Distributed flight control method for unmanned aerial vehicle cluster

Abstract

The invention discloses an unmanned aerial vehicle cluster distributed flight control method, and belongs to the technical field of unmanned aerial vehicle cluster control, and the method comprises the steps: distributing unique identity codes for all unmanned aerial vehicles in a cluster, employing an Ad-Hoc protocol to construct a centerless peer-to-peer communication network, collecting the state and environment data through an airborne sensor, carrying out the denoising processing of the original data, and carrying out the collection of all the unmanned aerial vehicles. Effective update data is only broadcasted to neighbor nodes in a communication range, the effective neighbor nodes are screened to generate a dynamic undirected topological graph, a coupling type consistency control law of'cooperative variable difference weighted summation of the local aircraft and the neighbor nodes' and a corresponding error function are constructed, and the flight control quantity of the local aircraft is solved. According to the control method, a centerless, fully-distributed and master-slave-free control framework is established, it is guaranteed that the unmanned aerial vehicle cluster does not depend on a ground station or a pilot, and the technical problems that a traditional unmanned aerial vehicle cluster flight control technology is low in cooperation precision, insufficient in survivability, free of self-healing of faults, conflicting in obstacle avoidance and cooperation and weak in large-scale expansion capacity are solved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, and more specifically, to a distributed flight control method for UAV swarms. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarm flight control technology is a core research direction in the fields of multi-agent collaboration, autonomous navigation, and distributed control, and is widely used in military and civilian scenarios such as formation performances, emergency search and rescue, geographic surveying, power line inspection, and airspace operations. As swarm operations become more large-scale, complex, and autonomous, higher demands are placed on the collaborative accuracy, survivability, real-time response capabilities, and environmental adaptability of UAV swarms. Existing swarm flight control architectures and algorithms are no longer sufficient to meet the needs of practical engineering applications.

[0003] Current UAV swarm flight control technology is mainly divided into three technical routes: centralized control, master-slave control, and traditional distributed control. Among them, traditional distributed swarm control abandons the central and master-slave architecture, and each UAV achieves coordinated flight only through local neighbor information exchange, which has become the current research mainstream. However, existing distributed solutions still have several key technical defects: First, the collaborative control is limited to a single dimension, designing consensus algorithms only for single variables such as position or velocity, without incorporating attitude variables into the coupled control system. This results in low cluster attitude synchronization accuracy and poor formation stability, making high-precision collaborative operations impossible. Second, the topology is statically fixed, failing to incorporate communication quality, relative distance, and remaining battery power for dynamic neighbor selection and topology updates. Changes in the relative position of the UAVs can easily lead to invalid interactions and control inaccuracies. Third, obstacle avoidance and collaborative control are disconnected. Traditional obstacle avoidance algorithms only focus on individual UAV safety, without introducing formation coordination constraints, making it highly susceptible to disrupting the cluster formation during obstacle avoidance. The system suffers from several drawbacks: First, it lacks a collaborative interruption mechanism, making it impossible to accurately determine faulty nodes through heartbeat verification and status data. Furthermore, it cannot quickly reconstruct communication and control links after a fault occurs, making it difficult to guarantee cluster continuity. Second, its single-machine computing resources are limited, with complex data fusion and control law calculation tasks all completed independently by a single machine. Under large-scale clusters, the computing load is too high, and real-time control cannot be guaranteed. Third, its aircraft compatibility is poor, supporting only homogeneous UAV cluster collaboration and lacking parameter adaptation mechanisms for heterogeneous aircraft such as multi-rotor, fixed-wing, and tiltrotor aircraft, thus limiting application scenarios.

[0004] In view of this, the present invention is proposed to solve the above-mentioned technical problems. Summary of the Invention

[0005] The purpose of this invention is to provide a distributed flight control method for unmanned aerial vehicle (UAV) swarms, in order to solve the technical problems of existing UAV swarm flight control technologies, such as insufficient survivability, low coordination accuracy, high communication load, lack of self-healing for faults, conflict between obstacle avoidance and coordination, low computing power utilization, poor heterogeneous adaptability, and weak large-scale expansion capability.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Distributed flight control methods for unmanned aerial vehicle (UAV) swarms include: Step 1: Assign a unique identification code to all drones in the cluster, build a decentralized peer-to-peer communication network using the Ad-Hoc protocol, complete sub-millisecond time synchronization and initial absolute position calibration, and autonomously establish an initial list of neighbor nodes; Step 2: Each drone collects its own status and environmental data through onboard sensors. After denoising the raw data, it broadcasts valid update data only to neighboring nodes within the communication range. It receives data from neighboring nodes and performs distributed fusion with its own data to form a local state consensus. Step 3: Calculate the node connection weight based on three indicators: communication quality, relative distance, and remaining power. Select effective neighbor nodes to generate a dynamic undirected topology graph. The topology update cycle is synchronized with the control cycle to provide a neighbor node basis for subsequent consistency control. Step 4: Using position, velocity, and attitude as cooperative control variables, and based on the neighbor node information in the dynamic topology generated in Step 3, construct a coupled consistency control law and corresponding error function of "weighted summation of the differences between the cooperative variables of the local machine and neighbor nodes", solve the local flight control quantity, and realize closed-loop constraint control of the cooperative error of the cluster position, velocity, and attitude. Step 5: Based on the environmental data from Step 2 and the location information of neighboring nodes in the dynamic undirected topology graph from Step 3, the improved artificial potential field method with cooperative potential field terms is used to correct the control quantity calculated in Step 4, so as to achieve inter-machine collision avoidance and environmental obstacle avoidance while maintaining cluster collaboration. Step 6: Perform fault determination through heartbeat message interaction and status data verification, identify faulty nodes, remove faulty nodes, and reconstruct the communication topology and control topology based on the breadth-first search algorithm to ensure cluster connectivity and control continuity. Step 7: Based on the remaining power collected in Step 2 and the node connection weights in Step 3, edge computing nodes are selected and high-complexity computing tasks are unloaded. Then, the control quantity corrected in Step 5 is converted into flight control execution commands to drive the UAV actuators to move. Steps 2 to 6 are repeated according to a fixed control cycle to form a closed-loop negative feedback control. Step 8: After receiving the mission completion instruction, the cluster exits the distributed collaborative control mode, and each UAV autonomously switches to return-to-home or hovering standby state, shutting down the self-organizing communication network.

[0007] Furthermore, in step 1, time synchronization adopts PPS pulse synchronization technology with a synchronization accuracy of ≤0.5ms; Initial position calibration is based on GPS / BeiDou dual-mode positioning, with a positioning accuracy of ≤1cm; The Ad-Hoc communication protocol supports node relay forwarding.

[0008] Furthermore, the sensing data in step 2 includes GPS / BeiDou positioning data, inertial measurement unit attitude angle data, lidar / visual obstacle distance data, remaining battery power, flight altitude, airspeed, and communication signal strength; The noise reduction process uses weighted mean filtering, with the weights dynamically allocated based on the sensor accuracy. Specifically, the weights for lidar data are 0.4, inertial measurement unit data are 0.3, GPS / BeiDou positioning data are 0.2, and communication signal strength data are 0.1. Valid update data is defined as abnormal data or status change data whose difference from the previous period's data exceeds a preset threshold. The threshold is set as follows: positioning data 0.5cm, attitude data 0.1°, speed data 0.1m / s, obstacle distance 0.2m, remaining battery power 1%, communication signal strength 2dBm, flight altitude 0.1m, and airspeed 0.05m / s.

[0009] Furthermore, the formula for calculating the node connection weight in step 3 is as follows: W_ij=α×S_ij+β×D_ij+γ×E_ij; Where α, β, and γ are weighting coefficients and α+β+γ=1, S_ij is the normalized value of the communication signal strength between UAV i and neighbor j, D_ij is the normalized value of the relative distance, and E_ij is the normalized value of the remaining battery power of neighbor j. The effective neighbor node determination thresholds are: communication RSSI ≥ -70dBm, relative distance ≤ 5m, and remaining power ≥ 20%. The topology update cycle is consistent with the control cycle in step 7.

[0010] Furthermore, the expression for the coupled consistency control law in step 4 is: u_i=k_p∑(p_j-p_i)+k_v∑(v_j-v_i)+k_θ∑(θ_j-θ_i); Where u_i is the total control quantity of the i-th UAV, and k_p, k_v, and k_θ are the position, velocity, and attitude coupling coefficients, respectively, with values ​​ranging from k_p ∈ [0.8, 1.2], k_v ∈ [0.3, 0.6], and k_θ ∈ [0.1, 0.3]. p_i, v_i, and θ_i are the position vector, velocity vector, and attitude angle vector of the i-th UAV in the local machine, respectively; p_j, v_j, and θ_j are the position vector, velocity vector, and attitude angle vector of the neighbor node j, respectively; and ∑ represents the summation of the data of all valid neighbor nodes.

[0011] Furthermore, the potential field function of the improved artificial potential field method in step 5 is: U_total = U_att + U_rep + U_coop; Where U_att is the target gravitational potential field, U_rep is the repulsive potential field, and U_coop is the cooperative potential field. The obstacle avoidance response time is ≤50ms, and the consistency control command is automatically restored after the collision risk is eliminated. U_att = 0.5 × k_att × ||p_d - p_i||², where k_att is the gravitational coefficient and p_d is the desired formation position; U_rep=0.5×k_rep×(1 / ||p_obst-p_i||-1 / R0)², where k_rep is the repulsion coefficient, p_obst is the location of the obstacle / neighbor node, and R0 is the safe distance threshold; U_coop=0.5×k_coop×||∑p_j / n-p_i||², where k_coop is the cooperation coefficient and n is the number of effective neighbor nodes; The general reference values ​​for the potential field coefficients in the improved artificial potential field method are k_att=5.0, k_rep=10.0, and k_coop=3.0. Multi-rotor UAVs maintain the reference values, fixed-wing UAVs have k_att=8.0 and k_rep=15.0, and tilt-rotor UAVs have k_att=6.5, k_rep=12.5, and k_coop=3.0.

[0012] Furthermore, in step 6, the sending period of the heartbeat message is consistent with the control period, and the fault determination adopts a dual threshold rule: if a neighbor node's heartbeat message is not received for three consecutive control periods, or if the status data transmitted by the node exceeds the physical reasonable range, it is determined to be a faulty node.

[0013] Furthermore, the edge computing node selection and task offloading strategy in step 7 is as follows: based on the remaining power collected in step 2 and the node connection weight in step 3, the top 20% of drones with the highest weight and ≥30% remaining power are selected as edge computing nodes, and high-complexity computing tasks within the cluster are allocated to the edge computing nodes, reducing the single-machine computing load by ≥40%; the control cycle is fixed at 20ms-50ms and dynamically adjusted according to the cluster size, with a control cycle of 20ms for a cluster of 10-50 drones and 50ms for a cluster of 51-100 drones.

[0014] Furthermore, the cluster size supports seamless expansion from 10 to 100 drones, adjusting the relative distance threshold in step 3 during expansion.

[0015] Furthermore, the control method is applicable to heterogeneous UAV swarms of multi-rotor, fixed-wing, and tilt-rotor aircraft. By adjusting the potential field coefficient in step 5 and the coupling coefficient in step 4, it can be adapted to different aircraft models, as detailed below: For the multi-rotor UAV, k_p=1.0, k_v=0.4, k_θ=0.2, k_att=5.0, k_rep=10.0; For the fixed-wing UAV, k_p=1.2, k_v=0.6, k_θ=0.3, k_att=8.0, k_rep=15.0; For tilt-rotor UAVs, k_p=1.1, k_v=0.5, k_θ=0.25, k_att=6.5, k_rep=12.5, and k_coop=3.0 for all models.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Fully distributed architecture with no central node or master-slave relationship. Each drone has independent decision-making capabilities. Single point of failure or communication interruption will not affect the overall operation of the cluster, significantly improving resilience.

[0017] 2. The three-variable coupled consistency control of position, velocity and attitude has high coordination accuracy, with position error ≤5cm and attitude synchronization error ≤0.5°, which is superior to existing single-dimensional coordination control methods.

[0018] 3. Dynamic topology maintenance and fault self-repair are organically combined, topology updates and control cycles are synchronized, and fault reconstruction time is ≤20ms, ensuring the collaborative continuity of the cluster when nodes change dynamically.

[0019] 4. Improve the artificial potential field method by introducing a cooperative potential field term to achieve seamless integration of obstacle avoidance and cooperative control, which not only ensures flight safety but also avoids formation breakup during obstacle avoidance.

[0020] 5. Edge computing offloading strategy reduces the computing power load of a single machine, supports large-scale cluster expansion, can be seamlessly adapted to 10-100 drones, and has low communication load, only transmitting valid update data, without redundant data interaction.

[0021] 6. By adjusting the coupling coefficient and potential field coefficient, it can be adapted to multi-rotor, fixed-wing, and tilt-rotor heterogeneous UAV swarms, with a wide range of applications and strong practicality. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation of the invention. Obviously, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings: Figure 1 This is a schematic diagram illustrating the principle architecture of the distributed flight control method for unmanned aerial vehicle (UAV) swarms provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] See Figure 1 As shown, the distributed flight control method for UAV swarms includes: Step 1: Assign a unique identification code to all drones in the cluster, build a decentralized peer-to-peer communication network using the Ad-Hoc protocol, complete sub-millisecond time synchronization and initial absolute position calibration, autonomously establish an initial list of neighbor nodes, establish communication links between nodes through decentralized self-organizing networking, ensure the timing consistency of data interaction through time synchronization, and provide a benchmark for collaborative control through initial position calibration. Step 2: Each UAV collects its own status and environmental data through onboard sensors. After denoising the raw data, it broadcasts valid update data only to neighboring nodes within the communication range. It receives neighboring node data and performs distributed fusion with its own data to form a local state consensus. The local state consensus formed after the distributed fusion of multi-source perception data is not only the basis for dynamic topology generation, but also the core input for consistency control law calculation and obstacle avoidance decision-making. Step 3: Calculate the node connection weight based on three indicators: communication quality, relative distance, and remaining battery power. Select effective neighbor nodes to generate a dynamic undirected topology graph. The topology update cycle is synchronized with the control cycle, providing a neighbor node basis for subsequent consistency control. The dynamic topology graph clarifies the effective neighbor node set for each UAV, ensuring that consistency control is based only on local neighbor information, while also providing a link basis for fault node identification and topology reconstruction. Step 4: Using position, velocity, and attitude as coordinated control variables, and based on the neighbor node information in the dynamic topology generated in Step 3, construct a coupled consistency control law and its corresponding error function that is a "weighted summation of the differences between the coordinated variables of the local machine and the neighbor nodes". Solve the local flight control variables to achieve closed-loop constraint control of the coordinated errors of the cluster's position, velocity, and attitude. The coupled consistency control law realizes the synchronous coordination of the three variables of position, velocity, and attitude, which is the key to the cluster maintaining its formation. Step 5: Based on the environmental data from Step 2 and the location information of neighboring nodes in the dynamic undirected topology graph from Step 3, the improved artificial potential field method with cooperative potential field terms is used to correct the control quantity calculated in Step 4. While maintaining the cluster's coordination, it completes inter-machine collision avoidance and environmental obstacle avoidance. By improving the artificial potential field method, obstacle avoidance and cooperative control are organically integrated, eliminating collision risks while maintaining formation stability. Step 6: Perform fault determination through heartbeat message interaction and status data verification, identify faulty nodes, remove faulty nodes, and reconstruct the communication topology and control topology based on the breadth-first search algorithm to ensure cluster connectivity and control continuity. Fault detection promptly identifies faulty nodes, and topology self-repair quickly reconstructs the control link to avoid the impact of faulty nodes on the overall UAV cluster. Step 7: Based on the remaining power collected in Step 2 and the node connection weights in Step 3, edge computing nodes are selected and high-complexity computing tasks are unloaded. Then, the control quantity corrected in Step 5 is converted into flight control execution commands to drive the UAV actuators. Steps 2 to 6 are repeated according to a fixed control cycle to form a closed-loop negative feedback control. The corrected control commands are output in real time, and each core step is iterated according to a fixed cycle to ensure the dynamic adjustment and stability of the UAV cluster status. Step 8: After receiving the task completion instruction, the cluster exits the distributed collaborative control mode. Each UAV autonomously switches to return-to-home or hovering standby state, shuts down the self-organizing communication network, and ensures the safe disbandment of the UAV cluster after the task is completed. Each UAV autonomously executes subsequent operations. The task completion instruction adopts a "relay feedback + broadcast confirmation" mechanism: 1. The completion instruction sent by the ground station is relayed by any UAV, and the relay node adds its own ID to the message; 2. After receiving the instruction, the UAV in the cluster sends confirmation messages to all valid neighbor nodes; 3. When a UAV receives confirmation messages from ≥50% of valid neighbor nodes, it is determined that the instruction transmission is complete and the disbandment process is initiated; if not enough confirmation messages are received within 3 control cycles, the UAV re-requests the relay node to broadcast the instruction.

[0025] In the above scheme, the control method establishes a decentralized, fully distributed, and masterless control architecture, ensuring that the UAV swarm does not depend on ground stations or navigators. It solves the technical problems of low collaborative accuracy, insufficient survivability, high communication load, lack of self-healing for faults, conflict between obstacle avoidance and collaboration, low computing power utilization, poor heterogeneous adaptability, and weak large-scale expansion capability of traditional UAV swarm flight control technology, making the entire flight process of the UAV swarm continuous and stable.

[0026] In some possible implementations, time synchronization in step 1 uses PPS pulse synchronization technology with a synchronization accuracy of ≤0.5ms; initial position calibration is based on GPS / BeiDou dual-mode positioning with a positioning accuracy of ≤1cm; the Ad-Hoc communication protocol supports node relay forwarding, and in environments with a communication distance of ≤1km and few obstructions, the network latency is ≤10ms and the communication packet loss rate is ≤1%. The relay forwarding of the Ad-Hoc protocol follows the rule of "hop count limit + high-weight node priority": 1. The maximum relay hop count is 3 hops to avoid communication latency and packet loss caused by multi-hop forwarding; 2. Forwarding nodes prioritize selecting effective neighbor nodes with a node connection weight W_ij≥0.5; 3. Each forwarded packet is marked with a hop count identifier, and packets exceeding 3 hops are discarded, controlling the communication packet loss rate to ≤1%.

[0027] In the above scheme, PPS pulses are used to achieve sub-millisecond time synchronization, GPS / BeiDou dual-mode positioning is used to achieve high-precision positioning, and Ad-Hoc centerless relay communication is used. The latency and packet loss rate indicators are limited, and parameter constraints are imposed on the UAV swarm and the communication infrastructure layer to ensure the real-time performance, reliability and timing consistency of the network. This solves the most basic problems of UAV swarm, such as clock asynchrony, inaccurate initial positioning, large communication latency and network unreliability. It provides a unified time and position reference for all subsequent data interaction, control calculation and topology update.

[0028] In some possible implementations, the sensing data in step 2 includes GPS / BeiDou positioning data, inertial measurement unit (IMU) attitude angle data, lidar / visual obstacle distance data, remaining battery power, flight altitude, airspeed, and communication signal strength. The noise reduction process uses weighted average filtering, with weights dynamically allocated based on sensor accuracy. LiDou data has a weight of 0.4, IMU data 0.3, GPS / BeiDou positioning data 0.2, and communication signal strength data 0.1. Valid update data is defined as abnormal data or status change data whose difference from the previous period's data exceeds a preset threshold. The threshold is set as follows: positioning data 0.5 cm, attitude data 0.1°, speed data 0.1 m / s, obstacle distance 0.2 m, remaining battery power 1%, communication signal strength 2 dBm, flight altitude 0.1 m, and airspeed 0.05 m / s.

[0029] Weighted mean filter weights (before normalization): LiDAR 0.4, IMU 0.3, GPS / BeiDou 0.2, communication signal strength 0.1, airspeed 0.1, remaining battery power 0.1, with a total weight of 1.2.

[0030] Divide the unnormalized weights of each sensor by the sum of the total weights to obtain the normalized weights, using the following formula: w′x = wx / ∑wx (w′x is the weight after normalization, wx is the weight before normalization, ∑w_x = 1.2), and the sum of all weights after normalization is 1.

[0031] In the above scheme, the perception data sources are clearly defined by GPS / BeiDou positioning data, inertial measurement unit (IMU) attitude angle data, lidar / visual obstacle distance data, remaining battery power, flight altitude, airspeed, and communication signal strength. This is responsible for standardizing the perception layer of the UAV swarm, ensuring the integrity and accuracy of the information acquired by each UAV. Filtering is used to solve the problems of high sensor noise and data jitter. An effective data broadcasting mechanism is used to solve the problems of insufficient communication and data congestion in large-scale UAV swarms. This provides clean, reliable, and low-redundancy input data for topology judgment, control calculation, and obstacle avoidance decision-making.

[0032] The formula for calculating node connection weights in step 3 is: W_ij=α×S_ij+β×D_ij+γ×E_ij; Where α, β, and γ are weighting coefficients and α+β+γ=1, S_ij is the normalized value of the communication signal strength between UAV i and its neighbor j, D_ij is the normalized value of the relative distance, and E_ij is the normalized value of the remaining battery power of neighbor j; the node connection weighting coefficients are allocated with general values ​​according to the principle of prioritizing communication quality, followed by relative distance, and then supplementing with remaining battery power, α=0.4 (communication signal strength weighting coefficient), β=0.4 (relative distance weighting coefficient), and γ=0.2 (remaining battery power weighting coefficient), satisfying α+β+γ=1; for large-scale clusters (≥50 UAVs), γ can be adjusted to 0.3, and α / β can each be reduced to 0.35 to adapt to the computing power and communication stability requirements of high-battery nodes; The effective neighbor node determination thresholds are: communication RSSI ≥ -70dBm, relative distance ≤ 5m, and remaining battery power ≥ 20%. The topology update cycle is consistent with the control cycle in step 7, and the communication signal strength is normalized as: S_ij = (RSSIij) ( 100)) / [( 50) ( 100)]=(RSSIij+100) / 50 (RSSI value ranges from -100dBm to -50dBm); Relative distance normalization: D_ij = (D_max) d_ij) / (D_max D_min)=(5 d_ij) / 5 (Default Dmax=5m, Dmin=0m, adjusted according to the new distance threshold after cluster expansion); Remaining power normalization: E_ij = Soc_j / 100 (Soc_j is the percentage of remaining power of neighbor node j, with a value of 0-100%).

[0033] The expression for the coupled consistency control law in step 4 is: u_i=k_p∑(p_j-p_i)+k_v∑(v_j-v_i)+k_θ∑(θ_j-θ_i); Where u_i is the total control quantity of the i-th UAV, and k_p, k_v, and k_θ are the position, velocity, and attitude coupling coefficients, respectively, with values ​​ranging from k_p∈[0.8,1.2], k_v∈[0.3,0.6], and k_θ∈[0.1,0.3]. The specific values ​​are determined according to the UAV model. p_i, v_i, and θ_i are the position vector, velocity vector, and attitude angle vector of the i-th UAV in the local machine, respectively; p_j, v_j, and θ_j are the position vector, velocity vector, and attitude angle vector of the neighbor node j, respectively; and ∑ represents the summation of the data of all valid neighbor nodes.

[0034] The above calculations address the problem of chaotic neighbor relationships caused by relative position changes, communication fluctuations, and uneven battery power during drone flight. This ensures that consistency control only performs calculations on valid, reliable, and close neighbors, avoiding invalid interactions.

[0035] The potential field function of the improved artificial potential field method in step 5 is: U_total = U_att + U_rep + U_coop; Where U_att is the target gravitational potential field, U_rep is the repulsive potential field, and U_coop is the cooperative potential field. The obstacle avoidance response time is ≤50ms, and the consistency control command is automatically restored after the collision risk is eliminated. U_att = 0.5 × k_att × ||p_d - p_i||², where k_att is the gravitational coefficient and p_d is the desired formation position; U_rep=0.5×k_rep×(1 / ||p_obst-p_i||-1 / R0)², where k_rep is the repulsion coefficient, p_obst is the location of the obstacle / neighbor node, and R0 is the safe distance threshold; U_coop=0.5×k_coop×||∑p_j / n-p_i||², where k_coop is the cooperation coefficient and n is the number of effective neighbor nodes.

[0036] The general benchmark values ​​for the potential field coefficients in the improved artificial potential field method are k_att=5.0 (gravitational coefficient), k_rep=10.0 (repulsive coefficient), and k_coop=3.0 (cooperation coefficient), which can be dynamically adjusted according to the aircraft type and environment. The adjustment rules for different aircraft types are as follows: multi-rotor UAVs maintain the benchmark values, fixed-wing UAVs, due to their high flight speed, have k_rep increased to 15.0 and k_att increased to 8.0, and tilt-rotor UAVs take the intermediate values ​​(k_att=6.5, k_rep=12.5, k_coop=3.0). In environments with dense obstacles, k_rep for all aircraft types is increased by 20%.

[0037] When multiple obstacles / neighboring nodes exist, the total potential force is calculated using the vector synthesis method: 1. Calculate the repulsive force vector of each obstacle / neighbor node to the local machine, the target's attractive force vector, and the cooperative potential force vector of all neighbor nodes; 2. Vector superposition is performed according to the principle of repulsive force first, synergy second, and gravity as the basis, with the repulsive force vector weight being 0.5, the synergistic potential field vector weight being 0.3, and the gravitational force vector weight being 0.2; 3. The direction of the synthesized total potential force is the core direction of the obstacle avoidance correction amount. Its magnitude is proportional to the magnitude of the total potential force. The larger the magnitude, the greater the correction amount.

[0038] In the above scheme, by improving the artificial potential field method, a potential field function combining attraction, repulsion and cooperative potential field is introduced. On the basis of consistency control, safety correction is superimposed to realize the integrated control of inter-aircraft collision avoidance and environmental obstacle avoidance. This effectively solves the technical pain points of traditional obstacle avoidance algorithms, such as "dispersal of formation when avoiding obstacles and conflict between cooperation and safety". Ultimately, it achieves the safety effect of obstacle avoidance response speed ≤50ms and no collision risk. Moreover, the formation of the cluster does not collapse and the cooperation is continuous during the obstacle avoidance process. It can adapt to complex obstacle environment and dense cluster flight scenarios.

[0039] In some possible implementations, the heartbeat message sending period in step 6 is consistent with the control period. Fault determination adopts a dual threshold rule: if a neighbor node's heartbeat message is not received for three consecutive control periods, or if the state data transmitted by the node exceeds the physically reasonable range (position change ≥ 5cm, attitude change ≥ 5°, speed change ≥ 1m / s, flight altitude change ≥ 0.5m, airspeed change ≥ 0.5m / s, any one of these conditions is considered data abnormal), it is determined to be a faulty node; topology self-repair uses a breadth-first search algorithm to reconstruct neighbor connection paths, with a reconstruction time ≤ 20ms, and the cluster connectivity retention rate after reconstruction ≥ 95%.

[0040] In the above scheme, heartbeat message interaction and dual-threshold fault judgment rules (continuous loss of heartbeats or abnormal data are judged as faults) are used to accurately identify faulty nodes. Then, a breadth-first search algorithm is used to quickly reconstruct the communication topology and control topology, which solves the problems of cluster loss of control, topology breakage and formation collapse that are prone to occur after UAV failure or loss of connection. This ensures that the "decentralized architecture" is truly resilient. Its technical effects are reflected in fast and accurate fault identification with no missed or false judgments, topology reconstruction time ≤20ms, almost no disturbance in cluster flight, and no impact on overall operation even if there is a single point or multiple point failure, thus greatly improving the resilience.

[0041] In some possible implementations, the edge computing node selection and task offloading strategy in step 7 is as follows: based on the remaining power collected in step 2 and the node connection weight in step 3, select the top 20% of drones with the highest weight and the remaining power ≥30% as edge computing nodes, and allocate high-complexity computing tasks (including data fusion, control law calculation, obstacle avoidance path planning) within the cluster to the edge computing nodes, thereby reducing the computing power load of a single machine by ≥40%; the control cycle is fixed at 20ms-50ms and dynamically adjusted according to the cluster size, with a control cycle of 20ms for a cluster of 10-50 drones and 50ms for a cluster of 51-100 drones. Low-battery nodes are handled in two tiers based on their remaining battery power: ① When the drone's remaining battery power is ≤20%, it automatically broadcasts a low-battery warning message to the cluster and exits the edge computing node queue, no longer undertaking cluster computing offload tasks. The node with the second highest weight in the edge computing node list takes its place. ② When the drone's remaining battery power is ≤10%, it automatically broadcasts an emergency return-to-home command. The cluster adjusts its formation based on the dynamic topology map according to the "nearest neighbor replacement" principle, redistributing the node's effective neighbor nodes to fill the formation gap. The return-to-home priority of low-battery nodes is higher than that of cluster collaborative tasks, and the return-to-home command is executed first to ensure the integrity of the cluster formation and the flight safety of low-battery nodes.

[0042] In the above scheme, by selecting high-power, high-weight nodes as edge computing nodes to offload complex computing tasks, and combining a control cycle design of 20-50ms that is dynamically adjusted according to the cluster size, the problems of weak onboard computing power of UAVs, insufficient real-time performance of complex algorithms, and high control latency of large-scale clusters are solved. This approach ensures real-time output and closed-loop correction of control commands, while achieving an optimization effect of reducing the computing load of a single machine by ≥40%. The hardware threshold is lower, and clusters of 10 to 100 UAVs can achieve real-time control by dynamically adjusting the control cycle. The closed-loop feedback mechanism continuously corrects errors, ensuring the flight accuracy and stability of the cluster.

[0043] It should be noted that: 1. Edge nodes are assigned computing tasks according to their connection weights, i.e., the task weight of an edge node = its own weight / the sum of the weights of all edge nodes; 2. The data interaction format between ordinary nodes and edge nodes is "UAV ID + Sensing Data Type + Data Value + Timestamp", which is transmitted synchronously according to the control cycle; 3. After the edge node completes the calculation, it sends "Calculation Result + Validity Identifier" back to the ordinary node. A validity identifier of 1 indicates that the result is usable, while a value of 0 indicates that the calculation is abnormal, and the ordinary node immediately switches to independent calculation on its own.

[0044] Task offloading rules: High-complexity computation tasks adopt a "partial offloading" mode, with 70% of the computational workload of data fusion and obstacle avoidance path planning offloaded to edge nodes, and 50% of the computational workload of control law solving offloaded to edge nodes, while single machines retain basic solving capabilities; the higher the weight, the greater the proportion of computational workload it undertakes. Data interaction rules: Ordinary nodes transmit filtered raw sensing data to edge nodes according to the control cycle, and edge nodes feed back the calculation results to ordinary nodes. Data interaction is synchronized with the topology update cycle and high-priority message transmission is adopted using the Ad-Hoc protocol.

[0045] In some possible implementations, the cluster size supports seamless expansion from 10 to 100 drones. During expansion, the relative distance threshold in step 3 is adjusted (5m for up to 50 drones, with a control cycle of 20ms; 8m for more than 50 drones, with a control cycle of 50ms) for adaptation. The cooperative position error is ≤5cm, the attitude synchronization error is ≤0.5°, and the speed synchronization error is ≤0.1m / s. Under all scales, the thresholds for effective neighbor RSSI ≥-70dBm and remaining battery power ≥20% remain unchanged.

[0046] The above solution supports seamless expansion from 10 to 100 drones. Large-scale cluster adaptation only requires adjusting the relative distance threshold in step 3 without modifying the core control logic. It provides a clear performance commitment for practical engineering applications and can be seamlessly expanded from small performance clusters to large operational clusters. Even after large-scale expansion, it can still maintain high precision with a collaborative position error of ≤5cm, an attitude synchronization error of ≤0.5°, and a speed synchronization error of ≤0.1m / s. The engineering implementation cost is low.

[0047] In some possible implementations, the control method is applicable to heterogeneous UAV swarms of multi-rotor, fixed-wing, and tilt-rotor aircraft. It adapts to different aircraft types by adjusting the potential field coefficient in step 5 and the coupling coefficient in step 4, as detailed below: For the multi-rotor UAV, k_p=1.0, k_v=0.4, and k_θ=0.2; For the fixed-wing UAV, k_p=1.2, k_v=0.6, and k_θ=0.3; The tilt-rotor UAV has k_p=1.1, k_v=0.5, and k_θ=0.25.

[0048] The above scheme explicitly supports the collaborative operation of heterogeneous UAV swarms, including multi-rotor, fixed-wing, and tiltrotor types. By configuring corresponding coupling coefficients and potential field coefficients for different UAV models, it achieves the design goal of adapting a single framework to all types of UAVs. This solves the problem that traditional swarm control can only collaborate with the same type of UAV and cannot achieve stable flight for heterogeneous types. It can achieve stable formation flight in mixed-type swarms without motion conflicts. It can adapt to different platforms simply by adjusting parameters, making it highly reusable. It can be widely applied to heterogeneous swarm tasks in multiple scenarios such as inspection, search and rescue, surveying, and performance.

[0049] Example 1 A cluster of 20 heterogeneous UAVs (15 multi-rotor, 3 fixed-wing, and 2 tiltrotor) was formed to perform formation flight and dynamic obstacle avoidance operations.

[0050] Step 1: Cluster initialization and decentralized self-organizing network formation 1.1 Assign a unique identification code (ID1-ID20) to each of the 20 UAVs. The code uses an 8-bit binary identifier, which includes a type-differentiation bit (00 for multi-rotor, 01 for fixed-wing, and 10 for tiltrotor) and a serial number bit. 1.2 After all drones are started, the Ad-Hoc decentralized communication protocol is automatically enabled. Each drone scans other drones within the communication range through its onboard communication module and sends a networking message containing its own ID, model, and initial positioning request. 1.3. Time synchronization is achieved using PPS pulse synchronization technology: Any drone is selected as a temporary synchronization reference (no need to specify, automatic negotiation). The reference drone periodically sends PPS pulse signals, and other drones receive them and calibrate their local clocks, achieving a synchronization accuracy of 0.3ms. The temporary reference drone for PPS pulse synchronization is automatically negotiated according to the "three priorities" rule: 1. Drones with remaining battery power ≥ 80% have priority; 2. Drones with communication signal strength RSSI ≥ -60dBm have priority; 3. Multi-rotor drones in heterogeneous clusters have priority (high hovering stability). If multiple drones meet the conditions, the one with the smallest node identification code number is taken as the reference. If the reference node fails, renegotiation is carried out according to the same rules. The negotiation process is completed within 3 seconds during the networking phase. 1.4 Initial absolute position data is collected through GPS / BeiDou dual-mode positioning module with a positioning accuracy of 0.8cm. Each UAV broadcasts its initial position data to other UAVs within the communication range and autonomously establishes an initial neighbor node list. The list includes neighbor ID, initial position, and communication signal strength information. The networking process does not require ground station intervention and the networking completion time is ≤3s.

[0051] Step 2: Multi-source sensing and distributed data fusion 2.1 Each UAV collects data in real time through onboard sensors: GPS / BeiDou positioning module collects location (longitude, latitude, altitude) data at a sampling frequency of 100Hz; IMU collects attitude angle (roll angle, pitch angle, yaw angle) and angular velocity data at a sampling frequency of 200Hz; LiDAR collects distance data between surrounding obstacles and neighboring UAVs at a sampling frequency of 50Hz; battery monitoring module collects remaining battery data at a sampling frequency of 10Hz; airspeed sensor collects flight speed data at a sampling frequency of 100Hz; communication module collects communication signal strength (RSSI) data at a sampling frequency of 50Hz. 2.2 Weighted mean filtering is performed on the collected raw data for noise reduction: The filtering window size is set to 5 sampling points, and the weights are allocated according to the sensor accuracy: LiDAR data weight 0.4, IMU data weight 0.3, GPS / BeiDou positioning data weight 0.2, communication signal strength data weight 0.1, airspeed data weight 0.1, battery data weight 0.1, and airspeed 0.1 (the total weight is 1.2, which is the allocation before normalization; normalization is performed during actual calculation). This filtering removes sensor noise and improves the accuracy of the data. To improve data reliability, the specific calculation steps for normalizing the filter weights are as follows: Divide the LiDAR (0.4), IMU (0.3), GPS / BeiDou (0.2), communication signal strength (0.1), airspeed (0.1), and remaining battery power (0.1) by the total weights and 1.2 respectively to obtain the normalized weights: LiDAR (0.333), IMU (0.25), GPS / BeiDou (0.167), communication signal strength (0.083), airspeed (0.083), and remaining battery power (0.083). The sum of all weights is 1, and the filtering operation is performed according to these normalized weights.

[0052] 2.3 Define valid update data: When the difference between the current period data and the previous period data exceeds a preset threshold, it is determined to be valid update data. The threshold is set as follows: 0.5cm for positioning data, 0.1° for attitude data, 0.1m / s for speed data, 0.2m for obstacle distance data, 1% for battery data, and 2dBm for communication signal strength data. Valid update data is only broadcast to neighboring nodes within the communication range to avoid redundant data transmission and reduce communication load. 2.4 After each UAV receives valid update data broadcast by neighboring nodes, it performs distributed fusion with its own filtered perception data. The fusion method is weighted averaging, and the weight of the neighboring data is determined based on the communication signal strength (weight 0.8 when RSSI ≥ -60dBm, weight 0.6 when -70dBm ≤ RSSI < -60dBm), forming a local state consensus and ensuring the UAV's accurate perception of the surrounding neighboring node states and environmental states. The distributed data fusion of multiple neighboring nodes adopts the "communication weighted average method," and the specific formula is as follows: =(w0×P i + × ) / (w0+ ),in The merged local state data has w0=1.0 as its base weight. w_j is the communication weight of neighbor node j (w_j=0.8 when RSSI≥-60dBm, w_j=0.6 when -70dBm≤RSSI<-60dBm), and n is the number of effective neighbor nodes; other state data such as velocity and attitude are fused according to this formula.

[0053] Step 3: Dynamic Topology Generation and Real-time Maintenance 3.1 Based on the communication signal strength, relative distance, and remaining battery power data obtained in step 2, calculate the node connection weight, the normalized value of the communication signal strength between UAV i and neighbor j (mapping RSSI from -100dBm to -50dBm to 0 to 1), the normalized value of the relative distance (mapping 0 to 10m to 1 to 0), and the normalized value of the remaining battery power of neighbor j (mapping 0% to 100% to 0 to 1). 3.2. Set the effective neighbor node judgment threshold: the communication RSSI ≥ -70dBm, the relative distance ≤ 5m, and the remaining power ≥ 20%. Nodes that meet the above three conditions are included in the effective neighbor node set. Nodes with a weight W_ij ≥ 0.5 are given priority as core neighbor nodes and participate in the consensus control law solution. 3.3 The topology update cycle is consistent with the subsequent control cycle. In this embodiment, the cluster size is 20 aircraft and the control cycle is set to 20ms. That is, the node connection weight is recalculated every 20ms, and the effective neighbor node set and dynamic topology graph are updated to ensure that the topology structure matches the dynamic changes of the UAV's flight status, communication status and power status in real time, so as to provide accurate neighbor node information for subsequent control steps.

[0054] Step 4: Solving the Coupled Distributed Consistency Control Law 4.1 Define the collaborative control objective: The cluster flies in a preset formation. The desired position, speed, and attitude of each UAV are determined by the formation mission command. In this embodiment, the preset formation is a rectangular formation with a distance of 3m between adjacent UAVs. 4.2 Based on the effective neighbor node information in the dynamic topology graph generated in step 3, a coupled consistency error function is constructed. The error function is defined as the weighted sum of the position difference, velocity difference, and attitude difference between each neighbor node and the local machine. 4.3 The total control quantity of the local machine is calculated using the control law expression. The control quantity calculation process only depends on the state data of the local machine and the effective neighbor nodes, without the need for global state information. Each UAV completes the calculation independently, and the calculation time is ≤5ms, ensuring the real-time performance of the control commands.

[0055] Step 5: Distributed obstacle avoidance and conflict resolution 5.1 Based on the lidar obstacle data collected in step 2 and the neighbor node location data in step 3, the collision risk is judged in real time: when the distance between the detected environmental obstacle and the local machine is ≤2m, or the distance between the neighbor node and the local machine is ≤1.5m, it is determined that there is a collision risk. 5.2. The total potential force is calculated using an improved artificial potential field method. In the potential field function, the gravitational coefficient k_att = 5.0, the repulsive coefficient k_rep = 10.0, the cooperative coefficient k_coop = 3.0, and the safe distance threshold R0 = 2m. The obstacle avoidance correction is calculated using the total potential force. The direction of the correction is towards the area with the lowest collision risk, and its magnitude is dynamically adjusted according to the collision risk level (the closer the distance, the larger the correction). 5.3 Add the obstacle avoidance correction to the consistency control quantity calculated in step 4 to form the final control quantity, ensuring that the UAV maintains its flight toward the desired formation position while avoiding obstacles and neighboring nodes, thus preventing formation collapse. 5.4 Obstacle avoidance response time ≤ 50ms. When the collision risk is eliminated (obstacle / neighbor node distance ≥ 2.5m), the obstacle avoidance correction weight is automatically reduced to 0, the original consistency control command is restored, and the continuity of formation coordination is ensured.

[0056] Step 6: Fault Detection and Topology Self-Healing 6.1 Each UAV sends a heartbeat message to its effective neighbor nodes every 20ms control cycle. The heartbeat message includes the UAV ID, current status identifier (normal / abnormal), and core status data (position, velocity, attitude summary). At the same time, it receives heartbeat messages from neighbor nodes and records the reception time and status data. 6.2 Fault determination adopts a dual threshold rule: If a heartbeat message from a neighboring node is not received for three consecutive control cycles (60ms), or if the core status data transmitted by the node exceeds the physical reasonable range (position change ≥ 5cm, attitude change ≥ 5°, velocity change ≥ 1m / s), then the node is determined to be a faulty node, removed from the set of valid neighboring nodes, and the faulty node ID is broadcast to other neighboring nodes. 6.3 Topology self-repair adopts breadth-first search algorithm: After removing faulty nodes, each UAV searches for new neighbor connection paths based on the location and communication status of the remaining valid neighbor nodes, and prioritizes the establishment of new connections with nodes with weight W_ij≥0.5 to ensure the connectivity of the cluster topology; In this embodiment, an ID5 multi-rotor UAV malfunction is simulated (communication module is turned off). The cluster completes fault node identification and topology reconstruction within 20ms. After reconstruction, the neighboring nodes of ID5 (ID4, ID6) quickly establish new connections with ID3 and ID7. The cluster connectivity is maintained at 100%, and there is no obvious disturbance to the flight status.

[0057] Step 7: Control command output and closed-loop iteration 7.1. Convert the total control quantity after the correction in step 5 into specific execution commands according to the UAV type: convert the multi-rotor UAV into the rotation speed command of the four rotors, convert the fixed-wing UAV into the control commands of throttle, aileron, elevator and rudder, and convert the tilt-rotor UAV into the rotor speed and tilt angle commands. 7.2 The execution command drives the actuator to move through the airborne flight control system, realizing the adjustment of the UAV's position, speed, and attitude. The response time of the actuator is ≤10ms. Calibration methods for the response parameters of actuators of different models: 1. Multi-rotor UAVs: Rotor speed response time ≤ 10ms, ensured by calibrating the output frequency of the electronic speed controller (ESC) (≥ 500Hz); 2. Fixed-wing UAVs: Control surface deflection angle response time ≤ 15ms, ensured by calibrating the speed and transmission ratio of the servo motor; 3. Tiltrotor UAVs: Tiltrotor angle response time ≤ 12ms, rotor speed response time ≤ 10ms, ensured by using dual servo motors for independent drive; The actuator response indicators of all models are calibrated during the cluster networking phase. Uncalibrated UAVs cannot be added to the list of valid neighbor nodes.

[0058] 7.3 The control cycle is fixed at 20ms. After each cycle, steps 2 to 6 are repeated to form a closed-loop negative feedback control, which corrects the UAV state deviation in real time and ensures that the cluster always maintains the desired formation and stable flight state.

[0059] Step 8: Task Termination and Cluster Disbandment 8.1 When a task termination instruction is received from the ground station (transmitted to the entire cluster via any drone relay), or when the remaining battery power of a drone is ≤10% (broadcasting a low battery alarm to the cluster), the cluster initiates the task termination process. 8.2 All drones exit the distributed collaborative control mode and switch to the following states according to preset rules: multi-rotor drones autonomously return to the preset take-off and landing points, fixed-wing drones maintain cruise mode and wait for further instructions, and tilt-rotor drones hover and stand by. 8.3 Shut down the Ad-Hoc decentralized communication network, stop data interaction between neighboring nodes, and complete the cluster disbandment. There is no risk of collision during the disbandment process.

[0060] Implementation verification An experimental platform was built using 20 heterogeneous UAVs (15 multi-rotor, 3 fixed-wing, and 2 tiltrotor) to perform rectangular formation flight, dynamic obstacle avoidance, and fault injection tests. The verification results are as follows: 1. Coordination accuracy: During formation flight, the relative position error between UAVs is ≤3cm, the attitude synchronization error is ≤0.3°, and the speed synchronization error is ≤0.08m / s, meeting the requirements for high-precision coordination; 2. Obstacle avoidance performance: When three fixed obstacles (1m diameter cylinders) are set in the flight path, the cluster responds within 50ms and successfully avoids the obstacles. During the obstacle avoidance process, the formation form retention rate is ≥95% and no collision occurs. 3. Damage resistance: Simulating the sequential failure of 3 drones (communication interruption), the cluster completed topology reconstruction within 20ms, the flight status was stable, and there was no formation collapse or loss of control. 4. Scalability: Tests were conducted on clusters of 10, 50, and 100 homogeneous multi-rotor UAVs, respectively. The coordinated position error was ≤5cm, and the control cycles were 20ms, 30ms, and 50ms, respectively. There were no communication congestion or excessive control delay. 5. Heterogeneous adaptability: When multi-rotor, fixed-wing, and tiltrotor UAVs are flying in a mixed swarm, all types of UAVs can maintain a coordinated state, with good attitude and speed matching and no obvious motion conflicts.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A distributed flight control method for unmanned aerial vehicle (UAV) swarms, characterized in that, include: Step 1: Assign a unique identification code to all drones in the cluster, build a decentralized peer-to-peer communication network using the Ad-Hoc protocol, complete sub-millisecond time synchronization and initial absolute position calibration, and autonomously establish an initial list of neighbor nodes; Step 2: Each drone collects its own status and environmental data through onboard sensors. After denoising the raw data, it broadcasts valid update data only to neighboring nodes within the communication range. It receives data from neighboring nodes and performs distributed fusion with its own data to form a local state consensus. Step 3: Calculate the node connection weight based on three indicators: communication quality, relative distance, and remaining power. Select effective neighbor nodes to generate a dynamic undirected topology graph. The topology update cycle is synchronized with the control cycle to provide a neighbor node basis for subsequent consistency control. Step 4: Using position, velocity, and attitude as cooperative control variables, and based on the neighbor node information in the dynamic topology generated in Step 3, construct a coupled consistency control law and the corresponding error function of "weighted summation of the differences between the cooperative variables of the local machine and the neighbor nodes", solve the flight control quantity of the local machine, and realize closed-loop constraint control of the cooperative error of the cluster position, velocity, and attitude. Step 5: Based on the environmental data from Step 2 and the location information of neighboring nodes in the dynamic undirected topology graph from Step 3, the improved artificial potential field method with cooperative potential field terms is used to correct the control quantity calculated in Step 4, so as to achieve inter-machine collision avoidance and environmental obstacle avoidance while maintaining cluster collaboration. Step 6: Perform fault determination through heartbeat message interaction and status data verification, identify faulty nodes, remove faulty nodes, and reconstruct the communication topology and control topology based on the breadth-first search algorithm to ensure cluster connectivity and control continuity. Step 7: Based on the remaining power collected in Step 2 and the node connection weights in Step 3, edge computing nodes are selected and high-complexity computing tasks are unloaded. Then, the control quantity corrected in Step 5 is converted into flight control execution commands to drive the UAV actuators to move. Steps 2 to 6 are repeated according to a fixed control cycle to form a closed-loop negative feedback control. Step 8: After receiving the mission completion instruction, the cluster exits the distributed collaborative control mode, and each UAV autonomously switches to return-to-home or hovering standby state, shutting down the self-organizing communication network.

2. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, In step 1, time synchronization uses PPS pulse synchronization technology with a synchronization accuracy of ≤0.5ms. Initial position calibration is based on GPS / BeiDou dual-mode positioning, with a positioning accuracy of ≤1cm; The Ad-Hoc communication protocol supports node relay forwarding.

3. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 2, characterized in that, The sensing data in step 2 includes GPS / BeiDou positioning data, inertial measurement unit attitude angle data, lidar / visual obstacle distance data, remaining battery power, flight altitude, airspeed, and communication signal strength. The noise reduction process employs weighted mean filtering, with weights dynamically allocated based on sensor accuracy. Specifically, the weights for lidar data are 0.4, inertial measurement unit data are 0.3, GPS / BeiDou positioning data are 0.2, and communication signal strength data are 0.

1. The valid update data is defined as abnormal data or status change data whose difference from the previous period's data exceeds a preset threshold. The threshold is set as follows: positioning data 0.5cm, attitude data 0.1°, speed data 0.1m / s, obstacle distance 0.2m, remaining battery power 1%, communication signal strength 2dBm, flight altitude 0.1m, and airspeed 0.05m / s.

4. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 3, characterized in that, The formula for calculating the node connection weight in step 3 is as follows: W_ij=α×S_ij+β×D_ij+γ×E_ij; Where α, β, and γ are weighting coefficients and α+β+γ=1, S_ij is the normalized value of the communication signal strength between UAV i and neighbor j, D_ij is the normalized value of the relative distance, and E_ij is the normalized value of the remaining battery power of neighbor j. The effective neighbor node determination thresholds are: communication RSSI ≥ -70dBm, relative distance ≤ 5m, and remaining power ≥ 20%. The topology update cycle is consistent with the control cycle in step 7.

5. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 4, characterized in that, The expression for the coupled consistency control law in step 4 is: u_i=k_p∑(p_j-p_i)+k_v∑(v_j-v_i)+k_θ∑(θ_j-θ_i); Where u_i is the total control quantity of the i-th UAV, and k_p, k_v, and k_θ are the position, velocity, and attitude coupling coefficients, respectively, with values ​​ranging from k_p ∈ [0.8, 1.2], k_v ∈ [0.3, 0.6], and k_θ ∈ [0.1, 0.3]. p_i, v_i, and θ_i are the position vector, velocity vector, and attitude angle vector of the i-th UAV in the local machine, respectively; p_j, v_j, and θ_j are the position vector, velocity vector, and attitude angle vector of the neighbor node j, respectively; and ∑ represents the summation of the data of all valid neighbor nodes.

6. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 5, characterized in that, The potential field function of the improved artificial potential field method in step 5 is: U_total = U_att + U_rep + U_coop; Where U_att is the target gravitational potential field, U_rep is the repulsive potential field, and U_coop is the cooperative potential field. The obstacle avoidance response time is ≤50ms, and the consistency control command is automatically restored after the collision risk is eliminated. U_att = 0.5 × k_att × ||p_d - p_i||², where k_att is the gravitational coefficient and p_d is the desired formation position; U_rep=0.5×k_rep×(1 / ||p_obst-p_i||-1 / R0)², where k_rep is the repulsion coefficient, p_obst is the location of the obstacle / neighbor node, and R0 is the safe distance threshold; U_coop=0.5×k_coop×||∑p_j / n-p_i||², where k_coop is the cooperation coefficient and n is the number of effective neighbor nodes; The general reference values ​​for the potential field coefficients in the improved artificial potential field method are k_att=5.0, k_rep=10.0, and k_coop=3.

0. Multi-rotor UAVs maintain the reference values, fixed-wing UAVs have k_att=8.0 and k_rep=15.0, and tilt-rotor UAVs have k_att=6.5, k_rep=12.5, and k_coop=3.

0.

7. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 6, characterized in that, In step 6, the heartbeat message sending period is consistent with the control period. The fault determination adopts a dual threshold rule: if a neighbor node's heartbeat message is not received for three consecutive control periods, or if the status data transmitted by the node exceeds the physical reasonable range, it is determined to be a faulty node.

8. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 4, characterized in that, The edge computing node selection and task offloading strategy in step 7 is as follows: Based on the remaining power collected in step 2 and the node connection weight in step 3, the top 20% of drones with the highest weight and ≥30% remaining power are selected as edge computing nodes. High-complexity computing tasks within the cluster are allocated to edge computing nodes, reducing the single-machine computing load by ≥40%. The control cycle is fixed at 20ms-50ms and dynamically adjusted according to the cluster size. The control cycle for a cluster of 10-50 drones is 20ms, and the control cycle for a cluster of 51-100 drones is 50ms.

9. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to claim 8, characterized in that, The cluster size supports seamless expansion from 10 to 100 drones, adjusting the relative distance threshold in step 3 during expansion.

10. The distributed flight control method for unmanned aerial vehicle (UAV) swarms according to any one of claims 1-9, characterized in that, The control method described is applicable to heterogeneous UAV swarms, including multi-rotor, fixed-wing, and tilt-rotor types. It adapts to different UAV models by adjusting the potential field coefficient in step 5 and the coupling coefficient in step 4, as detailed below: For the multi-rotor UAV, k_p=1.0, k_v=0.4, k_θ=0.2, k_att=5.0, k_rep=10.0; For the fixed-wing UAV, k_p=1.2, k_v=0.6, k_θ=0.3, k_att=8.0, k_rep=15.0; For tilt-rotor UAVs, k_p=1.1, k_v=0.5, k_θ=0.25, k_att=6.5, k_rep=12.5, and k_coop=3.0 for all models.

Images