboard central processing unit can perform per second; for example, the onboard CPU computing power of the Leader UAV L1 is 2.1 TFLOPS.

[0064] The environmental monitoring sensor group is used to collect the three-dimensional coordinates of obstacles, atmospheric visibility, real-time wind speed, and coordinates of the mission target area as environmental status data.

[0065] The three-dimensional coordinates of an obstacle refer to the longitude, latitude, and altitude of a static or dynamic obstacle in a low-altitude environment within a spatial coordinate system; for example, the three-dimensional coordinates of an obstacle detected by lidar are (31.24°, 121.48°, 60m). Atmospheric visibility refers to the maximum horizontal distance at which a target can be identified by normal visual perception in the atmosphere; for example, the current atmospheric visibility is 8km. Real-time wind speed refers to the instantaneous airflow rate in a low-altitude environment; for example, the real-time wind speed measured by an anemometer is 6m / s. The coordinates of the mission target area refer to the spatial location point where the UAV swarm needs to reach or perform the mission; for example, the coordinates of the mission target area are (31.25°, 121.49°, 30m).

[0066] The state fusion unit is used to perform weighted fusion of the individual state data and the environmental state data using a weighted average algorithm to obtain the fused standardized state data.

[0067] The weighted average algorithm refers to a mathematical method that calculates a weighted average value by assigning corresponding weights to data from different sources. For example, the state fusion unit uses the weighted average algorithm to assign a weight of 0.3 to the remaining power, 0.3 to the CPU computing power, and 0.4 to the flight status data of individual states, and assigns a weight of 0.5 to the obstacle distribution, 0.3 to the visibility, and 0.2 to the wind speed of the environmental state data, thereby calculating standardized state data.

[0068] In the above-mentioned optional methods, the UAV's three-dimensional coordinates, flight speed, heading angle, pitch angle, remaining battery power, and onboard CPU computing power are further collected as individual state data through the UAV's onboard sensor group. The environmental monitoring sensor group collects the three-dimensional coordinates of obstacles, atmospheric visibility, real-time wind speed, and coordinates of the mission target area as environmental state data. The state fusion unit uses a weighted average algorithm to perform weighted fusion on the above data, which solves the problem of insufficient information dimensions from a single data source leading to one-sided state representation, and improves the comprehensiveness of standardized state data and the reliability of scheduling decision input.

[0069] In one alternative approach, the two-tier self-organizing communication architecture includes an intra-group communication sublayer and an inter-group communication sublayer.

[0070] The intra-group communication sublayer refers to the layer in the two-level self-organizing communication architecture responsible for information exchange between drones within the same communication group. For example, the intra-group communication sublayer groups five drones with a spacing of less than 10 meters into one communication group, and constructs a star topology with the group's virtual center as the core to complete the sending and receiving of status data within the group. The inter-group communication sublayer refers to the layer in the two-level self-organizing communication architecture responsible for information exchange between leaders of different communication groups. For example, the inter-group communication sublayer establishes a fully connected communication link between the leaders of each group, and the leader encapsulates the standardized status data of their group and forwards it to other leaders.

[0071] The intra-group communication sublayer is used to divide communication groups according to a distance threshold and construct the star topology with the intra-group virtual center as the core. The inter-group communication sublayer is used to determine the node with the smallest Euclidean distance from the virtual center and the remaining power is higher than a preset percentage of the total power as the group leader, and the group leaders construct the fully connected communication link.

[0072] The virtual center within the group refers to the geometric center point calculated based on the three-dimensional coordinates of all drones within the group. For example, if the three-dimensional coordinates of five drones are (10m, 20m, 5m), (12m, 21m, 5m), (11m, 19m, 5m), (13m, 20m, 5m), and (12m, 22m, 5m), the virtual center coordinates are the arithmetic mean of these five coordinates (11.6m, 20.4m, 5m). The preset percentage of total battery power refers to the minimum remaining battery power threshold used to determine whether a drone is qualified to serve as the group leader. For example, if the preset percentage of total battery power is set to 60%, only drones with remaining battery power exceeding 60% of the total battery power can be designated as the group leader.

[0073] In the above-mentioned optional methods, communication groups are further divided according to distance thresholds through intra-group communication sub-layers and a star topology is constructed with the virtual center within the group as the core. The node with the smallest Euclidean distance from the virtual center within the group and a remaining power higher than a preset percentage of the total power is determined as the group leader through inter-group communication sub-layers. Fully connected communication links are then constructed between the group leaders. This solves the link redundancy and energy consumption imbalance problems caused by the rigidity of large-scale cluster communication topologies, and improves the self-organizing capability of the communication architecture and the stability of the group leader node operation.

[0074] In one alternative approach, the virtual center is determined based on the arithmetic mean of the three-dimensional coordinates of each UAV within the group.

[0075] The arithmetic mean refers to the sum of a set of values ​​divided by the number of values. For example, the sum of the x-coordinates of five drones is 58m. Dividing this by the number 5 gives the arithmetic mean of 11.6m, which is used as the x-coordinate of the virtual center within the group.

[0076] In the above-mentioned optional methods, the virtual center is further determined by the arithmetic mean of the three-dimensional coordinates of each UAV in the group. This solves the problems of group leader election bias and topology reconstruction inaccuracy caused by the ambiguity of the definition of the spatial reference point in the group, and improves the consistency of virtual center calculation and the rationality of communication group division.

[0077] In one alternative approach, the multi-level fractal formation structure is a Sierpinski fractal structure. The first-level formation of the Sierpinski fractal structure includes one leader drone and three follower drones. The leader drone is located at the center of the formation, and the three follower drones are evenly distributed around the leader drone with adjacent followers having a horizontal angle of 120 degrees. The second-level formation of the Sierpinski fractal structure consists of three first-level formations, and the leader of each first-level formation forms an equilateral triangle.

[0078] Among them, the Sierpinski fractal structure refers to a fractal pattern with self-similar properties formed by iteratively removing the central inverted triangle based on an equilateral triangle. For example, a first-level formation consists of a leader drone L1 and three follower drones F1, F2, and F3. L1 is located at the center of the formation, and F1, F2, and F3 are evenly distributed with L1 as the center and the horizontal angle between adjacent followers is 120°. After the three first-level formations are combined, each leader forms an equilateral triangle to form a second-level formation.

[0079] In this context, the leader drone refers to the drone located at the geometric center or a designated position within the formation, serving as a reference point for the movement of the followers. For example, in a primary formation, drone L1, with sufficient remaining battery power and computing power, is designated as the leader drone and positioned at the center of the formation. Follower drones refer to drones within the formation that maintain a predetermined relative position with the leader drone as a reference. For example, three follower drones F1, F2, and F3 are evenly distributed around the leader drone L1 on a circle, adjusting their positions in real time to maintain a relative distance of 20 meters and a 120° interval with L1. The formation center refers to the central point of the formation's geometric structure; for example, in a primary formation, the position of the leader drone L1 is the formation center. In the above-mentioned optional methods, the multi-level fractal formation structure is further limited to the Sierpinski fractal structure. The first-level formation includes one leader drone and three follower drones. The leader drone is located at the center of the formation, and the three follower drones are evenly distributed around the leader drone with the horizontal angle between adjacent followers being 120 degrees. The second-level formation consists of three first-level formations, and the leader of each first-level formation forms an equilateral triangle. This solves the problem of insufficient task adaptability caused by the fixed formation size not being scalable, and improves the self-similarity characteristics of the formation structure and the flexibility of cluster size expansion.

[0080] In one alternative, the multi-level fractal formation structure is a Menger sponge fractal structure. The first-level formation of the Menger sponge fractal structure includes one leader drone and four follower drones. The four follower drones are evenly distributed around the leader drone and the horizontal angle between adjacent followers is 90 degrees. The second-level formation of the Menger sponge fractal structure consists of four first-level formations.

[0081] Among them, the Menger sponge fractal structure refers to a three-dimensional fractal structure with self-similar properties formed by iteratively removing the central cube based on a cube. For example, a first-level formation consists of a leader drone L2 and four follower drones F4, F5, F6, and F7. F4, F5, F6, and F7 are evenly distributed with L2 as the center and the horizontal angle between adjacent followers is 90°. The four first-level formations combine to form a second-level formation.

[0082] Among the above-mentioned optional methods, the multi-level fractal formation structure is further limited to a Menger sponge fractal structure. The first-level formation includes one leader drone and four follower drones. The four follower drones are evenly distributed around the leader drone and the horizontal angle between adjacent followers is 90 degrees. The second-level formation consists of four first-level formations. This solves the problem of the limited application scenarios of the single fractal mode, which leads to the single formation type problem. It improves the diversity of fractal formation structure and its adaptability to different mission scenarios.

[0083] In one alternative approach, the simulation verification layer 105 includes a scene modeling submodule, a UAV control submodule, a communication simulation submodule, and a result visualization submodule.

[0084] The scene modeling submodule is used to construct a three-dimensional terrain scene and set the three-dimensional coordinates of obstacles and the coordinates of target points. The UAV control submodule is used to control the take-off, cruise, turning and landing of the UAV. The communication simulation submodule is used to simulate the two-level self-organizing communication architecture. The result visualization submodule is used to output the formation position error rate, obstacle avoidance speed change curve and formation reconstruction delay data corresponding to the scheduling algorithm.

[0085] The target point coordinates refer to the spatial location of the destination of the UAV swarm flight mission; for example, the target point coordinates are (31.25°, 121.49°, 30m).

[0086] The formation position error rate refers to the statistical measure of the deviation between the actual position of the UAV and the desired formation position. For example, during the simulation, the distance deviation between the actual position of the follower UAV and the predetermined circular position is calculated every 1 second, and the ratio of the average deviation of all followers to the safe distance is used as the formation position error rate. The obstacle avoidance speed change curve is a graphical representation of the change in flight speed of the UAV over time while avoiding obstacles. For example, the UAV decelerates to 5 m / s at a distance of 50 m from the obstacle, accelerates to 10 m / s after bypassing the obstacle, forming a speed change curve that first descends and then rises. The formation reconstruction delay data is a quantitative record of the time required for the formation to transform from the current formation to the target formation. For example, it takes 3.2 seconds for a primary formation to reconstruct from a line formation to a Sierpinski triangle formation; this delay data is recorded to evaluate the real-time performance of the scheduling algorithm.

[0087] In the above-mentioned optional methods, a three-dimensional terrain scene is further constructed through the scene modeling submodule and the three-dimensional coordinates of obstacles and the coordinates of target points are set. The drone control submodule controls the take-off, cruise, turning and landing of the drone. The two-level self-organizing communication architecture is simulated through the communication simulation submodule. The formation position error rate, obstacle avoidance speed change curve and formation reconstruction delay data corresponding to the scheduling algorithm are output through the result visualization submodule. This solves the problem of uncontrollable deployment risks caused by the lack of algorithm verification and improves the data support and simulation verification completeness of the scheduling algorithm optimization iteration.

[0088] like Figure 2 As shown, the UAV swarm collaborative scheduling system includes a perception layer 101, a communication layer 102, a scheduling decision layer 103, an execution layer 104, and a simulation verification layer 105. The perception layer 101 includes an UAV onboard sensor group, an environmental monitoring sensor group, and a state fusion unit. The communication layer 102 includes intra-group communication sublayers and inter-group communication sublayers. The scheduling decision layer 103 includes a fractal formation module, a leader selection unit, a client selection unit, and an improved multi-agent reinforcement learning scheduling module. The execution layer 104 includes a formation control unit and a task execution unit. The simulation verification layer 105 includes a scene modeling submodule, a UAV control submodule, a communication simulation submodule, and a result visualization submodule. The fractal formation module supports multi-level fractal formation structures based on Sierpinski fractal structure and Menger sponge fractal structure. An improved multi-agent reinforcement learning scheduling module integrates the G-MIXDDPG and MIX-MAPPO algorithms. The G-MIXDDPG algorithm is adapted to Sierpinski fractal formations, while the MIX-MAPPO algorithm is adapted to Menger sponge fractal formations. Both algorithms embed attention mechanisms and hybrid critic networks. The UAV swarm collaborative scheduling system also executes corresponding collaborative scheduling methods, including perception data acquisition and fusion, two-level self-organizing communication networking, fractal formation construction, improved multi-agent algorithm decision-making, task execution, and simulation verification.

[0089] like Figure 3 As shown, the workflow of the UAV swarm collaborative scheduling method includes: the perception layer 101 performs data acquisition and fusion; the communication layer 102 performs two-level self-organizing networking; the scheduling decision layer 103 performs fractal formation construction, scheduling algorithm decision-making, and node selection; the execution layer 104 performs formation control, task execution, and status feedback; and the simulation verification layer 105 performs scenario modeling, flight control simulation, and result visualization. Each layer operates collaboratively according to the data interaction sequence, forming a complete closed loop from perception acquisition to decision control and then to simulation verification.

[0090] In one embodiment: The perception layer 101 is the information acquisition center for cluster scheduling. The accuracy of data acquisition and the effectiveness of data fusion directly determine the reliability of upper-layer scheduling decisions. This embodiment addresses the heterogeneity and redundancy of multi-source sensor data in low-altitude environments. From hardware selection and acquisition process to algorithm implementation, it achieves accurate perception and standardized fusion of the individual UAV status and environmental status, providing unified data support for subsequent networking and formation.

[0091] In terms of hardware selection and deployment, the UAV's onboard sensor suite uses a GNSS sensor with an accuracy of ±0.5m to collect the UAV's three-dimensional coordinates, a three-axis velocity sensor with a range of 0-100m / s to collect flight speed, an attitude sensor with an accuracy of ±0.1° to collect heading and pitch angles, a battery monitoring chip with an error of ±1% to collect remaining battery power, and a computing power monitoring chip to collect the computing power of the onboard central processing unit in real time. The environmental monitoring sensor suite uses a lidar with a detection range of 0-200m to collect the three-dimensional coordinates of obstacles, a visibility meter with a measurement range of 0-50km to collect atmospheric visibility, an anemometer with a range of 0-60m / s to collect real-time wind speed, and a target positioning sensor with an accuracy of ±1m to collect the coordinates of the target area. The state fusion unit uses an embedded chip as the core processor, equipped with 1GB of RAM and 8GB of storage to store the fusion algorithm program and sensing data.

[0092] Regarding the data acquisition process, the individual UAV status acquisition process is as follows: After the UAV is powered on, the UAV's onboard sensor group enters the working mode. The GNSS sensor outputs the UAV's three-dimensional coordinate data every 100ms, the three-axis velocity sensor outputs flight speed data every 50ms, the attitude sensor outputs heading and pitch angle data every 50ms, the battery monitoring chip outputs remaining battery data every 1s, and the computing power monitoring chip outputs onboard central processing unit computing power data every 1s. All data is transmitted to the status fusion unit in real time. The environmental status acquisition process is as follows: The environmental monitoring sensor group and the UAV's onboard sensor group start up synchronously. The lidar outputs obstacle three-dimensional coordinate data every 200ms through 360° scanning, the visibility meter outputs atmospheric visibility data every 5s, the anemometer outputs real-time wind speed data every 100ms, and the target positioning sensor outputs the initial target area coordinates once when the mission starts, and updates them every 10s thereafter. The data is transmitted synchronously to the status fusion unit.

[0093] Regarding the practical implementation of the state fusion algorithm, to achieve effective integration of multi-source heterogeneous data, a weighted average fusion model was first built based on Python language combined with numerical computing libraries. This involved developing data standardization and weighted summation functions, and embedding individual state fusion values. Integration value with environmental status The calculation formula is then used. Subsequently, 1000 sets of individual state data and environmental state data are collected as a training set. The stability of the fused data is used as an evaluation metric to iteratively optimize the accuracy of the weight coefficients until the fluctuation range of the standardized state data output by the model is controlled within ±0.02. Finally, the trained fusion algorithm is compiled into an embedded executable program and burned into the storage chip of the embedded chip. Simultaneously, a data reception interrupt response mechanism is set to ensure real-time access and fusion calculation of sensor data.

[0094] In terms of algorithm execution, after sensor data is input to the state fusion unit, it first enters the standardization function section, where it is converted into values ​​between 0 and 1 according to a preset mapping rule. The standardized data then enters the weighted calculation function section, where it is substituted into the individual state fusion values. Integration value with environmental status The calculation formula completes the fusion of the individual unit state and the environment state. The fused individual unit state value is then calculated. Integration value with environmental status The data is uploaded to the scheduling decision layer 103 at a frequency of 100ms, serving as the basis for subsequent scheduling.

[0095] In this embodiment, by combining data complementarity from multiple sensors with a weighted average algorithm, the measurement error of a single sensor can be effectively offset, enabling accurate perception of environmental information such as obstacle distribution. Standardized processing eliminates the dimensional differences between different types of data, providing unified data support for upper-level scheduling decisions. At the same time, it covers the full-dimensional status of individual UAVs and the environment, achieving comprehensive perception of the cluster itself and the environment, laying a complete data foundation for subsequent formation and decision-making.

[0096] In yet another embodiment: Improving the multi-agent reinforcement learning scheduling module is the core of cluster collaborative decision-making, and the performance of the scheduling algorithm directly affects the response speed and control accuracy of the formation. This embodiment addresses the problem of input dimension explosion in traditional multi-agent algorithms by completing the architectural design and implementation adaptation of the G-MIXDDPG and MIX-MAPPO algorithms, respectively. By reducing network complexity through attention mechanisms and group information filtering strategies, precise collaborative control under fractal formations is achieved.

[0097] In the practical implementation of the G-MIXDDPG algorithm, to adapt to the topological characteristics of the Sierpinski fractal formation, the network architecture needs to be built first. The policy network and target policy network employ three fully connected layers. The input layer dimension includes eight-dimensional agent state features, encompassing the UAV's three-dimensional coordinates, remaining battery power, and onboard CPU computing power. The hidden layer has 256 neurons with ReLU activation. The output layer outputs the UAV's speed and heading angle control commands. The hybrid critic network with attention mechanism and the target critic network first integrate the agent state... Grouping information Then, an attention mechanism layer assigns weights to agents in different groups. These attention weights are calculated using a scaled dot product mechanism. This is followed by two fully connected layers with 256 hidden neurons and ReLU activation function, ultimately outputting the action value function. During network initialization, the policy network parameters are... and target policy network parameters The critic network parameters are initialized using a normal distribution. And target critic network parameters We initialized using Xavier with a learning rate of 0.001.

[0098] After the network is set up, a corresponding training process needs to be designed. First, set the capacity to... The experience replay pool is stored in a four-tuple format, containing state, action, reward, and next state, with a batch size of 1024 samples per iteration. Then, the loss function is applied... The formula calculates the network loss value of the strategy, uses the Adam optimizer to perform gradient updates, and simultaneously calculates the target action value formula. The target network parameters are updated every 100 iterations at an update rate of 0.005. Training terminates when the loss value fluctuates less than 0.001 over 100 consecutive iterations and the action-value function on the validation set converges to a stable interval. The loss function formula is: in, This represents the loss value of the policy network; These are the parameters of the policy network; This represents the total number of samples collected in the experience replay pool. The action value function output by the hybrid critic network with attention mechanism. As the current strategy, Let i be the state of the i-th agent. Let k be the group to which the i-th agent belongs, k be the group type, and att represent the introduction of an attention mechanism; Value of the target action.

[0099] The gradient update formula is: in, For policy network parameters The gradient is used for parameter iterative updates; This represents the policy network of the i-th agent with respect to the parameters. gradient, For the action of the i-th agent, For the observations of the i-th agent; The action value function with attention mechanism is expressed as a function of action. The gradient; This indicates that the action is generated by the current policy network based on observations.

[0100] The formula for the value of a target action is: in, This represents the immediate reward for the i-th agent; This represents the discount factor, used to weigh immediate rewards against future rewards; This represents the action-value function of the target critic network; This represents the state of the i-th agent at the next moment; Indicates the task type corresponding to the i-th agent. Grouping information; This represents the target action of the j-th agent; Represents the target policy network; Let represent the observation value of the j-th agent.

[0101] To meet the lightweight requirements of airborne deployment, the trained model needs to be processed. First, a model pruning technique is used to remove weights in the network whose absolute values ​​are less than a certain threshold. The network parameters are then quantized into 16-bit floating-point numbers through quantized perception training, compressing the model size to 40% of its original size. The lightweight model is then deployed to the UAV's onboard computing chip, with a decision cycle set to 100ms to ensure real-time issuance of collaborative decision-making commands. Simultaneously, the grouping information of the Sierpinski fractal formation needs to be... The input scheduling algorithm uses first-level formations as basic groups and second-level formations as aggregation groups, which allows the scheduling algorithm to focus only on the state characteristics of agents within the same formation unit, thus reducing the input dimensionality.

[0102] In the implementation of the MIX-MAPPO algorithm, to address the differentiated control requirements of Menger sponge fractal formations, the policy network and value network employ three fully connected layers. The input layer has a joint dimension of 10 dimensions, representing the agent's state and actions. The hidden layer contains 512 neurons, with ReLU activation. The policy network outputs the action probability distribution, and the value network outputs the action value. During network initialization, Xavier parameters are used, with a learning rate of 0.0003 and a regularization coefficient of... Let's take 0.01. The loss function formula is: in, It represents the mathematical expectation at time step t, used to characterize the long-term average performance of the strategy; Indicates the total number of agents; Let represent the advantage function of the i-th agent, which characterizes the degree of advantage of the current action relative to the average policy; Indicates the strategy at time step t. Generate Actions The logarithmic probability, Let this be the state of the i-th agent at time t; This represents the regularization coefficient, used to balance the weights for policy optimization and weight distribution optimization; This represents the weight of the i-th agent; This represents the weight distribution strategy and characterizes the state. Generate weights The probability of.

[0103] During the training phase, the generalized dominance estimation method is first used to calculate the dominance function. The GAE coefficient was set to 0.95, and then the loss function was applied. The loss value is calculated using a formula, and gradient updates are performed using the Adam optimizer. Policy pruning is performed every 200 iterations to ensure the stability of policy updates. Training terminates when the KL divergence of the policy network is less than 0.01 and the prediction error of the value network is less than 0.05. The lightweight deployment process is consistent with the G-MIXDDPG algorithm, while also needing to adapt to the grouping information of the Menger sponge fractal formation to achieve differentiated and precise control.

[0104] In this embodiment, by filtering key input data through grouping information and attention mechanisms, the input dimension of the scheduling algorithm is significantly reduced, thus reducing the amount of computation. The hybrid critic network accurately evaluates the action value of agents within the same formation unit, avoiding evaluation bias caused by full agent data and improving the convergence speed of the scheduling algorithm. The adaptation of the scheduling algorithm to the fractal formation structure enables the collaborative decision-making instructions to be accurately issued to each formation unit, improving the decision-making accuracy of the formation reconstruction task.

[0105] In yet another embodiment: Fractal formation modules are key to achieving orderly management and control of large-scale clusters, and their self-similarity properties support flexible expansion of formation levels. This embodiment addresses the formation requirements of different task scenarios by constructing formations using Sierpinski fractal structures and Menger sponge fractal structures, clarifying the positional arrangement rules and safe distance constraints for each formation level, thus achieving efficient collaborative management and control of the cluster.

[0106] In implementing the Sierpinski fractal formation, the basic parameters are first calculated, and then... (The sentence is incomplete and requires more context to be translated accurately.) Calculate the radius of the expected location region, where The radius of the expected location area. For flight speed, The unit of time is used. Flight speed is taken as the unit of time. unit time ,have to The safe distance is calculated using the formula. Calculation yields ,in This refers to the safe distance between drones. The standard distance between the leader drone and the follower drone is... Next, the first-level formation is arranged, with one leader drone placed at the center. Three follower drones are evenly distributed around the leader drone in a circle with a radius of 20 meters. The horizontal angle between adjacent followers is 120°, ensuring a distance of [missing information - likely a specific distance between drones]. Maintain a safe distance to avoid collisions.

[0107] To build a secondary formation, first select three completed primary formations as basic units, then calculate the coordinates of the three primary formation leader drones, ensuring that the leader drones form an equilateral triangle, with the centroid of the triangle at a distance from each leader drone. The second-level formation has been completed.

[0108] In implementing the Menger sponge fractal formation, the basic parameters of the first-level formation are consistent with the Sierpinski fractal structure. , The standard distance between the leader drone and the follower drones is 20m. One leader drone is placed at the center of the formation, and four follower drones are evenly distributed around the leader drone in a circle with a radius of 20m. The horizontal angle between adjacent follower drones is 90°, and the distance between adjacent follower drones is [missing information]. This meets the safety distance requirements.

[0109] When building a secondary formation, select four completed primary formations, adjust the positions of the leader drones of each primary formation to form a square arrangement, and complete the construction of the secondary formation. It can be expanded into a tertiary or higher formation according to mission requirements.

[0110] In this embodiment, both fractal structures possess self-similarity characteristics. The primary formation can be freely combined into higher-level formations as basic units, enabling orderly control of drone swarms ranging from dozens to hundreds of drones. The geometric symmetry of the fractal structure ensures uniform force distribution during flight, and the presence of the leader drone at the center of the formation reduces airflow disturbance and enhances formation stability. Different fractal structures can adapt to different mission types; the Sierpinski fractal structure is suitable for large-area coverage, while the Menger sponge fractal structure is suitable for high-precision inspection, thus improving the mission adaptability of the swarm.

[0111] In yet another embodiment: Communication layer 102 is the core hub for information exchange among UAV swarms. Traditional radio frequency communication suffers from limitations in bandwidth and weak anti-interference capabilities, making it difficult to support real-time collaboration in large-scale swarms. This embodiment, based on free-space optical communication technology, establishes a two-level self-organizing communication architecture. Through a layered design of intra-group star topology and inter-group fully connected communication links, it ensures low-latency information exchange within the group while achieving global information exchange across the swarm, providing stable communication support for upper-level scheduling decisions.

[0112] In terms of hardware selection and deployment, the group's communication sublayer hardware is equipped with wavelengths for each drone. The free-space optical communication transceiver module is equipped with an angle deviation compensation function with an accuracy of ±0.1°, ensuring the stability of the communication link under high-speed movement. It also features a ranging function with an accuracy of ±0.5m, used for real-time sensing of the distance between drones, providing data support for communication group division. The inter-group communication sublayer hardware upgrades each communication group leader drone with a high-power optical communication module, increasing the transmission power to twice that of a standard module, and adding a communication relay function to extend the inter-group communication distance. It also deploys a remaining battery monitoring function with an accuracy of ±1%, used to select qualified group leader nodes. The communication control unit uses an FPGA embedded chip as the communication scheduling core, equipped with a real-time operating system, supporting dynamic communication group division, group leader determination, and link status monitoring, with a response latency of ≤50ms.

[0113] Regarding the implementation of the intra-group communication sublayer, the distance threshold for communication group division is determined by the formula... Calculation, where Distance thresholds for dividing communication groups, The maximum flight speed of the drone. This is the communication latency tolerance threshold. The maximum flight speed of the drone is taken as the threshold. Communication delay tolerance threshold The distance threshold is calculated. This means that drones with a spacing of 10m or less can be classified into the same communication group.

[0114] Taking a communication group as an example, the group contains 5 drones with 3D coordinates of (10,20,5), (12,21,5), (11,19,5), (13,20,5), and (12,22,5). The 3D coordinates of the virtual center within the group are determined by the formula... Calculation, where The three-dimensional coordinates of the virtual center within the communication group. , , Let i be the three-dimensional coordinates of the i-th UAV in the group. The number of drones in the group is given. The coordinates of the virtual center within the group are calculated to be (11.6, 20.4, 5). A star topology is constructed using this as the core, and all drones in the group establish optical communication links with the drone that acts as the virtual center node within the group.

[0115] In terms of signal attenuation control, atmospheric visibility is collected in real time. ,when At that time, according to the formula Calculate the signal attenuation coefficient, where This represents the signal attenuation coefficient for free-space optical communication within the group. Atmospheric visibility, For communication wavelength, This is the atmospheric attenuation coefficient. (The atmospheric attenuation coefficient is taken as...) The attenuation coefficient of the optical communication signal within the group was calculated. When atmospheric visibility drops to 5km, the optical power compensation mechanism is automatically activated, increasing the transmission power by 30% to ensure that the link signal-to-noise ratio is greater than or equal to 15dB and maintain communication stability.

[0116] Regarding the implementation of the inter-group communication sublayer, the process for determining the group leader involves iterating through all drones within each communication group and selecting the node with the smallest Euclidean distance to the group's virtual center and remaining battery power exceeding 60% of the total battery power as the group leader. Taking a certain communication group as an example, drone A is 0.8m from the group's virtual center and has 75% remaining battery power, while drone B is 1.2m from the group's virtual center and has 62% remaining battery power. Ultimately, drone A is determined to be the group leader.

[0117] The inter-group full-connectivity networking process establishes full-connectivity communication links between all group leader nodes to achieve inter-group information exchange. If the cluster is divided into 3 communication groups, with 3 group leader nodes A, B, and C, then bidirectional optical communication links are established between A and B, A and C, and B and C, with a link bandwidth greater than or equal to 100Mbps to meet the cluster's global information synchronization requirements.

[0118] The inter-group information forwarding process involves the group leader node receiving standardized status data from drones within the group, encapsulating it into a unified format according to a preset protocol, and forwarding it to other group leaders via a fully connected inter-group communication link. Simultaneously, it receives data forwarded by other group leaders, splits it, and distributes it to each drone within the group, thereby achieving global sharing of cluster status.

[0119] In this embodiment, the two-level self-organizing communication architecture divides the cluster into multiple communication groups. The star topology within each group reduces the communication complexity of a single group and ensures low-latency information interaction within the group. The fully connected communication links between groups enable global information exchange and ensure the stability of data transmission between group leaders. The application of free-space optical communication technology effectively avoids the electromagnetic interference problem of radio frequency communication and can maintain high communication link availability even in complex low-altitude environments, providing stable and low-latency information transmission support for upper-level scheduling decisions.

[0120] In yet another embodiment: The leader selection unit and client selection unit are crucial components of cluster collaborative scheduling. Inappropriate node selection can lead to decreased formation stability and reduced training efficiency. This embodiment combines fuzzy logic and multi-armed machine algorithms to achieve precise selection of leader drone nodes and dynamic optimization of client nodes, ensuring efficient training of the collaborative control and scheduling algorithm for fractal formations.

[0121] In terms of hardware and algorithm deployment preparation, the hardware support includes equipping each drone with remaining power monitoring, node degree perception, and distance sensors to collect the remaining energy of the nodes in real time. Node degree Node centrality Core data is collected; edge computing chips are deployed to run fuzzy logic algorithms and multi-armed gambling machine algorithms, with a decision latency of ≤30ms. The algorithm model is built using Python language combined with a fuzzy logic library to build a fuzzy logic model, and a multi-armed gambling machine confidence upper bound model is built based on a deep learning framework. After being solidified, the models are burned into the edge computing chip to support real-time decision-making.

[0122] Regarding the implementation of the leader selection unit, the input parameter calculation process is as follows: Remaining energy calculation: Take the initial total energy of a certain UAV. Calculated data transmission energy consumption Energy consumption for receiving data Model aggregation energy consumption Model training energy consumption According to the formula Calculate the remaining energy, where Let be the remaining energy of the m-th node. Let be the initial total energy of the m-th node. Let m be the energy consumption for transmitting data at the m-th node. Let m be the energy consumption for receiving data at the m-th node. The aggregated energy consumption of the model for the m-th node is... The energy consumption for training the model at the m-th node. The remaining energy. After standardization, it is 0.6.

[0123] Node degree and node centrality calculation: the number of neighboring nodes of the drone The Euclidean distances to adjacent nodes are 8m, 10m, 9m, and 11m, respectively, according to the formula... Calculate the node centrality, where Let m be the node centrality of the m-th node. Let m be the number of neighboring nodes of the m-th node. Let be the Euclidean distance between the m-th node and its neighboring node i. Then, calculate the node centrality. Node degree After standardization, it is 0.8.

[0124] The fuzzy logic decision-making process is as follows: Fuzzification process: The remaining energy domain of discourse 0 to 1 is divided into three fuzzy sets: low, medium, and high; the node degree domain of discourse 0 to 1 is divided into three fuzzy sets: few, medium, and many; and the node centrality domain of discourse 0 to 15m is divided into three fuzzy sets: far, medium, and near. The input parameters are fuzzified by the triangular membership function.

[0125] Rule reasoning: The core rule is set as follows: if the remaining energy is high, the node degree is high, and the node center degree is close, then the leader fitness is high. Through reasoning, the leader fitness of this drone is found to be 0.85.

[0126] Deblurring: Deblurring is performed using the centroid method. All drones in the cluster are traversed, and nodes with a fitness score greater than or equal to 0.8 are selected as the leader drone of the fractal formation. If it is a secondary formation, the overall leader is determined to be the drone with the highest fitness score among the leader drones of each primary formation.

[0127] Regarding the implementation of the client-side selection unit, the confidence upper bound algorithm parameters are initialized to set the exploration factor. The number of times each node is selected in the initial stage Historical average reward .

[0128] The client-side dynamic selection process is as follows: Reward function setting: The training efficiency of the client node is converted into a reward value. The shorter the training time and the lower the latency, the higher the reward value. The reward range is 0 to 1.

[0129] Confidence upper bound score calculation: The average historical reward of node i in the t-th round of training. Number of times selected According to the formula Calculate the upper bound of the confidence score, where Score the upper bound of the confidence of the i-th node in the t-th round of training. This is the historical average reward. To explore factors, Let be the number of times the i-th node is selected. The upper bound of the confidence score is then calculated as follows: .

[0130] Node selection: Nodes are sorted from high to low according to their confidence upper bound scores, and the top 30% of nodes are selected as clients to participate in the collaborative training of the scheduling algorithm. After each round of training, the historical average reward and the number of times each node is selected are updated to achieve dynamic selection of client nodes.

[0131] In this embodiment, the fuzzy logic algorithm comprehensively selects leader drones based on multi-dimensional parameters, significantly improving the stability of the leader drone node's formation control and avoiding formation collapse caused by a single parameter selection; the multi-arm gambling machine confidence upper bound algorithm realizes dynamic selection of client nodes, accelerates the training convergence speed of the scheduling algorithm, and reduces training latency; the dynamic switching mechanism between leader drones and client nodes ensures the task continuity of the cluster in the event of node failure, greatly enhancing the overall robustness of the cluster.

[0132] In summary, the improved multi-agent reinforcement learning scheduling module integrates the G-MIXDDPG and MIX-MAPPO algorithms, which are adapted to different fractal formations. Both algorithms embed attention mechanisms and hybrid critic networks. By combining the grouping information of the fractal formations to filter key agent state features, it integrates only features within the same formation unit, avoiding the input of all agent data, significantly reducing the input dimensionality, and improving the convergence speed of the scheduling algorithm. At the same time, it matches the formation topology characteristics to simplify network complexity, enabling real-time and accurate collaborative decision-making for large-scale clusters. The communication layer 102 adopts a two-level self-organizing communication architecture based on free-space optical communication. The intra-group communication sublayer divides communication groups according to distance thresholds and constructs a star topology. The inter-group communication sublayer determines the group leader that meets the conditions and constructs a fully connected communication link to avoid the electromagnetic interference defects of radio frequency communication. At the same time, the hierarchical networking reduces the communication complexity of a single group, ensuring low-latency and high anti-interference information interaction in complex low-altitude environments, and providing stable communication support for cluster collaboration. The fractal formation module supports multi-level fractal formation structures based on Sierpinski and Menger sponge fractal structures. First-level formations can be freely combined into higher-level formations using basic units. Self-similarity allows for flexible expansion of formation levels. Different fractal structures can adapt to different task types. Geometric symmetry ensures uniform force distribution during formation flight. The central leader UAV reduces airflow disturbance, improving both formation stability and swarm task adaptability. The leader selection unit uses a fuzzy logic algorithm to comprehensively consider node remaining energy, node degree, and node centrality to select the leader UAV. The client selection unit uses a multi-armed gambler algorithm to dynamically select collaboratively trained client nodes. This dynamic resource adaptation mechanism ensures the stability of fractal formation collaborative control, accelerates the training convergence speed of the scheduling algorithm, and enables dynamic switching in case of node failure, improving the overall robustness of the swarm. This solves the problems of input dimension explosion in multi-agent reinforcement learning algorithms, limited and easily interfered radio frequency communication bandwidth, and insufficient formation structure flexibility inherent in traditional UAV swarm control technologies.

[0133] Figure 4 This diagram illustrates a flowchart of an embodiment of a drone swarm collaborative scheduling method provided by the present invention. This drone swarm collaborative scheduling method employs a drone swarm collaborative scheduling system 100 as provided by the present invention. Figure 4 As shown, the method includes the following steps: S1. The perception layer 101 performs weighted fusion of the individual state data collected by the UAV airborne sensor group and the environmental state data collected by the environmental monitoring sensor group, and outputs the fused standardized state data. S2. The communication layer 102 constructs a two-level self-organizing communication architecture based on free space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to the distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer 102 is used to transmit the standardized state data via the star topology and the fully connected communication link. S3. The scheduling decision layer 103 constructs a multi-level fractal formation structure with self-similar characteristics, receives the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generates collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure. S4. The execution layer 104 controls the drone cluster to perform formation actions and task actions according to the collaborative decision-making instructions; S5. The simulation verification layer 105 performs scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

[0134] It should be noted that the beneficial effects of the UAV swarm collaborative scheduling method provided in the above embodiments are the same as those of the UAV swarm collaborative scheduling system 100 described above, and will not be repeated here. Furthermore, the method and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the system embodiments, and will not be repeated here.

[0135] The UAV swarm collaborative scheduling system 100 of the present invention can be a computer program (including program code) running on a computer device. For example, the UAV swarm collaborative scheduling system 100 of the present invention is an application software that can be used to execute the corresponding steps in the UAV swarm collaborative scheduling method of the present invention.

[0136] In some embodiments, the UAV swarm collaborative scheduling system 100 of the present invention can be implemented in a combination of hardware and software. As an example, the UAV swarm collaborative scheduling system 100 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the UAV swarm collaborative scheduling method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0137] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0138] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned UAV swarm collaborative scheduling methods. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the UAV swarm collaborative scheduling method shown in any embodiment of the present invention by calling the computer program.

[0139] In one alternative embodiment, an electronic device is provided, such as Figure 5 As shown, Figure 5 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0140] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0141] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0142] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0143] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0144] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0145] It should be noted that, Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0146] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described UAV swarm collaborative scheduling methods.

[0147] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0148] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned UAV swarm cooperative scheduling method.

[0149] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0150] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0151] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0152] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0153] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0154] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0155] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0156] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A drone swarm collaborative scheduling system, characterized in that, include: The layers are: perception layer, communication layer, scheduling decision layer, execution layer, and simulation verification layer. The perception layer is used to perform weighted fusion of the individual state data collected by the UAV onboard sensor group and the environmental state data collected by the environmental monitoring sensor group, and output the fused standardized state data. The communication layer is used to construct a two-level self-organizing communication architecture based on free space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to a distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer is used to transmit the standardized state data via the star topology and the fully connected communication link. The scheduling decision layer is used to construct a multi-level fractal formation structure with self-similar characteristics, receive the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generate collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure. The execution layer is used to control the UAV swarm to perform formation actions and mission actions according to the collaborative decision-making instructions; The simulation verification layer is used to perform scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

2. The UAV swarm collaborative scheduling system according to claim 1, characterized in that, The perception layer includes an airborne sensor group for the UAV, an environmental monitoring sensor group, and a state fusion unit; The UAV's onboard sensor group is used to collect the UAV's three-dimensional coordinates, flight speed, heading angle, pitch angle, remaining battery power, and onboard CPU computing power as the individual unit's status data; The environmental monitoring sensor group is used to collect the three-dimensional coordinates of obstacles, atmospheric visibility, real-time wind speed, and coordinates of the mission target area as the environmental status data. The state fusion unit is used to perform weighted fusion of the individual state data and the environmental state data using a weighted average algorithm to obtain the fused standardized state data.

3. The UAV swarm collaborative scheduling system according to claim 1, characterized in that, The two-level self-organizing communication architecture includes an intra-group communication sublayer and an inter-group communication sublayer. The intra-group communication sublayer is used to divide communication groups according to a distance threshold and construct the star topology with the intra-group virtual center as the core. The inter-group communication sublayer is used to determine the node with the smallest Euclidean distance from the virtual center and the remaining power is higher than a preset percentage of the total power as the group leader, and the group leaders construct the fully connected communication link.

4. The UAV swarm collaborative scheduling system according to claim 3, characterized in that, The virtual center is determined based on the arithmetic mean of the three-dimensional coordinates of each UAV in the group.

5. The UAV swarm collaborative scheduling system according to claim 1, characterized in that, The multi-level fractal formation structure is a Sierpinski fractal structure. The first-level formation of the Sierpinski fractal structure includes one leader drone and three follower drones. The leader drone is located at the center of the formation. The three follower drones are evenly distributed around the leader drone, and the horizontal angle between adjacent followers is 120 degrees. The second-level formation of the Sierpinski fractal structure consists of three first-level formations, and the leader of each first-level formation forms an equilateral triangle.

6. The UAV swarm collaborative scheduling system according to claim 1, characterized in that, The multi-level fractal formation structure is a Menger sponge fractal structure. The first-level formation of the Menger sponge fractal structure includes one leader drone and four follower drones. The four follower drones are evenly distributed around the leader drone and the horizontal angle between adjacent followers is 90 degrees. The second-level formation of the Menger sponge fractal structure consists of four first-level formations.

7. The UAV swarm collaborative scheduling system according to any one of claims 1 to 6, characterized in that, The simulation verification layer includes a scene modeling submodule, a UAV control submodule, a communication simulation submodule, and a result visualization submodule. The scene modeling submodule is used to construct a three-dimensional terrain scene and set the three-dimensional coordinates of obstacles and the coordinates of target points. The UAV control submodule is used to control the take-off, cruise, turning and landing of the UAV. The communication simulation submodule is used to simulate the two-level self-organizing communication architecture. The result visualization submodule is used to output the formation position error rate, obstacle avoidance speed change curve and formation reconstruction delay data corresponding to the scheduling algorithm.

8. A method for collaborative scheduling of unmanned aerial vehicle (UAV) swarms, employing the UAV swarm collaborative scheduling system as described in any one of claims 1 to 7, characterized in that, include: The perception layer performs weighted fusion of the individual state data collected by the UAV's onboard sensor group and the environmental state data collected by the environmental monitoring sensor group, and outputs the fused standardized state data. The communication layer constructs a two-level self-organizing communication architecture based on free-space optical communication. In the two-level self-organizing communication architecture, a star topology is constructed within the communication groups divided according to a distance threshold, and a fully connected communication link is constructed between the determined group leaders. The communication layer is used to transmit the standardized state data via the star topology and the fully connected communication link. The scheduling decision layer constructs a multi-level fractal formation structure with self-similar characteristics, receives the standardized state data transmitted via the star topology and the fully connected communication link by running a scheduling algorithm, and generates collaborative decision instructions by combining the grouping information corresponding to the multi-level fractal formation structure. The execution layer controls the drone swarm to perform formation actions and mission actions according to the collaborative decision-making instructions; The simulation verification layer performs scenario modeling, flight control simulation, and result visualization verification of the scheduling algorithm.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory, the memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the UAV swarm collaborative scheduling method as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the UAV swarm collaborative scheduling method as described in claim 8.

Images