Unmanned aerial vehicle cluster safety control and steering consensus method based on starling group

Abstract

The invention discloses a starling group-based unmanned aerial vehicle cluster safety control and steering consensus method, which comprises the following steps of: firstly, constructing a three-layer leader-follower network simulating a pigeon group hierarchical structure, and establishing a dynamic physical neighbor topology in combination with a communication radius; an improved artificial potential field method is designed, a damage area, a rejection area, an alignment area and an attraction area are divided, a smooth concave-convex function is introduced to adjust a potential field, and cooperative force and obstacle rejection force for maintaining an expected distance are generated; an inertial spin mechanism is introduced, when a maneuvering intention is detected, expected spin and perceptual spin are calculated, a spin kinetic equation is utilized to update a state, and a transverse consensus coupling force for driving a cluster to rapidly and synchronously steer is generated. And finally, for different levels of unmanned aerial vehicles, differentially fusing the control force, and updating the cluster state through numerical integration. According to the method, the internal safety distance of cluster high-density flight is guaranteed, the overall steering speed and consistency of the cluster facing sudden obstacles are remarkably improved, and the robustness of the system in a dynamic environment is enhanced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of unmanned aerial vehicle cluster control, and particularly relates to a safety control and turning consensus method for unmanned aerial vehicle cluster based on starling flocks. BACKGROUND

[0002] Unmanned aerial vehicle cluster has become a key direction for the development of future intelligent unmanned systems due to its multifunctionality, high flexibility and strong reconfigurability in the process of performing collaborative tasks. Especially when performing tasks such as reconnaissance, logistics and formation performance in low-altitude complex environments, the unmanned aerial vehicle cluster often presents the characteristics of high density and large scale, which puts higher requirements on the autonomous obstacle avoidance ability, formation stability and rapid maneuvering response ability of the cluster.

[0003] Currently, the collaborative control methods for unmanned aerial vehicle cluster mainly include methods based on consensus protocol and artificial potential field method. Among them, the control method based on consensus gradually makes the cluster state consistent by designing local interaction rules, achieving formation formation and maintenance; the artificial potential field method simulates gravity and repulsion by constructing a virtual potential field, guiding the unmanned aerial vehicle to avoid obstacles and move towards the target. Although these methods are relatively mature in theory, they still have the following obvious shortcomings when dealing with high-density and dynamic mutation in actual scenarios:

[0004] First, the traditional artificial potential field method usually adopts a simple "repulsion-attraction" binary action mechanism, which is prone to local minimum value problems in high-density clusters, causing unmanned aerial vehicles to be trapped in oscillation or even stagnation, and unable to effectively cross the obstacle area. At the same time, this method is difficult to accurately maintain the distance between the unmanned aerial vehicles, and is prone to loose formation or collision due to the imbalance between repulsion and attraction, with poor safety.

[0005] Second, the traditional consensus control protocol relies on the successive diffusion of neighbor information, and its dynamic response is similar to the process of heat diffusion, with inherent time hysteresis. When the cluster needs to make a rapid turn or emergency obstacle avoidance, unmanned aerial vehicles at the edge of the network or the end of the communication often cannot obtain the turning instructions in time, resulting in delayed response of the cluster, broken formation, and even some individuals falling behind or colliding, making it difficult to meet the real-time and consistency requirements in high-maneuvering scenarios.

[0006] Therefore, the existing unmanned aerial vehicle cluster control methods still have significant bottlenecks in simultaneously ensuring flight safety and rapid collaborative maneuvering under high-density conditions. In view of the above problems, it is necessary to draw lessons from the efficient and robust cluster interaction mechanism exhibited by bird flocks (such as starling flocks and pigeon flocks) in nature, and design a collaborative control method for unmanned aerial vehicle cluster that can integrate bionics principles and have the capabilities of safe obstacle avoidance, formation maintenance and rapid turning consensus. SUMMARY

[0007] The unmanned aerial vehicle cluster safety control and turning consensus method based on the starling group aims to overcome the shortcomings of the prior art, and solves the technical problems that the traditional artificial potential field method is prone to local minimum in high-density clusters and is difficult to accurately maintain a safe distance, and the information transmission of the traditional consensus algorithm is delayed, resulting in slow cluster turning response and poor maneuvering consistency.

[0008] To achieve the above-mentioned purpose, the unmanned aerial vehicle cluster safety control and turning consensus method based on the starling group is constructed by constructing a bionic hierarchical network and a multi-mechanism fusion controller, realizing the autonomous cooperation of the cluster.

[0009] Firstly, a bionic hierarchical command network and a UAV dynamics model are constructed. A UAV motion model based on particle dynamics is established, and according to the information transmission hierarchical characteristics of pigeon flight, the UAV cluster is logically divided into three layers of static "leader-follower" structure: the first layer is the global leader, which directly obtains environmental information and navigation instructions; the second layer is the relay leader, which is responsible for information transmission and local guidance; the third layer is the ordinary follower, which executes local following. At the same time, a physical neighbor interaction topology is constructed combined with the onboard communication range, forming a hybrid network architecture of "logical layering and physical interconnection", laying a structural foundation for cooperative control.

[0010] Secondly, the cluster safety distance maintenance and obstacle avoidance mechanism based on the improved artificial potential field is designed to ensure individual safety. In order to solve the problem of unbalanced force of traditional potential field method under high density, the interaction space between UAVs is finely divided into four continuous regions: damage zone, repulsion zone, alignment zone and attraction zone, and a smooth concave-convex function is introduced to adjust the potential field strength between regions. Based on this, a piecewise smooth pairwise potential function is constructed, and the negative gradient can generate a neighbor interaction force for maintaining a fixed desired distance. At the same time, virtual nodes are generated for static environmental obstacles, and an obstacle repulsion potential field is constructed, so as to synthesize an obstacle repulsion force that can effectively avoid local minimum and smoothly guide the UAV to bypass the obstacle.

[0011] Then, a turning consensus protocol based on the inertia spin of starling flocks is designed to improve the maneuvering consistency. To overcome the delay of information transmission, the "inertia spin" variable simulating the physical process of bird turning is introduced. When the cluster detects the need for a large maneuver, the turning consensus module is triggered: first, the "expected spin" of the individual itself is calculated according to the deviation of the "maneuvering intention vector" synthesized by navigation, obstacle avoidance and neighbor force from the current speed; at the same time, the spin state of the neighbor is perceived, and the "perceived spin" is calculated. Subsequently, the spin state of each unmanned aerial vehicle is updated through a second-order spin dynamics equation containing a damping term and a noise term. Finally, the spin state is fed back to the speed control loop and converted into a coupling force driving the lateral motion of the cluster, so that the turning intention can be quickly transmitted in the hierarchical network in the form of waves, realizing the synchronization and rapid turning of the cluster.

[0012] Finally, multi-layer fusion control and cluster state update are performed. According to the hierarchy of the unmanned aerial vehicle, the above control components are differentially fused: for the first level leader, the control input is weighted and integrated with the neighbor force, obstacle repulsion force and global navigation force; for the lower level follower, the control input mainly fuses the neighbor force and the speed consensus force generated by the spin mechanism. In the calculation of the following force, the communication time delay is introduced to simulate the biological reaction lag, enhancing the bionic authenticity of the model. The total control input is obtained by vector synthesis of all components, and the position and speed state of each unmanned aerial vehicle in the cluster are updated by numerical integration method, completing the distributed cooperative control closed loop.

[0013] The technical scheme of the present application is:

[0014] A starling flock-based unmanned aerial vehicle cluster safety control and turning consensus method, comprising the following steps:

[0015] Step 1: Construct a bionic control network of unmanned aerial vehicles based on biological cluster behavior: establish a particle dynamics model of unmanned aerial vehicles, and according to the preset hierarchical logic, divide the unmanned aerial vehicles in the cluster into a three-layer structure including at least first level leaders, second level relay leaders and third level ordinary followers, and construct a physical neighbor interaction topology based on the on-board communication radius, forming a network architecture combining logical hierarchy and physical interconnection;

[0016] Step 2: Design and calculate the control force for maintaining a safe distance and avoiding obstacles: based on the improved artificial potential field method, divide the interaction space between unmanned aerial vehicles into damage zone, repulsion zone, alignment zone and attraction zone, and introduce a smooth concave-convex function to adjust the strength of the interval potential field, construct a piecewise continuous pairwise potential function, and obtain the neighbor cooperative force for maintaining the desired distance between the unmanned aerial vehicles by calculating the negative gradient; for the environmental obstacles, a virtual node is generated on the surface of the obstacle and a repulsion potential field is constructed to obtain the obstacle repulsion force;

[0017] Step 3: Design and calculate the control force for achieving rapid turn consensus: When the swarm maneuver demand exceeds a set threshold, activate the consensus mechanism based on the starling flock's inertial spin; calculate the individual's expected spin based on the deviation between the maneuver intention vector synthesized from navigation, obstacle avoidance, and neighbor forces and the current speed; simultaneously, calculate the perceived spin by sensing the spin state of the neighbors; update the spin state of each UAV through the spin dynamics equation including a damping term, and transform it into a coupling force that drives the swarm's synchronized lateral movement;

[0018] Step 4: Perform multi-level fusion control and state update: For the first-level navigator, the neighbor cooperative force, obstacle repulsion force, and navigation force pointing to the global target are weighted and fused to generate its total control input; for the second-level and third-level UAVs, the neighbor cooperative force and the steering consensus coupling force generated in step 3 are weighted and fused to generate their respective total control inputs; using the obtained total control inputs, the position and velocity state of each UAV in the cluster are updated through numerical integration.

[0019] In a further preferred embodiment, in step 1, the particle dynamics model of the UAV is expressed as follows:

[0020]

[0021] in For drones The total control input vector, For drones The position vector, For drones The velocity vector.

[0022] In a further optimized scheme, in step 1, the first-level navigator is located at the front of the cluster and can directly obtain global navigation information and obstacle information, and is responsible for guiding the overall course of the cluster; the second-level relay leader is located in the middle of the cluster and acts as an information relay node, following the first-level navigator and transmitting maneuver information backward; the third-level ordinary follower is located at the rear of the cluster and only interacts locally with its neighbors and superior navigators to maintain a compact formation.

[0023] In a further preferred embodiment, step 2 uses the relative distance between drones. Based on this, define the normalized distance. According to the damage radius Expected Spacing Alignment area bandwidth The interaction space is divided into four continuous regions: the damage zone. Rejection region: Alignment area: Attraction Zone: ;in This refers to the communication coverage radius of the drone.

[0024] A further preferred embodiment is that the concavity / convexity function is:

[0025]

[0026] in , , , The adjustment coefficient controls the attenuation rate of the potential field strength in the repulsive and attractive regions.

[0027] In a further preferred embodiment, the paired potential energy function is:

[0028]

[0029] in, Indicates from drones Pointing to the neighbor The unit direction vector; the neighbor cooperative force is:

[0030] .

[0031] A further preferred embodiment is that the repulsive force of the obstacle is:

[0032]

[0033] in For drones The distance to the nearest point on the surface of the obstacle. The preset avoidance distance, Indicates from drones The unit direction vector pointing to the virtual node.

[0034] In a further preferred embodiment, step 3, the process of determining whether the cluster mobility demand exceeds a set threshold, is as follows:

[0035] Calculate the direction change function :

[0036]

[0037] in For drones The number of neighbors, and drones and Velocity vector; setting steering threshold ,like Greater than the preset threshold If this happens, the cluster is determined to be in a redirection state.

[0038] In a further preferred embodiment, step 3, the process of calculating the desired spin and the sensed spin, updating the spin state, and converting it into a coupling force that drives the synchronous lateral motion of the cluster, is as follows:

[0039] First, calculate the drone's motion intention. :

[0040]

[0041] in To foster neighborly cooperation The repulsive force of the obstacle, For global navigation power, Based on following force;

[0042] Based on motion intention Calculate the expected spin of the drone :

[0043]

[0044] The average spin deviation of the neighboring drones is calculated to obtain the sensed spin. :

[0045]

[0046] in, The coupling strength coefficient is... For the neighbors ;

[0047] Update the UAV based on the spin dynamics equations including damping terms. spin vector :

[0048]

[0049] in The damping coefficient; For random noise that follows a normal distribution;

[0050] Using rotational inertia The spin vector is converted into tangential acceleration and superimposed with the basic following force, thus transforming the spin information into an actual control force as the coupling force driving the synchronous lateral motion of the cluster. :

[0051] .

[0052] In a further optimized approach, step 4, for the first-level navigator, its control input... It is composed of a weighted average of neighbor cooperation force, obstacle repulsion force, and navigation force pointing towards the global goal:

[0053]

[0054] in These are weighting coefficients; for second- and third-level UAVs, their control inputs... It is composed of a weighted average of neighbor cooperation force and coupling force:

[0055]

[0056] in The weighting coefficients are the coupling force; Introducing communication delay Simulating biological response lag, using a superior navigator exist Velocity state at any moment As a reference source:

[0057] .

[0058] Beneficial effects:

[0059] The advantages and beneficial effects of this invention are as follows:

[0060] 1. This invention establishes a four-layer spatial interaction model comprising a damage zone, a repulsion zone, an alignment zone, and an attraction zone. It introduces an improved artificial potential field and concave-convex functions with four-region division, thereby achieving precise and smooth maintenance of safe distance under high-density clustering. This effectively avoids collisions and oscillation problems of traditional potential field methods, and enhances the inherent safety of dense formation flight.

[0061] 2. This invention introduces an inertial spin mechanism inspired by starling flocks, which transforms the turning command into a physical quantity (spin) that can spread rapidly in the cluster. This significantly accelerates the propagation speed of information at the cluster edge, overcomes the slow information transmission defect of traditional consensus algorithms, and enables the cluster to quickly and consistently complete the overall turning maneuver when encountering sudden obstacles, thus overcoming the hysteresis defect of traditional consensus protocols.

[0062] 3. This invention integrates the hierarchical information structure of pigeon flocks with the distributed speed matching mechanism of starling flocks, achieving decoupling between global tasks and local coordination. This hybrid architecture not only reduces communication and computational complexity but also enables the cluster to maintain a high level of formation consistency and throughput efficiency when facing obstacles of different sizes and layouts, demonstrating good environmental adaptability and system robustness.

[0063] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0064] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0065] Figure 1 : Schematic diagram of regional division;

[0066] Figure 2 : Schematic diagram of concave and convex functions;

[0067] Figure 3 Hierarchical division diagram;

[0068] Figure 4 : Schematic diagram of the dynamic update process;

[0069] Figure 5 : Schematic diagram of a multi-level control algorithm based on starling flocks;

[0070] Figure 6 : Schematic diagram of the obstacle avoidance process; (a) (b) ;

[0071] Figure 7 Simulation results of distance-related parameters;

[0072] Figure 8 Simulation results of velocity-related parameters;

[0073] Figure 9 : Simulation results of consistency order parameters;

[0074] Figure 10 Monte Carlo simulation results of obstacle avoidance parameters. Detailed Implementation

[0075] The embodiments of the present invention are described in detail below. These embodiments are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0076] The present invention discloses a method for security control and steering consensus of unmanned aerial vehicle (UAV) swarms based on starling flocks, the core implementation architecture of which is as follows: Figure 5 As shown, by constructing a biomimetic hierarchical network, a refined secure potential field, an inertial spin consensus protocol, and layered fusion control, secure and efficient collaboration of high-density UAV swarms is achieved. The following detailed explanation, in conjunction with the accompanying diagrams and specific implementation steps, provides further details.

[0077] Step 1: Constructing a biomimetic control network for drones based on biological swarm behavior: Establish a point mass dynamics model for the drones, and according to a pre-defined hierarchical logic, divide the drones in the swarm into a three-layer structure containing at least a first-level leader, a second-level relay leader, and a third-level ordinary followers. Simultaneously, construct a physical neighbor interaction topology based on the airborne communication radius, forming a network architecture that combines logical hierarchy with physical interconnection. This step aims to establish the physical motion model and logical organizational structure of the swarm, laying the foundation for collaborative control. Specifically, it includes the following sub-steps:

[0078] Step 1.1: Establish the UAV particle dynamics model

[0079] Unmanned aerial vehicles (UAVs) are equipped with positioning systems and communication devices. To simplify the analysis and focus on swarm cooperative behavior, UAVs are treated as point masses on a two-dimensional plane, ignoring their specific geometric dimensions and aerodynamic characteristics. For those containing... A swarm of fixed-wing drones, any number The state variables of the UAV include the position vector. and velocity vector Its second-order particle dynamics model in continuous time is as follows:

[0080]

[0081] in, For drones The total control input vector. In actual simulation, a fixed step size is used. Discretized numerical integration is performed by setting the simulation step size. Total task duration Initialize the starting positions of all drones. With initial speed Set the state of the task objective point as In this embodiment, a time step is set. Set the total simulation time to 0.03 seconds. The time limit is 600 seconds. The initial positions of the drones are randomly distributed in the starting area, and the initial velocity is 0.

[0082] Step 1.2: Construct a command and control network consisting of a static three-tier leader-follower topology.

[0083] To simulate the hierarchical information transmission structure of biological groups such as pigeon flocks, a static logical hierarchical network is pre-constructed, such as... Figure 3 As shown. Based on the information transmission delay characteristics during pigeon flock flight, the cluster nodes are divided into three logical levels. This structure is allocated before the task begins and remains unchanged during the task. The specific process is as follows:

[0084] Step 1.2.1: Allocation of Hierarchical Numbers and Definition of Roles:

[0085] First, the hierarchical size of the cluster is parameterized to construct a Leader-Follower organizational structure that conforms to the biological characteristics of starlings.

[0086] The first tier (Navigator), located at the front of the cluster, can directly obtain global navigation and obstacle information, and is responsible for guiding the overall course of the cluster; the number of first-tier Navigators is set. The drone is assigned the highest level of privileges (1), pre-installed with global mission path information, and equipped with a detection radius of [missing information]. The radar equipment is responsible for sensing obstacles in the environment and making initial steering decisions.

[0087] The second tier (relay leader), located in the center of the cluster, acts as an information relay node, following the first-tier leader and relaying movement information backward; the number of second-tier relay leaders is set. Number 8 serves as a relay node, responsible for receiving and forwarding the navigator's maneuvering instructions.

[0088] The third tier (ordinary followers): Located at the tail of the cluster, they only interact locally with their neighbors and superior leaders, maintaining a compact formation; the number of ordinary followers under each second-level leader is set. It is rated 6. It only interacts with local neighbors and direct superiors.

[0089] Based on the above preset parameters, the total number of ordinary followers is 48, and the total size of the swarm is 57 drones. The parameters for the number of our drone swarm are shown in Table 1.

[0090] Table 1. Hierarchical Unmanned and Quantity Parameters

[0091]

[0092] Step 1.2.2: Generate a list of node indexes:

[0093] After determining the total size, an identity index sequence from UAV-1 to UAV-57 is generated for subsequent matrix operations and state updates. Index UAV-1 is assigned to the first-level leader, indices UAV-2 to UAV-9 are assigned to second-level leaders, and indices UAV-10 to UAV-57 are assigned to ordinary followers.

[0094] A complete cluster index list is created, and index ranges are divided hierarchically. A mapping function is defined. Indicates drone The superior navigator ID. If the drone... If it is the second level, then This is the first-level node; if the drone As a regular follower, This is the second-level node. Next, all second-level leader nodes are traversed, and their direct leader indices are uniformly mapped to first-level leaders, forming a backbone network centered on first-level leaders; all ordinary follower nodes are traversed, and they are evenly divided into... Each subgroup maps drones within it to a corresponding secondary leader as its direct leader, forming a distributed local follower network.

[0095] Step 1.3: Constructing a physical neighbor interaction topology based on airborne communication radius:

[0096] In addition to the logical hierarchy, each drone interacts in real time with individuals in physical proximity based on its onboard communication equipment.

[0097] Set the communication coverage radius of the UAV's onboard radio to be For any drone Define its real-time neighbor set The set of all nodes whose Euclidean distance is less than the communication radius:

[0098]

[0099] Constructing the adjacency matrix of the cluster , the elements The specific formula for calculating the connectivity between nodes is as follows:

[0100]

[0101] This physical topology is dynamically changing and forms the basis for achieving local cooperative obstacle avoidance and velocity matching.

[0102] Step 2: Design and calculate the control forces for maintaining safe distance and avoiding obstacles: Based on the improved artificial potential field method, the interaction space between UAVs is divided into damage zone, repulsion zone, alignment zone, and attraction zone. A smooth concave-convex function is introduced to adjust the potential field strength within each zone, constructing a piecewise continuous pairwise potential energy function. The neighbor cooperative force for maintaining the desired distance between UAVs is obtained by calculating its negative gradient. For environmental obstacles, virtual nodes are generated on their surfaces, and a repulsive potential field is constructed to obtain the obstacle repulsion force. This step, based on the improved artificial potential field method, generates forces to maintain safe distance within the cluster and avoid external obstacles. Its core process is as follows: Figure 4 As shown.

[0103] Step 2.1: Define the interaction region and the convex / concave functions:

[0104] To achieve smooth and precise distance control, the relative distance between drones is used. Based on this, define the normalized distance. Used for interval determination of subsequent potential field functions. Based on preset safety parameters: damage radius. Expected Spacing Alignment area bandwidth The interaction space is divided into four continuous areas, such as Figure 1 As shown:

[0105] Damage Zone: If the distance is too close, a strong repulsive force will be generated to prevent a collision.

[0106] Rejection zone: They need to repel each other to restore the desired distance.

[0107] Alignment area: It mainly performs speed alignment, with zero distance holding force.

[0108] Attraction Zone: When the distance is too great, attraction is generated to maintain formation.

[0109] To smoothly transition the potential field intensity between different regions, an adjustable concavity / convexity function is introduced. Its shape is like Figure 2 As shown, based on the normalized distance The output adjustment factor is set according to the different regions, realizing the transition of potential field strength between the damage region, repulsion region, alignment region, and attraction region. The calculation formula is as follows:

[0110]

[0111] in , , , The adjustment coefficient controls the attenuation rate of the potential field strength in the repulsive and attractive regions;

[0112] Step 2.2: Calculate neighbor cooperation force:

[0113] Based on concave and convex functions, a drone is defined. with neighbors Pair potential functions between for:

[0114]

[0115] in, Indicates from drones Pointing to the neighbor The unit direction vector; the fractional part of the integral term represents the basic value of the potential energy gradient based on the distance deviation. This potential energy function is relative to the position. The negative gradient is the value of the drone. Because of the neighbor The resulting force. Summing over all neighbors yields the drone's value. Total neighbor cooperative force received:

[0116]

[0117] in, The force is positive (repulsive) in the repulsion region, negative (attractive) in the attraction region, and zero in the alignment region. This causes the drone to be attracted when it is far from its neighbor and repelled when it gets too close, thus maintaining the desired distance. .

[0118] Step 2.3: Calculate the repulsive force of the obstacle:

[0119] For static obstacles in the environment, targeting drones on the obstacle surface. Generate a virtual node Define the obstacle potential energy function for:

[0120]

[0121] in For drones The distance to the nearest point (virtual node) on the obstacle surface. The preset avoidance distance, Indicates from drones The unit direction vector pointing to the virtual node.

[0122] Potential energy function of obstacle Find the negative gradient to obtain the repulsive force of the obstacle. :

[0123]

[0124] The force is directed away from the obstacle, and its intensity increases sharply as the distance decreases.

[0125] Step 3: Design and compute the control force for achieving fast turn consensus. Traditional consensus algorithms suffer from latency when the cluster needs to perform emergency obstacle avoidance or sharp turns. This step introduces a starling flock-inspired inertial spin mechanism to rapidly transmit turn intentions in a wave-like manner.

[0126] When the swarm maneuvering demand exceeds a set threshold, a consensus mechanism based on the inertial spin of the starling flock is activated; the expected spin of an individual is calculated based on the deviation between the maneuvering intention vector synthesized from navigation, obstacle avoidance, and neighbor forces and the current speed; at the same time, the perceived spin is calculated by sensing the spin state of the neighbors; the spin state of each UAV is updated through the spin dynamics equation containing a damping term, and it is transformed into a coupling force that drives the synchronous lateral movement of the swarm.

[0127] Step 3.1: Turning state triggered:

[0128] To identify whether a drone needs to enter a state of drastic maneuvering, a steering consensus mechanism is triggered by assessing the degree of local consensus within the cluster; here, a direction change function is defined. The mean norm of the normalized velocity direction deviation among neighbors:

[0129]

[0130] in, For drones The number of neighbors, and drones and The velocity vector. Setting the steering threshold. ,like Greater than the preset threshold If the cluster enters a turning state, the inertial spin calculation module is activated, and subsequent spin calculations and consistency updates are performed.

[0131] Step 3.2: Calculate spin-related variables, including desired spin and sensed spin;

[0132] When the drone is determined to be in a turning state according to step 3.1, the drone's motion intention is first calculated. Mainly due to neighborly cooperation Basic following force Repulsive force of obstacles and global navigation capabilities constitute:

[0133]

[0134] Neighbor collaboration According to the calculation in step 2.2, The global navigation force is calculated based on step 2.3. According to the formula:

[0135]

[0136] Received, among which For position coefficients, The velocity coefficient is used; the basic following force is determined by the formula.

[0137]

[0138] Calculated.

[0139] Based on this movement intention Calculate the expected spin of the drone (Angular momentum required to characterize the alignment intention):

[0140]

[0141] Simultaneously, the average spin deviation of the neighboring drones, i.e., the sensed spin, is calculated. (Characterizing the deviation from the neighboring spin):

[0142]

[0143] in, The coupling strength coefficient is... For the neighbors The spin vector.

[0144] Step 3.3: Update spin state and synthesize uniform control input:

[0145] The UAV is updated according to the following spin dynamics equations including damping terms. spin vector :

[0146]

[0147] in This is the damping coefficient, used to simulate steering resistance; It is a random noise that follows a normal distribution, used to simulate environmental disturbances.

[0148] Using rotational inertia The spin vector is converted into tangential acceleration and superimposed with the basic following force, thus transforming the spin information into an actual control force as the coupling force driving the synchronous lateral motion of the cluster. This yields the final consistency control input:

[0149]

[0150] Step 4: Perform multi-level fusion control and state update; This step is based on the hierarchical structure determined in Step 1, calculates the total control input of UAVs at different levels, and updates the cluster state.

[0151] Step 4.1: Layered control input fusion, including the first-level navigator control input and the second and third-level follower control inputs.

[0152] The first-level drone possesses global information, and its control inputs It is composed of a weighted average of neighbor cooperation force, obstacle repulsion force, and navigation force pointing towards the global goal:

[0153]

[0154] in These are the weighting coefficients.

[0155] Subsequent-level drones primarily follow, and their control inputs... It is composed of a weighted average of neighbor coordination force and turning consensus coupling force:

[0156]

[0157] in These are weighting coefficients. In the calculation... At that time, communication delay is introduced. Simulating biological response lag, using a superior navigator exist Velocity state at any moment As a reference source:

[0158]

[0159] Step 4.2: Cluster state iterative update:

[0160] Summarize all control inputs of drones Substitute the values ​​into the dynamic equations and perform numerical integration to update the state at the next time step:

[0161]

[0162] Repeat the above steps until the preset simulation duration is completed. .

[0163] To further verify the effectiveness and superiority of the multi-level control algorithm proposed in this invention in actual tasks, this embodiment selects a typical "narrow-spacing continuous obstacle avoidance" scenario for simulation experiments, and performs quantitative analysis of the implementation effect by defining behavioral indicators and comparative experiments.

[0164] Construction of simulation scenarios and evaluation metrics:

[0165] To quantitatively evaluate the collaborative performance of the cluster, an order parameter is defined. The calculation formula is as follows:

[0166]

[0167] The range of values ​​for this indicator is: [ 0 , 1 ] The closer the value is to 1, the more consistent the speed direction of all drones in the cluster, and the neater the formation. Simultaneously, the average distance between the drone and its nearest neighbor, and the minimum distance to the obstacle surface, are recorded to evaluate formation maintenance capability and obstacle avoidance safety, respectively.

[0168] This embodiment presents the following scenarios for analysis and research:

[0169] Obstacle avoidance by drone swarms in a single scenario;

[0170] Monte Carlo simulation of a drone swarm under multiple obstacles of varying sizes;

[0171] (1) Parameter and result analysis in a single scenario:

[0172] Depend on Figure 6 It can be seen that the control methods described in steps 1 to 4 above are applied during simulation. In the initial stage of the simulation ( The swarm maintains stable flight, with the order parameter remaining near 1. When the first-level leader detects an obstacle, it utilizes the obstacle's repulsive force to begin turning. At this point, through the inertial spin mechanism established in step 3, the turning intention is rapidly transmitted between levels. Simulation results show that the swarm... The cluster exhibits a distinct "fluid-like" splitting behavior: centered on an obstacle, the cluster automatically splits into two streams that flow around the obstacles from both sides, and then quickly recombine after passing the obstacle.

[0173] In this process, such as Figure 7 As shown, the average neighbor spacing between drones remained stable around the expected value of 12 meters. Even during intense maneuvers traversing obstacles, the average spacing did not fluctuate drastically. Furthermore, the obstacle distance curve indicates that no individual drone entered the 3-meter damage radius of the obstacle, verifying the robustness and safety of the improved artificial potential field mechanism in dynamic environments. Figure 8 As shown, significant hierarchical transmission characteristics can be observed. During obstacle avoidance around steps 300 and 1130, a sharp bidirectional divergence in Y-axis velocity occurs (maximum lateral velocity reaches -7.66 m / s), with the first-level leader (blue curve) deflecting velocity first, followed closely by the second and third levels (green and red curves). This velocity diversion phenomenon replicates the avoidance behavior of starling flocks when encountering predators, demonstrating the effective balance between maintaining formation and maneuverability within a multi-level command and control architecture.

[0174] Regarding obstacle avoidance consistency, such as Figure 9 As shown, the method of this invention benefits from the inertial spin mechanism. Although the cluster order parameter briefly decreases (to a minimum of 0.81) when encountering an obstacle, it quickly recovers and stabilizes above 0.99. This indicates that the consensus mechanism can rapidly decay and oscillate within the cluster in a wave-like manner, overcoming the hysteresis problem of traditional consensus algorithms.

[0175] (2) Monte Carlo analysis in scenarios with changing obstacles:

[0176] like Figure 10 As shown in the obstacle radius Under various test conditions ranging from 20 meters to 40 meters, statistical results show that, with different obstacle sizes, the average compactness of the cluster remains within a very narrow range of 10.55m to 10.60m, and the average order parameter remains above 0.9, indicating that the algorithm has strong robustness to environmental changes. Furthermore, as the obstacle radius increases, the average time for the cluster to traverse the obstacle only shows a slight linear increase, without exponential growth or deadlock.

[0177] In summary, the starling flock-inspired distributed multi-layer control and obstacle avoidance method for UAV swarms proposed in this invention decouples high-level navigation and guidance from low-level consistency control tasks by constructing a static multi-level Leader-Follower network architecture, thereby achieving a comprehensive improvement in safety, consistency, and task efficiency.

[0178] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. A method for security control and steering consensus of unmanned aerial vehicle (UAV) swarms based on starling flocks, characterized in that: Includes the following steps: Step 1: Establish a point mass dynamics model of the UAV, and according to the preset hierarchical logic, divide the UAVs in the cluster into a three-layer structure that includes at least a first-level navigator, a second-level relay leader, and a third-level ordinary follower. At the same time, construct a physical neighbor interaction topology based on the airborne communication radius to form a network architecture that combines logical hierarchy and physical interconnection. Step 2: Based on the improved artificial potential field method, the interaction space between UAVs is divided into damage zone, repulsion zone, alignment zone and attraction zone. A smooth concave-convex function is introduced to adjust the potential field strength of the interval, and a piecewise continuous pairwise potential energy function is constructed. The neighbor cooperative force to maintain the expected distance between UAVs is obtained by calculating its negative gradient. For environmental obstacles, virtual nodes are generated on their surface and a repulsive potential field is constructed to obtain the obstacle repulsion force. Step 3: When the swarm maneuver demand exceeds a set threshold, activate the consensus mechanism based on the inertial spin of the starling flock; calculate the expected spin of an individual based on the deviation between the maneuver intention vector synthesized from the global navigation force, obstacle repulsion force, and neighbor cooperative force and the current speed; at the same time, calculate the perceived spin by sensing the spin state of the neighbors; update the spin state of each UAV through the spin dynamics equation containing a damping term, and transform it into a coupling force that drives the synchronous lateral movement of the swarm; Step 4: For the first-level navigator, the neighbor cooperative force, obstacle repulsion force and global navigation force it receives are weighted and fused to generate its total control input; For the second and third level drones, the neighbor cooperative force they are subjected to and the turning consensus coupling force generated in step 3 are weighted and fused to generate their respective total control inputs; using the obtained total control inputs, the position and velocity state of each drone in the cluster are updated by numerical integration.

2. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 1, characterized in that: In step 1, the particle dynamics model of the UAV is expressed as follows: in For drones The total control input vector, For drones The position vector, For drones The velocity vector.

3. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 1, characterized in that: In step 1, the first-level navigator is located at the front of the cluster and can directly obtain global navigation information and obstacle information, and is responsible for guiding the overall course of the cluster; The second-level relay leader is located in the middle of the cluster, acting as an information relay node, following the first-level navigator and transmitting maneuver information backward; the third-level ordinary followers are located at the tail of the cluster, interacting only with their neighbors and superior navigators locally, maintaining a tight formation.

4. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 1, characterized in that: In step 2, the relative distance between drones is used. Based on this, define the normalized distance. According to the damage radius Expected Spacing Alignment area bandwidth The interaction space is divided into four continuous regions: the damage zone. Rejection region: Alignment area: ; Attraction Zone: ;in This refers to the communication coverage radius of the drone.

5. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 4, characterized in that: The concavity / convexity function is: in , , , The adjustment coefficient controls the attenuation rate of the potential field strength in the repulsive and attractive regions.

6. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 4, characterized in that: The pairwise potential energy function is in, Indicates from drones Pointing to the neighbor The unit direction vector; the neighbor cooperative force is 。 7. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 4, characterized in that: The repulsive force of the obstacle is: in For drones The distance to the nearest point on the surface of the obstacle. The preset avoidance distance, Indicates from drones The unit direction vector pointing to the virtual node.

8. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 1, characterized in that: In step 3, the process of determining whether the cluster mobility demand exceeds the set threshold is as follows: Calculate the direction change function in For drones The number of neighbors, and drones and Velocity vector; setting steering threshold ,like Greater than the preset threshold If this happens, the cluster is determined to be in a redirection state.

9. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 8, characterized in that: In step 3, the process of calculating the desired spin and the sensed spin, updating the spin state, and converting it into the coupling force that drives the cluster's synchronized lateral motion is as follows: First, calculate the drone's motion intention. : in To foster neighborly cooperation The repulsive force of the obstacle, For global navigation power, Based on following force; Based on motion intention Calculate the expected spin of the drone : The average spin deviation of the neighboring drones is calculated to obtain the sensed spin. : in, The coupling strength coefficient is... For the neighbors ; Update the UAV based on the spin dynamics equations including damping terms. spin vector : in The damping coefficient; For random noise that follows a normal distribution; Using rotational inertia The spin vector is converted into tangential acceleration and superimposed with the basic following force, thus transforming the spin information into an actual control force as the coupling force driving the synchronous lateral motion of the cluster. : 。 10. The method for security control and steering consensus of UAV swarms based on starling flocks according to claim 9, characterized in that: In step 4, for the first-level navigator, its control input... It is composed of a weighted average of neighbor cooperation force, obstacle repulsion force, and navigation force pointing towards the global goal: in These are weighting coefficients; for second- and third-level UAVs, their control inputs... It is composed of a weighted average of neighbor cooperation force and coupling force: in The weighting coefficients are the coupling force; Introducing communication delay Simulating biological response lag, using a superior navigator exist Velocity state at any moment As a reference source: 。

Images