Intelligent unmanned aerial vehicle cluster scheduling system and application thereof

Abstract

The application relates to the technical field of unmanned aerial vehicle control, and discloses an intelligent unmanned aerial vehicle cluster scheduling system and application thereof, the system comprising a working cluster composed of multiple unmanned aerial vehicles, each unmanned aerial vehicle being provided with a flight control module, a computing module, a battery and a communication module; the system triggers the state machine logic of the computing module through a task token data packet, and automatically allocates node roles; in flight operation, the system calculates kinematic energy gradient parameters in combination with residual power, spatial distance and wind field vector, triggers autonomous defense switching logic by using the gradient difference value of front and rear nodes, and adaptively adjusts the following distance according to the attenuation slope of the accompanying node; when the backhaul link is disconnected, the system cuts in key frame acquisition to reduce power consumption and storage occupation; in the low power working condition, the system calculates the landing cost in combination with the terrain elevation and the wind field penalty term, and guides the unmanned aerial vehicle to autonomously optimize and land. The application maintains the operation continuity of the unmanned aerial vehicle cluster under the condition of complex wind field and network disconnection and the reliability of hardware recovery.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, specifically to an intelligent UAV swarm scheduling system and its application. Background Technology

[0002] In recent years, swarm systems composed of multiple drones have been widely used in large-scale operations such as power transmission line inspection. These tasks require long-distance continuous flight under complex terrain and variable weather conditions. Most existing drone swarm scheduling relies on centralized command issuance from ground control stations or pre-set fixed programs to maintain formation and task allocation.

[0003] However, in actual operating environments, drones within a cluster often face inconsistent aerodynamic disturbances. The drone in the lead position experiences forward wind resistance, causing its battery to consume more energy than the trailing nodes. Traditional cluster control schemes often employ role handover strategies based on fixed time intervals or preset heading points. This static scheduling mechanism fails to consider the real-time physical wear and tear of the drone and the environmental wind field, leading to the lead drone losing power prematurely and thus disrupting the entire cluster's topology. Furthermore, existing fixed-distance following control cannot dynamically adjust the physical distance based on the energy decay of preceding nodes. In strong winds, the large relative displacement between drones can cause local communication links to be interrupted.

[0004] Furthermore, complex terrain often causes fluctuations or disconnections in the backhaul routing link from the cluster to the ground control station. During communication interruptions, the existing system maintains the original continuous video stream sampling frequency, which quickly fills up the limited onboard storage space, resulting in the overwriting or loss of critical operational data in the network outage area. When the battery level of a single drone triggers the return-to-home condition, the conventional landing and recovery procedure plans the return route based solely on the straight-line distance, lacking a comprehensive consideration of the three-dimensional terrain elevation difference and real-time wind field vectors. This single recovery strategy causes low-battery drones to accidentally enter areas with high wind resistance or drastic terrain undulations during landing, increasing the risk of equipment crashes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent drone swarm scheduling system and its application, which solves the problems of topological dispersion, data loss, and high crash risk caused by static scheduling strategies, communication disconnections, and single recovery mechanisms in complex operating conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent unmanned aerial vehicle (UAV) swarm scheduling system, comprising a work swarm consisting of multiple UAVs; each UAV is equipped with a flight control module, a computing module, a battery module, and a communication module; each UAV carries a sensing component, a camera module, and a storage module; the sensing component includes a positioning module and an inertial measurement module; the flight control module establishes a physical connection with the computing module through an internal communication bus; the battery module, communication module, and sensing component are respectively connected to the computing module through a data bus; the camera module and storage module are connected to the computing module through a data interface; based on the above hardware structure, the system generates a task token data packet containing a flight path coordinate sequence and a network routing table; the communication module is configured to receive the task token data packet and transmit it to the computing module to trigger state machine logic to complete node scheduling.

[0007] The system of this invention triggers logical switching of the underlying state machine through the internal flow of task token data packets, thereby constructing a distributed scheduling framework.

[0008] Furthermore, the computing module embeds a finite state machine logic program and a state register. During the system initialization phase, when the computing module detects that the task token data packet with a complete checksum has been successfully loaded into memory, it modifies the flag bit of the state register and sets the operating mode of the UAV to the master inspection node state. The computing module of the UAV that has not received the task token data packet executes the default preset branch logic of the finite state machine within the timeout waiting window and sets the operating mode of the UAV to the escort relay node state.

[0009] This invention utilizes packet status detection and timer timeout determination to automatically assign roles to cluster nodes.

[0010] Furthermore, the computing module in the accompanying relay node state reads its own topology sequence number in the cluster and receives the initial three-dimensional spatial coordinate vector and the unit direction vector along the predetermined operation route broadcast by the computing module in the main inspection node state; the computing module in the accompanying relay node state retrieves the preset initial spatial reference distance parameter, calculates the initial expected physical occupancy coordinates in combination with the topology sequence number, generates the expected position offset command and sends it to the flight control module to drive the UAV to deploy, and completes the physical construction of the linear following topology network.

[0011] The system transforms network hierarchical logic into distributed control commands in three-dimensional physical space through operations on topology sequence numbers and baseline distance scalars.

[0012] Furthermore, during flight operations, the battery module reads the remaining power integral data of the UAV's internal battery pack, and the sensing components output three-dimensional spatial coordinate data and apparent wind field vector data; the calculation module aligns the heterogeneous sensor data using absolute timestamps, and calculates and outputs kinematic energy gradient parameters by using the physical spatial distance and environmental drag coefficient of the denominator as attenuation weighting factors for energy consumption, and instructs the communication module to call the media access control layer protocol to locally broadcast the kinematic energy gradient parameters to neighboring UAVs.

[0013] This invention aligns and models the available battery power with real-time wind field vectors and spatial distances to output quantitative indicators that reflect the actual endurance of the drone.

[0014] Furthermore, the computing module in the accompanying relay node state receives the kinematic energy gradient parameters broadcast by the preceding node, establishes a sliding time window, and uses the least squares method to linearly fit the timestamp and energy gradient, outputting the attenuation slope data of the current time segment; the computing module transmits the attenuation slope data to the flight control module, and the flight control module substitutes the attenuation slope data into the position control closed loop to update the target following distance expectation value, so as to adaptively adjust the physical distance between the UAVs.

[0015] The system extracts the energy decay rate of the preceding node and feeds it back to the position control closed loop. When the energy consumption at the preceding node intensifies, the following distance is shortened to maintain the communication link.

[0016] Furthermore, the calculation module in the main inspection node state periodically extracts its own kinematic energy gradient parameters and the kinematic energy gradient parameters broadcast by the adjacent escort relay node and performs difference calculation; when the comparison difference increases positively and reaches the fault tolerance threshold for switching, the switching logic is triggered, and the command communication module sends the mission token data packet to the rear UAV via unicast; the original calculation module that handed over the mission token data packet issues a stop maneuver command to the flight control module to switch the UAV to the static communication relay node state and enter the fixed-point hovering control program; the rear calculation module that received the mission token data packet changes its operating mode to the main inspection node state and takes over the subsequent waypoint coordinate sequence.

[0017] The system uses the difference in energy gradient parameters between the preceding and following nodes as a trigger condition, and transfers the task token data packet according to the difference in physical loss to complete the dynamic handover of node roles.

[0018] Furthermore, the computing module in the main inspection node state monitors the global backhaul routing link to the ground control terminal through the communication module; when it is determined that the backhaul routing link is disconnected, the computing module sends an instruction set containing a speed attenuation coefficient to the flight control module to limit the output power of the underlying motor, cut off the continuous video stream sampling process of the camera module, and when the relative displacement increment reaches the set spatial sampling step size threshold or the time span reaches the forced trigger time period, it instructs the camera module to perform a key frame acquisition operation based on spatial coordinates and write the acquired image data into the storage module.

[0019] When the system detects a communication interruption, it reduces motor power to slow down energy consumption and converts video stream sampling into keyframe sampling triggered by spatial location to adapt to the storage capacity during the network outage.

[0020] Furthermore, when the original main inspection node triggers the switchover logic to switch to the static communication relay node state and verifies that the aircraft is in a stationary hovering attitude, the calculation module verifies that the backhaul route link is reconnected by sending a link probe message to the ground control terminal, thereby removing the preset transmission bandwidth limit in the network protocol stack of the communication module, and retrieving the lingering image data that has not been returned from the storage module, instructing the communication module to perform full-bandwidth burst data transmission using the current physical environment.

[0021] The system utilizes the relatively stable radio frequency channel environment when the aircraft is physically hovering to transmit the delayed data in batches after the link is restored.

[0022] Furthermore, when the calculation module determines that its own energy alarm margin parameter is less than the system's preset landing trigger threshold, it modifies the single-machine disconnection status bit of the status register to activate the underlying multi-level return-to-home recovery logic closed loop; the calculation module extracts the current three-dimensional spatial coordinates and three-axis attitude angle data, uses a heuristic search algorithm, and calculates the comprehensive landing cost function value of the multi-level landing cost model by taking the underlying dynamic constraints of the fuselage, the external environment wind field vector, and the terrain elevation difference as independent penalty terms. It selects the candidate ground recovery point with the smallest comprehensive landing cost function value as the best landing anchor point for the current UAV, guiding the UAV to leave the current network and enter an independent landing route.

[0023] This invention uses terrain elevation difference and wind field vector as penalty terms to calculate landing cost, enabling low-power drones to avoid high-energy-consuming flight areas and enter selected landing anchor points.

[0024] The second aspect of this invention provides an application of the aforementioned intelligent drone swarm scheduling system, which is applied to a high-altitude mountainous power transmission line inspection scenario. Under the complex wind field conditions of the inspection scenario, multiple drones in the work swarm construct a linear following topology network based on the node identity of the task token data packet. The system utilizes the potential energy comparison difference caused by the real-time calculated kinematic energy gradient parameters in the strong wind area to trigger a decentralized autonomous task role handover. During the disconnection and restoration of the backhaul routing link, the system performs adaptive keyframe image acquisition and full-bandwidth burst data clearing and backhaul. It also autonomously optimizes its landing to the selected optimal landing anchor point based on the external environment wind field and terrain undulations.

[0025] This invention deploys the scheduling system in actual inspection scenarios, maintaining the operational continuity of the drone swarm under complex wind conditions and network outages.

[0026] This invention provides an intelligent unmanned aerial vehicle (UAV) swarm scheduling system and its application. It has the following beneficial effects: 1. This invention directly triggers the state machine logic switching of the computing module through the flow of task token data packets within the system, and combines the difference in kinematic energy gradient parameters between adjacent UAVs as the switching condition. This design does not rely on the continuous issuance of commands from the ground control terminal, but directly transforms the actual physical wear difference of the aircraft into the basis for the handover of roles of the underlying nodes, reducing scheduling delays caused by long-distance communication delays and maintaining the topological continuity of the cluster during operation.

[0027] 2. This invention couples the apparent wind field vector, physical space distance, and remaining power of the UAV into a model, and calculates the kinematic energy gradient parameters that reflect the dynamic aerodynamic environment. The UAV in the state of a follower relay node extracts these parameters to calculate the energy attenuation slope of the preceding node, and substitutes it into the position control closed loop to dynamically update the target following distance. When the leading node encounters a complex wind field that leads to increased energy consumption, the follower node can actively shorten the physical distance to avoid the interruption of the internal communication link of the cluster due to excessive relative displacement.

[0028] 3. When the backhaul route link is determined to be disconnected, this invention reduces power consumption and storage space occupation during network outages by limiting the motor output power and switching the continuous video stream to key frame acquisition based on spatial location triggering. At the same time, when the UAV triggers the landing logic, the system introduces a multi-level landing cost model that includes terrain elevation difference and wind field vector penalty terms to guide the low-battery aircraft to avoid areas with high wind resistance and drastic undulations, and autonomously find the candidate ground recovery point with the lowest overall cost, thereby improving the recovery success rate of the hardware equipment. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the internal hardware module structure and connection of the system of the present invention; Figure 2 This is a schematic diagram of the main process of drone swarm operation execution according to the present invention; Figure 3 This is a schematic diagram of the system initialization and follow-topology network construction process of the present invention; Figure 4 This is a schematic diagram of the asynchronous node switching and decentralized handover logic flow of the present invention; Figure 5 This is a schematic diagram of the closed-loop process for communication link interruption degradation response and data clearing in this invention; Figure 6 This is a schematic diagram of the landing cost model evaluation and multi-level autonomous return control process of the present invention; Figure 7 This is a comprehensive schematic diagram of the simulation results of the system operation process of the present invention. In this diagram, a is a schematic diagram of the flight trajectory of the UAV swarm escort and defense replacement in three-dimensional space, and b is a schematic diagram of the comparison curve between the standardized operation time of the system and the absolute remaining power consumption trend.

[0030] Among them, 110 is the flight control module; 120 is the computing module; 130 is the battery module; 140 is the communication module; 150 is the sensing component; 151 is the positioning module; 152 is the inertial measurement module; 160 is the camera module; and 170 is the storage module. Detailed Implementation

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] See attached document Figure 1 The present invention provides an intelligent unmanned aerial vehicle (UAV) swarm scheduling system, which may include a work swarm composed of multiple UAVs; the system is composed of basic hardware and the means of implementing the basic hardware.

[0033] Each drone is equipped with a flight control module 110, a computing module 120, a battery module 130, and a communication module 140. The drone 100 is equipped with a sensor component 150, a camera module 160, and a storage module 170. The sensing component 150 includes a positioning module 151 and an inertial measurement module 152; Flight control module 110 establishes a physical connection with computing module 120 via internal communication bus; The battery module 130, communication module 140 and sensing component 150 are respectively connected to the computing module 120 via a data bus; The camera module 160 and the storage module 170 are connected to the computing module 120 via a data interface.

[0034] See attached document Figure 2 This invention provides an intelligent unmanned aerial vehicle (UAV) swarm scheduling method, based on the aforementioned hardware system, comprising the following steps: S100, during the system initialization phase, the system generates a task token data packet containing a flight path coordinate sequence and a network routing forwarding table. The communication module 140 receives the task token data packet and transmits it to the computing module 120. The computing module 120 runs finite state machine logic to switch the UAV 100 with the task token data packet loaded to the main inspection node state, and the UAV 100 without the task token data packet loaded enters the escort relay node state and establishes a linear following topology network. S200, when performing flight operations, the battery module 130 reads the remaining power integral data, the sensing component 150 outputs three-dimensional spatial coordinate data and apparent wind field vector data, the calculation module 120 extracts the above data to calculate and output kinematic energy gradient parameters, and instructs the communication module 140 to broadcast the kinematic energy gradient parameters to the adjacent UAV 100. S300, the computing module 120 in the escort relay node state receives the kinematic energy gradient parameters broadcast by the forward node, establishes a sliding time window to calculate the attenuation slope data, and transmits the attenuation slope data to the flight control module 110. The flight control module 110 substitutes the attenuation slope data into the position control closed loop and updates the target following distance expectation value to adjust the physical distance between the UAVs 100. S400, the calculation module 120 in the main inspection node state compares its own and the kinematic energy gradient parameters of the rear node. When the comparison difference reaches the fault tolerance threshold, the switchover logic is triggered. The command communication module 140 transmits the mission token data packet to the rear UAV 100. The original calculation module 120 that handed over the mission token data packet issues a stop maneuver command to the flight control module 110 and switches the UAV 100 to the static communication relay node state. The rear calculation module 120 that received the mission token data packet changes the UAV 100 to the main inspection node state and takes over the flight control authority. In S500, the computing module 120 in the main inspection node state monitors the backhaul routing link through the communication module 140. When the backhaul routing link is disconnected, the computing module 120 sends a speed attenuation command to the flight control module 110, cuts off the continuous video stream sampling process of the camera module 160, and instructs the camera module 160 to perform a keyframe acquisition operation triggered by spatial coordinates, writing the acquired image data into the storage module 170. When the UAV 100 switches to the static communication relay node state and is in a physical hovering attitude, the computing module 120 removes the network bandwidth limitation, retrieves the image data in the storage module 170, and instructs the communication module 140 to perform data transmission. In S600, each node's computing module 120 continuously monitors its own energy status and mission progress. When the energy of a single unit reaches the lower limit set for safe return, the computing module 120 clears the network routing table, the flight control module 110 executes the underlying autonomous return procedure, and the ground control terminal dispatches the backup UAV 100 to fly to the missing relay spatial coordinates to access the network and complete the cluster topology link replacement. When the main inspection node completes the traversal of all route coordinates, it triggers the cluster topology disbandment, and each node executes the autonomous return landing procedure.

[0035] See attached document Figure 3 In specific implementations, step S100 provided by the present invention may include the following steps: S110, the system generates a task token data packet containing a specific data structure. The task token data packet is generated by the ground control terminal. Its underlying data structure includes a header identifier, a flight path coordinate sequence group, and a network routing table. The flight path coordinate sequence group consists of a series of waypoints with 3D geographic location markers. The network routing table contains the media access control address mapping relationship of the communication modules 140 of all UAVs 100 in the operation cluster. In this embodiment, the specific network addressing and underlying encapsulation protocol of the media access control address mapping can be completed using the standard wireless ad hoc network addressing specification. This communication encoding format is a well-known technology in the field.

[0036] S120, the communication module 140 receives the mission token data packet and triggers a finite state machine logic switch. The communication node of the ground control end unicasts the mission token data packet to the UAV 100 at the designated first position in the formation. After receiving the data packet, the communication module 140 of the UAV 100 unpacks it through the internal data bus and loads it into the memory space of the computing module 120. The computing module 120 embeds a finite state machine logic program. After power-on reset, the finite state machine continuously polls the status register in the memory space. When it detects that the mission token data packet with a complete check code has been successfully loaded into the memory, the computing module 120 modifies the flag bit of the status register and sets the operating mode of the UAV 100 to the master inspection node state.

[0037] The computing modules 120 of the other drones 100 in the work cluster that have not received the task token data packet do not detect a valid frame header checksum within the timeout waiting window. In this embodiment, the timeout waiting window ranges from 3 seconds to 10 seconds. The computing modules 120 execute the default preset branch logic of the finite state machine and directly set the operating mode of the drone 100 to the accompanying relay node state, forming a state machine switching mechanism based on data packet detection triggering.

[0038] S130, establish a linear following topology network. Each computing module 120 in the escort relay node state reads its own topology sequence number in the cluster. The topology sequence number is incremented and allocated by the communication module 140 according to the order of radio frequency network access time during the system power-on network handshake phase and is fixed in the local register. At the same time, the computing module 120 in the main inspection node state instructs the communication module 140 to broadcast a network handshake packet containing its own initial 3D spatial coordinate vector and the unit direction vector along the predetermined operation route to the cluster. The calculation module 120 in the escort relay node state receives the coordinate vector and the unit direction vector, and retrieves the preset initial spatial reference distance parameter. It calculates the initial expected physical occupancy coordinates in combination with the topology sequence number. This calculation process combines the communication radio frequency link hierarchy between the UAVs 100 to transform the abstract network hierarchy hops into a fixed length extension of the physical space dimension, so that the subsequently deployed UAV cluster is arranged linearly in the air physical distribution.

[0039] The specific formula for calculating the position of the linear following topology network is as follows, based on the initial spatial occupancy of each node. ; In the formula, For the first in the sequence The target 3D spatial coordinate vector of a companion relay node; The current 3D spatial coordinate vector of the main inspection node; This is the integer sequence number of the accompanying relay node in the network topology sequence, and its value is a positive integer; The initial spatial reference distance scalar is preset. In this embodiment, the value of the initial spatial reference distance scalar ranges from 5 meters to 20 meters. It is the unit direction vector along the predetermined operation route.

[0040] After the computing module 120, which is in the accompanying relay node state, obtains the corresponding 3D target coordinate vector according to the above solution model, it generates the desired position offset command and sends it to the flight control module 110. The flight control module 110 parses the desired position offset command and drives the underlying brushless motor flight actuator, so that all UAVs 100 that have entered the accompanying relay node state can sequentially perform fixed-distance hovering or synchronous flight along the flight path of the main inspection node, thus completing the physical construction of the linear following topology network.

[0041] In this embodiment, step S200 provided by the present invention may include the following steps in specific implementation: S210, the computing module 120 performs time synchronization and acquisition of multi-source heterogeneous sensor data. During flight operations, the battery module 130 continuously reads the discharge current and voltage data of the battery pack inside the UAV 100 and generates residual power integral data. The positioning module 151 in the sensing component 150 outputs 3D spatial coordinate data including latitude, longitude and altitude. The inertial measurement module 152 outputs apparent wind field vector data based on the aircraft attitude estimation. The computing module 120 adds absolute timestamps to the above heterogeneous data sources through the internal high-precision hardware clock bus. For timestamp deviations caused by inconsistent sensor sampling frequencies, the computing module 120 uses a zero-order hold or linear interpolation algorithm to resample the discrete data and puts the physical sensor data into the data queue of the same time segment for alignment and matching.

[0042] S220, the calculation module 120 extracts the aligned data queue and performs physical modeling and calculation of the kinematic energy gradient. The kinematic energy gradient comprehensively quantifies the mission redundancy of the UAV 100 in the current dynamic environment. This calculation process uses the physical spatial distance and environmental wind resistance coefficient of the denominator as the attenuation weight factor of energy consumption. When facing headwinds or when the distance to the return point is far, the value of the denominator increases, causing the overall gradient parameter to show an accelerated attenuation trend, thereby mapping the environmental response capability of the physical body. The calculation module 120 retrieves the safe return energy margin constant and the aerodynamic drag gain coefficient preset by the model from the internal memory, and combines it with the currently extracted remaining power integral data, 3D spatial coordinate data and apparent wind field vector data, and substitutes them into the kinematic energy gradient algorithm model.

[0043] The specific formula of the kinematic energy gradient algorithm model built into the calculation module 120 is as follows: ; In the formula, In order to be in The kinematic energy gradient parameters calculated at time step; It is an absolute timestamp variable; In order to be in The drone's remaining battery power score data is read in real time; As a preset safe return-to-home energy margin constant, in this embodiment, the value of the safe return-to-home energy margin constant ranges from 15% to 25% of the total battery capacity; In order to be in The 3D Euclidean distance from the current position of the mech to the nearest effective counterattack point, calculated at each moment; The aerodynamic drag gain constant is a pre-calibrated aerodynamic drag gain constant. In this embodiment, the value of the aerodynamic drag gain constant ranges from 0.05 to 0.15. In order to be in The apparent wind field vector estimated at time; In order to be in The modulus dimension of the apparent wind field vector estimated at time.

[0044] S230, the calculation module 120 encapsulates the kinematic energy gradient parameters output by the solution and triggers the state heartbeat broadcast mechanism. The calculation module 120 packages the current absolute timestamp, the current 3D spatial coordinate data of the UAV 100 and the corresponding kinematic energy gradient parameters into a custom application layer short frame message and sends it to the data transmission buffer of the communication module 140. The communication module 140 calls the underlying Media Access Control layer protocol and dynamically calculates the non-collision broadcast frequency by combining the current relative flight speed of the UAV 100 and the upper limit of the available wireless channel bandwidth. The wireless channel is periodically occupied by the broadcast frequency to broadcast short frame messages to neighboring UAVs 100 within the single-hop radio frequency communication range. In this embodiment, a carrier sense multiple access mechanism can be used to realize the channel contention and data packet broadcast of the Media Access Control layer protocol. This mechanism is a well-known technology in the field.

[0045] In this embodiment, step S300 provided by the present invention may include the following steps in specific implementation: S310, the computing module 120 in the accompanying relay node state receives the kinematic energy gradient parameters broadcast by the forward node, and establishes a sliding time window of fixed length. The computing module 120 extracts the discrete kinematic energy gradient parameters within the window, uses the least squares method to linearly fit the timestamp and energy gradient, outputs the energy decay slope data of the current time section, and quantifies the current comprehensive energy consumption rate of the forward node. S320, the calculation module 120 transmits the calculated attenuation slope data to the flight control module 110. The flight control module 110 has a built-in dynamic spacing mapping function, which substitutes the attenuation slope data into the position control closed loop. When the attenuation slope increases, that is, when the energy consumption ahead increases, the target following distance is shortened to the expected value. Conversely, the preset safe following distance is maintained, thereby adaptively adjusting the physical spacing between the UAVs 100.

[0046] See attached document Figure 4In specific implementations, step S400 provided by the present invention may include the following steps: S410, the computing module 120 in the main inspection node state periodically listens to the kinematic energy gradient parameters broadcast by the first adjacent escort relay node. The computing module 120 extracts its own current parameters and the parameters of the rear node through an internal comparator and performs difference calculation. This comparison logic quantifies and separates the power of the two adjacent UAVs 100 in physical space from environmental constraints. The calculated difference represents the additional energy attenuation potential of the lead node compared to the escort node, so that the cluster triggers role handover based on the actual physical loss deviation rather than a fixed time. The computing module 120 continuously performs logical verification between the comparison difference and the preset fault tolerance threshold in memory. When the difference increases positively and reaches or exceeds the fault tolerance threshold, the hardware interrupt signal inside the computing module 120 is triggered and the asynchronous fault tolerance logic closed loop is invoked.

[0047] The specific formula for calculating the kinematic energy gradient comparison difference and its relationship with triggering a switch is as follows: ; In the formula, Indicates in The kinematic energy gradient calculated at each time step is compared with the difference; Indicates in The kinematic energy gradient parameters received at each time from the rear relay node; Indicates in The kinematic energy gradient parameters calculated by the main inspection node itself at any given moment; This represents the preset fault tolerance threshold constant for the troop rotation. The system presets that when the comparison difference is greater than or equal to this fault tolerance threshold constant, a state switching judgment will be executed. The specific value of this constant is combined with the wind load limit calibration of the UAV 100 specific airframe platform and its value range is between 0.15 and 0.25.

[0048] S420, after triggering the asynchronous defense switching logic, the computing module 120 of the original master inspection node reassembles the flight path coordinate sequence and network routing forwarding table mounted in memory into a task token data packet in handshake frame format. The computing module 120 instructs the communication module 140 to send the task token data packet to the communication module 140 of the corresponding UAV 100 behind it via unicast according to the forwarding table address mapping. After receiving and completing the redundancy cyclic code verification, the rear node sends an acknowledgment message back to the front end through its own communication module 140. If the original master inspection node does not detect an acknowledgment message within the set waiting clock cycle, it executes... In this embodiment, the waiting clock period ranges from 50 milliseconds to 200 milliseconds for channel backoff and message retransmission operations. When the number of retransmissions reaches the preset maximum retransmission limit, which ranges from 3 to 5 times in this embodiment, the calculation module 120 extracts the next-level node in the network routing table and performs a fault-tolerant procedure to redirect and send the task token data packet across levels. In this embodiment, the standard wireless LAN underlying handshake protocol can be used to perform the backhaul of the acknowledgment message and channel retransmission operations. This protocol mechanism is a well-known technology in the field.

[0049] S430, after receiving a complete acknowledgment message or hearing the topology update frame broadcast by the rear node, the original computing module 120 modifies the status register flag bit to change the operating mode of its UAV 100 to the static communication relay node state. Simultaneously, the original computing module 120 sends a stop maneuver command to the flight control module 110 of the physical layer. The flight control module 110 of the original main inspection node blocks the existing waypoint tracking position control loop and switches to the fixed-point hovering closed-loop control program to drive the underlying flight actuator to deviate from the predetermined operation route to the preset safe avoidance position, and maintains fixed-point hovering in 3D spatial coordinates at the safe avoidance position. After unpacking the task token data packet, the computing module 120 of the rear node changes its operating mode to the main inspection node state and instructs the communication module 140 to broadcast the topology update frame to the other accompanying relay nodes in the cluster to trigger the update of the relative topology sequence number. After receiving the topology update frame, the other accompanying relay nodes extract the original topology sequence number of the new main inspection node in the frame. For accompanying relay nodes whose own topology sequence number is greater than the original topology sequence number, they subtract the original topology sequence number from their own topology sequence number to maintain the physical continuity of the linear following topology network. The flight control module 110 of the rear node takes over the subsequent waypoint coordinate sequence and drives the actuator to maneuver along the predetermined operation route. The operation cluster relies on the above-mentioned channel handshake and control loop switching timing logic to realize decentralized autonomous task role handover.

[0050] See attached document Figure 5 In specific implementations, step S500 provided by the present invention may include the following steps: S510, the computing module 120 in the main inspection node state monitors the global backhaul routing link to the ground control terminal through the communication module 140. The computing module 120 periodically sends link probe messages to the ground control terminal. The probe messages are forwarded via multi-hop routing through the back topology network. The round-trip delay time of the probe messages and the packet loss rate parameters within a unit time period are recorded. The computing module 120 performs a logical comparison between the real-time extracted round-trip delay time and packet loss rate parameters and the corresponding communication interruption thresholds that are internally fixed. In this embodiment, the communication interruption threshold includes an upper limit for delay time and an upper limit for packet loss rate ranging from 70% to 90%. When the packet loss rate parameter exceeds 80% for three consecutive detection cycles, the round-trip delay time exceeds the upper limit of the delay time, or no response is received, a hardware interrupt signal for link disconnection is triggered. The calculation module 120 extracts the status word of the underlying register of the communication module 140 to quantify the validity of application layer data transmission and performs logical determination of the backhaul route link disconnection status.

[0051] S520, after the backhaul routing link is disconnected, the calculation module 120 sends an instruction set containing the speed attenuation coefficient to the flight control module 110. In this embodiment, the speed attenuation coefficient ranges from 0.5 to 0.8. The flight control module 110 parses the instruction set and limits the output power of the underlying motor to within 60% of the rated cruise power. The calculation module 120 cuts off the continuous video stream sampling process of the camera module 160 through the internal bus, stores the 3D spatial coordinates and absolute timestamp at the moment of disconnection as initial reference values ​​in memory, and instructs the camera module 160 to perform keyframe acquisition operation based on spatial coordinates. The calculation module 120 extracts the current 3D spatial coordinates output by the positioning module 151 and performs a difference operation with the recorded coordinates at the last acquisition time. When the displacement increment reaches the set step size threshold or the time span from the last sampling reaches the forced trigger time period, the camera module 160's shutter action is triggered and the image data is written to the storage module 170.

[0052] The specific formula for triggering keyframe acquisition with the added time fallback mechanism is as follows: ; ; In the formula, In order to be in The relative displacement increment parameter of the body calculated at any time; For absolute time cutoff variables; In order to be in The X-axis component value of the current 3D spatial coordinate vector output at any given time; In order to be in The Y-axis component value of the current 3D spatial coordinate vector output at any given time; In order to be in The Z-axis component value of the current 3D spatial coordinate vector output at any given time; The X-axis component value of the historical 3D spatial coordinate vector when the ignition gate was last triggered; The Y-axis component value of the historical 3D spatial coordinate vector when the ignition gate was last triggered; The Z-axis component value of the historical 3D spatial coordinate vector when the ignition gate was last triggered; In order to be in The relative displacement increment parameter of the body calculated at any time; For absolute time cutoff variables; The preset spatial sampling step size threshold constant is used. In this embodiment, the value of the spatial sampling step size threshold constant ranges from 10 meters to 50 meters. This is the absolute timestamp of the last time the fire gate action was triggered; The forced trigger time period is a preset constant. In this embodiment, the value of the forced trigger time period constant ranges from 5 seconds to 15 seconds.

[0053] S530, after the original main inspection node triggers the switchover logic in step S400 above, when the UAV 100 to which it is located switches to the static communication relay node state and is in a physical hovering attitude, it executes the asymmetric data clearing logic. The calculation module 120 extracts the attitude stability flag state inside the flight control module 110 for verification. After confirming that the airframe is in a static hovering attitude, the system uses the static state of the physical position to eliminate Doppler frequency shift and multipath fading channel interference in dynamic environments. After the calculation module 120 verifies that the backhaul route link has been reconnected by sending a link probe message to the ground control terminal and receiving an acknowledgment response, the calculation module 120 removes the preset transmission bandwidth limit in the network protocol stack of the communication module 140, retrieves the lingering image data that has not been transmitted back from the storage module 170, and instructs the communication module 140 to perform full-bandwidth burst data transmission using the current high signal-to-noise ratio radio frequency physical environment. In this embodiment, the packet aggregation transmission mechanism of the media access control layer can be used to perform the full-bandwidth burst data transmission operation, which is a well-known technology in the field.

[0054] See attached document Figure 6 In specific implementations, step S600 provided by the present invention may include the following steps: S610, the calculation module 120 reads the absolute remaining power integral data output by the battery module 130. If the calculation module 120 is in the main inspection node state, it synchronously reads the waypoint traversal status flag in the task token data packet mounted in memory. The calculation module 120 uses its internal high-precision arithmetic unit to perform difference calculation between the absolute remaining power integral data and the fixed safe return energy margin constant in memory to generate energy alarm margin parameters. The calculation module 120 uses these energy alarm margin parameters and, when in the main inspection node state, combines the waypoint traversal status flag to perform the logical determination of the current task cluster life cycle status. When the energy alarm margin parameter is less than the system's preset landing trigger threshold, in this embodiment, the landing trigger threshold ranges from 5% to 10% of the total energy, and the current calculation module 120 is in the main inspection node state, the calculation module 120 forcibly calls the task token data packet handover logic of the above step S420 to transmit the task token data packet to the first rear escort relay node. After completing the task token handover or being in a non-master inspection node state, the computing module 120 does not execute the hovering command to switch the static communication relay node. Instead, it directly modifies the single-machine offline status bit of the status register through the hardware bus, triggering a single-machine offline return hardware interrupt request and activating the underlying multi-level return recovery logic closed loop. If the computing module 120 in the master inspection node state determines that the waypoint traversal status flag shows that all waypoint coordinates have been visited and traversed, the computing module 120 modifies the global task status bit of the status register through the hardware bus and instructs the communication module 140 to broadcast a control message containing the cluster disbandment flag to all nodes in the job cluster, triggering a cluster topology disbandment hardware interrupt request and activating the underlying multi-level return recovery logic closed loop. After receiving the control message, the other nodes in the job cluster synchronously modify the global task status bit in the status register, trigger the cluster topology disband hardware interrupt request, and activate the underlying multi-level return and recovery logic closed loop.

[0055] S620, the calculation module 120 extracts the current 3D spatial coordinates output by the positioning module 151 and the 3-axis attitude angle data output by the inertial measurement module 152. The calculation module 120 retrieves the elevation digital map data stored in the non-volatile memory and combines it with the current 3D spatial coordinates to calculate the return landing trajectory vector using a heuristic search algorithm. This optimization model takes the fuselage's underlying dynamic constraints, the external environment wind field vector, and the terrain elevation difference as independent penalty terms. The cross product of the wind field and the heading physically represents the lateral wind resistance disturbance perpendicular to the flight direction. This disturbance will induce the flight control to adjust the attitude at high frequency, which will consume a lot of power. The calculation module 120 performs traversal calculations and ascending sorting on multiple pre-loaded candidate ground recovery points based on the output estimate of the multi-level landing cost model, and selects the candidate ground recovery point with the smallest comprehensive landing cost function value as the best landing anchor point for the current UAV 100.

[0056] The objective function formula for the multi-level landing cost model with terrain penalty term is as follows: ; In the formula, For the first The comprehensive landing cost function value of one alternative ground recovery point; This is the traversal index number of the candidate ground recycling points, and its value is a positive integer; The distance penalty weighting coefficient is a preset value. In this embodiment, the value of the distance penalty weighting coefficient ranges from 0.4 to 0.6. The current 3D spatial coordinate vector; For the first The 3D spatial coordinate vectors of the candidate ground recycling points; The crosswind disturbance penalty weighting coefficient is a preset value. In this embodiment, the crosswind disturbance penalty weighting coefficient ranges from 0.3 to 0.5. For continuous integration time variables; To estimate the arrival time of the first Theoretical flight time parameters required for each alternative ground recovery point; In order to be in The apparent wind field vector estimated at time; In order to be in At any moment, the machine body is facing the first Unit flight direction vector of one candidate ground recovery point; For time integral infinitesimal elements; The preset ground formation undulation penalty weight coefficient is used. In this embodiment, the value range of the ground formation undulation penalty weight coefficient is 0.1 to 0.3. Towards the first The maximum relative altitude difference of the routes of the alternative ground recovery points on the digital map projection trajectory.

[0057] S630, the computing module 120 removes the closed-loop command that limited the motor output power due to the disconnection of the backhaul routing link, restores the full power output permission of the underlying flight control, and sends the coordinates of the best landing anchor point to the flight control module 110. The flight control module 110 drives the flight actuator to guide the UAV 100 to leave the current network and enter an independent landing route. The computing module 120 synchronizes the command communication module 140 to broadcast a node disconnection message containing a disconnection identifier. After receiving the message, the adjacent nodes update the network routing table and send the node disconnection message back to the ground control terminal via a multi-hop route. If the current state is determined to be a single drone disconnected from the network and returning home, after the ground control terminal receives the node disconnection message, it dispatches the backup drone 100 to fly to the dynamically corresponding expected physical location coordinates or static hovering spatial coordinates according to the topology attributes of the missing node and connect to the network to complete the cluster topology link replacement; if the current state is determined to be a cluster topology disbandment, the dispatching task of the backup drone is not executed, and each node executes the landing procedure in sequence. When the sensing component 150 detects ground stress, the calculation module 120 cuts off the power supply to all peripherals except the communication module 140. In this embodiment, the advanced configuration and power interface specification can be used to control the hardware to enter a low-power standby mode. This sleep control specification is a well-known technology in the field.

[0058] Application Examples: Please refer to the appendix. Figure 7 To better understand the technical solution of this invention, the following explanation uses the application scenario of high-altitude mountain power transmission line inspection as an example: The operational cluster consists of 5 multi-rotor UAVs. The ground control terminal generates a mission token data packet containing a 10-kilometer cruise trajectory and 50 waypoints and sends it via unicast to UAV number 1. UAV number 1 takes off as the main inspection node and travels along the designated route. UAVs numbered 2 to 5 enter the escort relay node state and sequentially expand to establish a linear following topology network with an initial expected physical occupancy coordinate spacing of 15 meters. The computing modules of each UAV perform real-time time synchronization and align absolute timestamps. The sensing components continuously output spatial coordinates and wind field vector data.

[0059] The cluster enters the canyon area to perform flight operations. The main inspection node's computing module continuously compares its own kinematic energy gradient parameters with those of the rear-flying relay node numbered 2. When the difference between the comparison values ​​increases positively and exceeds the 0.2 fault tolerance threshold for switching, an internal hardware interrupt signal is triggered, and the asynchronous switching logic is invoked. The main inspection node directs the task token data packet to the drone numbered 2 via unicast and switches to the fixed-point hovering control loop, changing to the static communication relay node state. After the drone numbered 2 completes the token unpacking, its state changes to the main inspection node state, taking over the subsequent inspection waypoints.

[0060] The main inspection node detected that the packet loss rate of the backhaul routing link reached 85% for three consecutive detection cycles. The calculation module determined that the backhaul routing link was disconnected and sent a speed attenuation coefficient of 0.6 to the flight control module. The flight control module limited the motor power to within 60% of the rated power. The calculation module cut off the continuous video stream sampling process and instructed the camera module to perform key frame image acquisition based on a 30-meter step threshold. When the defense switching logic was triggered again and the UAV became a static communication relay node and maintained a physical hovering state, the calculation module confirmed the link connection through probe messages and lifted the network bandwidth limitation. The communication module used the current radio frequency physical environment to perform full-bandwidth burst data transmission and sent back the image data that was stuck during the network outage to the ground control terminal.

[0061] When the single-unit energy alarm margin parameter of static communication relay node No. 1 drops to the landing trigger threshold of 8% of the total energy, the node triggers a single-unit disconnection and return-to-home interruption request. The calculation module combines the elevation digital map and heuristic search algorithm to calculate the comprehensive landing cost function value, selects the candidate ground recovery point with the smallest comprehensive landing cost function value as the optimal landing anchor point, enters the independent landing route, and broadcasts the node disconnection message to the network. After receiving the node disconnection message, the ground control terminal dispatches the backup UAV No. 6 to fly to the missing hovering space coordinate position to access the network topology. After the main inspection node completes the traversal of all route coordinates, it triggers the cluster topology disbandment and autonomous landing procedure.

[0062] The experiment used a physical platform with 6 multi-rotor UAVs to conduct off-site terrain verification tests and output simulation visualization plots containing three-dimensional spatial waypoint distribution and energy consumption curve features. The control group adopted a fixed time interval rotation scheduling strategy and a return strategy based on communication signal strength. The experimental group adopted an asynchronous rotation logic based on kinematic energy gradient and a scheduling mechanism that integrates a multi-level landing cost model. The experiment performed 5 complete inspection cycles on a 15-kilometer test route with the same wind field disturbance model and undulating elevation profile, and recorded the system task completion rate and system topology reconstruction time data.

[0063] The experimental results data were extracted from the non-volatile storage logs of the computing module at the test node and generated through offline statistical mean filtering. The test physical environment was set to an average wind speed of 6 m / s with a gust limit of 10 m / s. The physical units of the test groups were configured with the same batch of initial battery modules, communication RF antennas and camera payloads. The quantitative comparison test data corresponding to different scheduling strategies are shown in the table below.

[0064] Based on the test data table, the following comparative conclusions were drawn: the minimum redundancy power of a single machine in the control group was 3.2%, while the minimum redundancy power of a single machine in the experimental group was 9.8% based on kinematic energy gradient parameters and fault tolerance thresholds for triggering the switchover logic and executing task role handover. The task completion rate of the experimental group was 98.2%, and the data retention rate during network outages was 1.5% due to the use of spatial coordinate triggering mechanism and full-bandwidth burst data transmission mechanism.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent unmanned aerial vehicle (UAV) swarm scheduling system, characterized in that, The system comprises a cluster of multiple drones. Each drone is equipped with a flight control module, a computing module, a battery module, and a communication module. Each drone carries sensing components, a camera module, and a storage module. The sensing components include a positioning module and an inertial measurement module. The flight control module establishes a physical connection with the computing module via an internal communication bus. The battery module, communication module, and sensing components are connected to the computing module via data buses. The camera module and storage module are connected to the computing module via data interfaces. Based on the above hardware structure, the system generates a task token data packet containing a flight path coordinate sequence and a network routing table. The communication module is configured to receive the task token data packet and transmit it to the computing module to trigger state machine logic to complete node scheduling.

2. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 1, characterized in that, The computing module embeds a finite state machine logic program and a state register. During the system initialization phase, when the computing module detects that the task token data packet with a complete verification code has been successfully loaded into memory, it modifies the flag bit of the state register and sets the operating mode of the UAV to the main inspection node state. The computing module of a drone that has not received the mission token data packet executes the default preset branch logic of the finite state machine within the timeout waiting window, and sets the operating mode of the drone to the escort relay node state.

3. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 2, characterized in that, The computing module in the accompanying relay node state reads its own topology sequence number in the cluster and receives the initial three-dimensional spatial coordinate vector and the unit direction vector along the predetermined operation route broadcast by the computing module in the main inspection node state. The computing module in the accompanying relay node state retrieves the preset initial spatial reference distance parameters, calculates the initial expected physical occupancy coordinates in combination with the topology sequence number, generates the expected position offset command and sends it to the flight control module to drive the UAV to deploy, and completes the physical construction of the linear following topology network.

4. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 2, characterized in that, During flight operations, the battery module reads the remaining battery power integral data of the UAV's internal battery pack, and the sensing components output three-dimensional spatial coordinate data and apparent wind field vector data. The calculation module aligns the heterogeneous sensor data using absolute timestamps and calculates and outputs kinematic energy gradient parameters by using the physical spatial distance and environmental drag coefficient of the denominator as attenuation weighting factors for energy consumption. The module then instructs the communication module to call the Media Access Control layer protocol to locally broadcast the kinematic energy gradient parameters to neighboring UAVs.

5. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 4, characterized in that, The computing module in the accompanying relay node state receives the kinematic energy gradient parameters broadcast by the preceding node, establishes a sliding time window, and uses the least squares method to linearly fit the timestamp and energy gradient, outputting the attenuation slope data of the current time segment; the computing module transmits the attenuation slope data to the flight control module, and the flight control module substitutes the attenuation slope data into the position control closed loop to update the target following distance expectation value, so as to adaptively adjust the physical distance between the UAVs.

6. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 4, characterized in that, The calculation module in the main inspection node state periodically extracts its own kinematic energy gradient parameters and the kinematic energy gradient parameters broadcast by the adjacent follower relay node and performs difference calculation. When the comparison difference increases positively and reaches the fault tolerance threshold for switching, the switching logic is triggered. The command communication module sends the mission token data packet to the rear UAV via unicast. The original calculation module that handed over the mission token data packet issues a stop maneuver command to the flight control module, switches the UAV to the static communication relay node state and enters the fixed-point hovering control program. The rear calculation module that received the mission token data packet changes its operating mode to the main inspection node state and takes over the subsequent waypoint coordinate sequence.

7. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 6, characterized in that, The computing module in the main inspection node state monitors the global backhaul routing link to the ground control terminal through the communication module. When it is determined that the backhaul routing link is disconnected, the computing module sends an instruction set containing a speed attenuation coefficient to the flight control module to limit the output power of the underlying motor, cut off the continuous video stream sampling process of the camera module, and when the relative displacement increment reaches the set spatial sampling step size threshold or the time span reaches the forced trigger time period, it instructs the camera module to perform a key frame acquisition operation based on spatial coordinates and write the acquired image data into the storage module.

8. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 7, characterized in that, When the original main inspection node triggers the switchover logic to switch to the static communication relay node state and verifies that the aircraft is in a stationary hovering attitude, the calculation module verifies that the backhaul route link is reconnected by sending a link probe message to the ground control terminal, thereby removing the preset transmission bandwidth limit in the network protocol stack of the communication module, and retrieving the lingering image data that has not been returned from the storage module, instructing the communication module to perform full-bandwidth burst data transmission using the current physical environment.

9. The intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in claim 2, characterized in that, When the calculation module determines that its own energy alarm margin parameter is less than the system's preset landing trigger threshold, it modifies the single-machine disconnection status bit of the status register to activate the underlying multi-level return and recovery logic closed loop. The calculation module extracts the current three-dimensional spatial coordinates and three-axis attitude angle data, uses a heuristic search algorithm, and calculates the comprehensive landing cost function value of the multi-level landing cost model by taking the underlying dynamic constraints of the fuselage, the external environmental wind field vector, and the terrain elevation difference as independent penalty terms. It selects the candidate ground recovery point with the smallest comprehensive landing cost function value as the best landing anchor point for the current UAV, and guides the UAV to leave the current network and enter an independent landing route.

10. An application of the intelligent unmanned aerial vehicle (UAV) swarm scheduling system as described in any one of claims 1 to 9, characterized in that, The system is applied to the inspection of power transmission lines in high-altitude mountainous areas. In the complex wind field conditions of the inspection scenario, multiple drones in the operation cluster construct a linear following topology network based on the node identity of the node carrying the task token data packet. They use the potential energy comparison difference caused by the real-time calculated kinematic energy gradient parameters in the strong wind area to trigger decentralized autonomous task role handover. During the disconnection and restoration of the backhaul route link, they perform adaptive keyframe image acquisition and full-bandwidth burst data clearing and backhaul. They also combine the external environment wind field and terrain undulations to autonomously optimize landing to the selected optimal landing anchor point.

Images