Unmanned aerial vehicle cluster task dynamic allocation method imitating prey selection behavior of coyote group
By employing a distributed task allocation method that mimics the prey selection behavior of coyote packs, the problem of efficient resource allocation for UAV swarms under the coordinated threat of enemy swarms was solved. This enabled precise strikes against vulnerable targets with limited resources, thereby improving the combat effectiveness and system reliability of UAV swarms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2023-11-02
- Publication Date
- 2026-05-15
AI Technical Summary
Existing drone swarm mission allocation methods are inefficient in allocating strike resources when there is a coordinated threat from enemy swarms, resulting in high resource consumption, significant losses, and a lack of flexibility and reliability, making them unsuitable for dynamic battlefield situations.
A distributed task allocation method based on coyote pack prey selection behavior is adopted. By establishing a UAV swarm control model and combining prey selection and coordinated encirclement behavior in coyote hunting, the potential coordinated threat of enemy swarms is dynamically assessed, and weaker targets are prioritized for attack, generating an efficient strike formation and reducing resource consumption and losses.
It improves the combat capability and flexibility of drone swarms in aerial combat, reduces the possibility of enemy swarm support and counter-encirclement, and enhances combat effectiveness and system robustness and reliability.
Smart Images
Figure CN117850442B_ABST
Abstract
Description
Technical Field
[0001] This invention is a method for dynamic task allocation in UAV swarms that mimics the prey selection behavior of coyotes, belonging to the field of autonomous control of UAV swarms. Background Technology
[0002] Currently, unmanned aerial vehicles (UAVs) have evolved from performing combat support missions such as aerial reconnaissance, target surveillance, and battle damage assessment to becoming major combat equipment capable of carrying offensive weapons to suppress enemy ground and air forces. Tactical strike integrated UAVs combine reconnaissance and attack capabilities, possessing surveillance, reconnaissance, target encirclement, and timely strike capabilities. They can effectively suppress enemy defense systems and conduct collaborative operations against high-risk targets, significantly enhancing combat capabilities in future unmanned military confrontation environments. Therefore, this invention aims to promote the development of tactical strike integrated combat strategies by utilizing UAV swarm collaborative combat modes, proposing a highly robust collaborative task dynamic allocation method to achieve intelligent team tactical planning for tactical strike integrated UAV swarm operations.
[0003] Since future unmanned combat environments are highly adversarial, distributed collaborative architectures are crucial for achieving large-scale group collaboration. Currently, while centralized collaborative architectures, widely researched and applied, offer the advantage of simple design, they also suffer from limitations such as high demands on communication networks and central computing power. Furthermore, if the central node fails, the entire unmanned system will be paralyzed. In contrast, distributed architectures lack a clearly defined central node, resulting in higher reliability and scalability. Specifically, this invention proposes a highly reliable distributed architecture implementation method, using the UAV performing the reconnaissance mission that discovers the first enemy target as the local center for subsequent strike missions, completing tasks such as pre-allocation and allocation of strike missions.
[0004] Traditional UAV swarm mission allocation primarily considers factors such as UAV maneuvering distance, area coverage, path planning, and no-fly zone constraints, but rarely takes into account potential collaborative threats and support time constraints between enemy platforms. For example, distributed mission allocation methods based on decentralized auction mechanisms use individual strike probability, resource availability, range, and target threat as mission allocation indicators, selecting the UAV with the optimal indicators to execute the mission. Another example is allocation methods based on response thresholds, using indicators such as flight path and relative situational advantage as stress probability factors, allocating missions based on probability. These methods mostly assess and allocate missions based on the current static situation, but in actual air combat, it is often necessary to consider the dynamic changes and potential threats of the combat scenario and enemy swarm situation. Therefore, this invention considers factors such as geometric and topological distances between UAVs within the enemy swarm to assess potential collaboration and cooperative support within the enemy swarm.
[0005] Ecological studies indicate that coyote packs often hunt cooperatively, working together to surround and share prey to ensure energy conservation for their daily activities and hunting activities. Furthermore, during hunts, coyotes tend to target vulnerable members of the prey group, such as prey isolated from the pack's protection, females, or young prey. This invention models the intelligent behavior of coyote packs by analyzing their prey selection behavior and dynamic resource allocation during large-scale predation, mapping the relationships as follows: Figure 1 As shown, a dynamic task allocation method for tactical strike integrated UAV swarm operations based on the prey selection behavior of coyote packs is proposed. It aims to solve the task allocation problem in the process of air combat, cut off or destroy the enemy's potential for coordinated operations, and strike the enemy UAVs when they are in a weak position as much as possible, thereby reducing our strike resource consumption and damage losses, and avoiding the incalculable losses caused by the enemy UAVs in the subsequent favorable situation, so as to complete the air combat mission with minimal losses. Summary of the Invention
[0006] Current drone swarm mission allocation methods often assume that there is no coordination among enemy target groups, treating the enemy as a simple target and only the friendly swarm possessing intelligent and autonomous characteristics. However, in future battlefield confrontations involving unmanned systems, this assumption is often unsatisfactory. Therefore, the purpose of this invention is to provide a distributed drone swarm dynamic mission allocation method that mimics the prey selection behavior of coyote packs. This method aims to solve the problem of efficient mission allocation under limited strike resource constraints in tactical integrated operations, where enemy swarms may potentially coordinate in drone swarm aerial combat environments. It also addresses how to identify and prioritize vulnerable target subgroups for attack. This approach reduces risk and cost while improving the flexibility and reliability of drone swarms in executing combat missions, further enhancing the intelligence and autonomy of drone swarms.
[0007] This invention is a method for dynamic task allocation in UAV swarms that mimics the prey selection behavior of coyote packs. The specific implementation steps are as follows:
[0008] Step 1: Initialization of Aerial Combat Conditions for Drone Swarm
[0009] S11. Initialization of UAV Control Model
[0010] First, a nonlinear model of the UAV is established based on aircraft dynamics and kinematics. The decomposition forms of forces and moments differ in different coordinate systems, and the descriptions of aircraft attitude angles and velocity vectors also vary. The fixed-wing model used in this invention is described below in a volume coordinate system. The control variables U of the UAV include the elevator deflection angle δ. e rudder deflection δ r Aileron deflection angle δ aand throttle lever δ T :
[0011] U T =[δ T δ e δ r δ a (1)
[0012] The controlled variables of the UAV are 12 state variables, namely the UAV's three orientation parameters x. g ,y g ,h g Roll angle φ, pitch angle θ, yaw angle ψ, airflow velocity angle V, angle of attack α, sideslip angle β, roll angular velocity p, pitch angular velocity q, yaw angular velocity r, i.e.:
[0013] X T =[x g ,y g ,h g ,φ,θ,ψ,V,α,β,p,q,r] T (2)
[0014] The mathematical description of the drone control model is as follows:
[0015]
[0016] In flight control, airflow velocity angle V, angle of attack α, and sideslip angle β are also frequently used. The formulas are supplemented below:
[0017]
[0018]
[0019]
[0020] S12, Initialization of UAV swarm combat mission
[0021] If the drone neither tracks nor engages a target, it is in a reconnaissance mission state. Based on the reconnaissance capabilities of our drone, the probability of detecting a target during coordinated reconnaissance is as follows:
[0022]
[0023] Let the numbers be z = i1, i2, ..., i Q If Q UAVs conduct cooperative reconnaissance of a target j, the probability of detecting the target is P(Q,j). μ is the cooperative reconnaissance capability gain coefficient to reflect the capability gain of multi-UAV cooperation, and PD... U (i z () represents the reconnaissance capability index of a single drone.
[0024]
[0025] Among them, C zj This indicates whether the z-th drone has detected target j. If it successfully detects the target drone, the function value is recorded as 1; otherwise, it is 0. If the z-th drone successfully detects the target drone or the target drone cluster, it switches to a tracking task, broadcasts the target information, and begins to pre-determine the attack range.
[0026] If the UAV is in tracking mission mode, it first determines the combat environment near the target within the reconnaissance range, including location, total number of enemy aircraft, and status of enemy aircraft, and broadcasts the information to nearby UAVs. At the same time, it receives status information transmitted back by the UAVs, including distance and combat readiness status, to determine the pre-formation queue.
[0027] If the drone is in a strike mission state, first determine the drone number j of the target drone it is responsible for attacking. att And the time t for launching the attack att Once an attack is launched, the probability of a successful strike is as follows:
[0028]
[0029] Where η is the capability gain coefficient generated by multi-machine cooperative attacks, PA U (i z ) represents the probability of a single drone destroying a target.
[0030]
[0031] Among them, A zj This indicates whether the z-th drone successfully destroyed target j. If it successfully destroyed the target and completed the strike mission, the function value is recorded as 1; otherwise, it is 0. If the target drone cluster was not successfully destroyed, the total resource E is checked. sum Does it meet the strike capability requirements for this attack? att If the conditions are met, the next attack will be organized; otherwise, the attack marker on the target will be removed and the relevant threat information on the map will be updated. Specifically, during the organization of the attack, if the strike capability of the tracking drone meets the attack capability requirements, it will be included in the strike queue; otherwise, the tracking mission will continue to coordinate the attack until the enemy cluster is successfully destroyed or the mission is canceled due to insufficient strike resources.
[0032] S13, Initialization of the drone swarm aerial combat map
[0033] In this invention, the combat scenario is a three-dimensional aerial combat environment. The mission execution area is divided into a three-dimensional grid, and the three-dimensional coordinates of the UAV are determined based on the grid. For example, the coordinate determination method for the ind-th grid is as follows:
[0034]
[0035] Wherein, ind is the grid number, Length is the number of 3D grids in the length direction, and Width is the number of 3D grids in the width direction. Furthermore, the Q targets are in the operational environment, either distributed or in clusters, and may have potential cooperative relationships among them; their positions and states change over time during aerial combat.
[0036] Step Two: Execute the target selection task based on the coyote pack behavior mechanism.
[0037] S21, UAV swarm reconnaissance and combat environment
[0038] In a combat environment, there are Q enemy drones distributed throughout the area. The incentive is to eliminate a larger swarm of enemy drones as quickly as possible. The design objective value is as follows:
[0039]
[0040] in, V represents the initial value of the objective, which gradually decreases over time. z (t) represents the value at the current moment. α is the value decay coefficient.
[0041] As mentioned above, the drone swarm has three mission types: reconnaissance, tracking, and strike. When no target is detected, the drones are in reconnaissance mode. Once a target is detected, they broadcast information and begin to pre-determine the attack range based on coyote prey selection behavior.
[0042] S22. Predetermine attack range based on prey selection behavior.
[0043] This study maps coyote packs to drone swarms, treating the target swarm as the prey population. Based on the cooperative hunting behavior of coyotes, the connected components of the target swarm are considered as a local whole, serving as the target of a single strike mission. The behaviors of any two individuals within the swarm directly or indirectly influence each other, thus achieving cooperation across the entire swarm. Geometric distance is used as one of the conditions for determining the range of interacting neighbors, with variables representing the interactivity between individuals.
[0044]
[0045] Where, r i The location of the target is represented by , and 'l' is the maximum geometric distance for interaction. Targets are considered neighbors if their geometric distance is less than 'l'; otherwise, no interaction occurs, indicating symmetry in the interaction. However, when a one-sided connection exists, further judgment using topological distance is necessary to avoid accumulating errors and causing repeated task assignment failures.
[0046] Topological distance refers to the number of other individuals existing between two individuals. Distinguishing it from geometric distance and considering the encirclement and killing behavior of coyotes in hunting, the topological distance from the attacker's perspective during a predator-killing hunt is defined as follows: For individual q... i The other individuals in the cluster are arranged according to their relationship with individual q. i The distance between them is numbered, that is, the distance from individual q i The most recent individual is numbered 1, the second most recent individual is numbered 2, and so on. Therefore, there are interactive variables.
[0047]
[0048] in, Let be the topological distance between the two targets, and m be the maximum topological distance. The pre-attack range is determined through interactive variables, and the target risk is assessed through ranking to determine the priority allocation of strike resources in encirclement strikes. Target risk is normalized using relative ranking to avoid excessive weighting of some function values due to the increasing number of drones within the target cluster. Furthermore, considering the mechanical energy possessed by the drones during flight, including potential and kinetic energy, we have:
[0049]
[0050] Among them, H e This reflects the mechanical energy per unit mass of the enemy fighter jet, and is based on the drone's q. i With subgroups having maxH e The definition of drone energy-to-altitude ratio:
[0051]
[0052] Construct a mechanical energy evaluation function:
[0053]
[0054] In heterogeneous UAV swarms, considering the differences in combat resources among different UAV models, a kill rate matrix is introduced:
[0055]
[0056] Among them, att zq This represents the kill rate of a Z-type fighter against a Q-type target. Taking all the above considerations into account, regularization and sorting are used to determine the pre-attack range. By detecting enemy swarms of target drones, relevant information such as the pre-attack range and enemy drone types is broadcast. The geometric distance of friendly drones is considered, and arrival and attack initiation times are negotiated, etc.
[0057] S23, Reconnaissance UAV broadcasts pre-selection results information
[0058] After pre-selecting the attack range, the threat coefficient, danger level, and distribution of the target subgroups are assessed and classified based on the coyote prey selection behavior mechanism. This provides sufficient selection basis for subsequently selecting weaker subgroups and coordinating a coordinated attack. Through a local central node, i.e., a reconnaissance drone, relevant information is broadcast to summon drones meeting the strike capability requirements into the pre-attack formation queue, with an agreed arrival time, to quickly launch an attack on the target cluster, thereby disrupting the enemy's coordinated combat effectiveness and the possibility of reinforcement support.
[0059] Step 3: Generate strike formations based on the effective time window constraints of coyote hunting
[0060] S31, Statistics on Drone Strike Capability in Areas Related to Attack Range
[0061] Other drones on reconnaissance missions receive broadcast information from the tracking drone and acquire detection information about the target cluster. If the drone is not currently detecting other targets or is in an attack queue (i.e., has no tracking or attack missions), it determines whether its own attack capabilities meet the requirements of this attack mission. If so, it sends a response message and waits to join the pre-attack formation queue.
[0062] S32. Generating strike pre-formations based on the coyote strike resource allocation mechanism.
[0063] Compared to various task allocation methods based on auction mechanisms and probability, this invention allocates strike tasks based on strike resources, which is more conducive to forming coordinated strike formations. The former, based on a greedy approach, often generates task allocations that primarily involve single-machine execution, which is not conducive to leveraging the advantages of drone swarms and can lead to an intractable predicament in the later stages of a strike mission due to insufficient strike resources.
[0064] First, based on the response messages from the drone swarm, the drones in the queue are traversed to generate all feasible solutions. Then, for each feasible formation, the total strike resources and the minimum total strike resource requirement are considered, which are related to the drones' combat readiness status and drone type. The total strike resource calculation method is as follows:
[0065]
[0066] Where n is the total number of drones in the feasible formation, E z This refers to the resource requirements for a single-machine strike. The minimum total strike resource requirement is:
[0067] E′ sum = (1+ρ)∑E(z,q) (20)
[0068] Where E(z,q) represents the resource consumption of a z-type UAV striking a q-type target, and ρ is the safety protection gain coefficient. If a feasible strike formation meets the total strike resource requirements, it is added to the pre-strike formation, until all feasible formations are traversed. Then, the geometric distances of the UAVs in the feasible formations are further considered, and their arrival times are estimated to determine the final strike formation and calculate the balanced consumption of strike resources. Through cooperative strikes, the consumption of individual UAVs during the strike process is reduced, thereby providing more possibilities and a higher upper limit for the later stages of the strike mission.
[0069] S33. Generating formal strike formations based on the effective time window constraints of coyote hunting.
[0070] Considering the average geometric distance and average topological distance of the enemy cluster:
[0071]
[0072]
[0073] in, The average geometric distance, The average topological distance is taken as an example. Considering the segmentation criteria of enemy cluster subgroups and the determination of pre-attack range, the calculation method for the potential enemy reinforcement time window is as follows:
[0074]
[0075] in, To estimate the minimum time for potential reinforcements, The airspeed of a single enemy drone is given by ζ, which is the estimated speed gain coefficient. The estimated time for a feasible formation to engage the target is... If the time constraint is met, the strike mission can be completed and the unit safely withdrawn before enemy reinforcements arrive. If multiple feasible formations satisfy the time window constraint, then... With minimum arrival time mint arr The formation carried out this strike mission.
[0076] Step 4: The drone swarm executes strike / tracking missions.
[0077] S41, Radio Strike Formation
[0078] Once the strike formation is identified, the drone numbers of the strike formation are broadcast, and the mission status of the relevant drones is changed to strike mission. The drones then arrive at the strike location at the designated time. The strike location is broadcast and updated by the tracking drones. The remaining drones continue their reconnaissance missions.
[0079] For the resource consumption of a strike formation, the consumption of each UAV within the formation during coordinated strikes is calculated using the following formula. The resource consumption calculation method for each UAV belonging to the strike formation is as follows:
[0080]
[0081] S42, Strike Mission Path Planning
[0082] The scheduled attack time is calculated as follows:
[0083]
[0084] Where t0 is the current time, This calculates the time it takes for the drone with the longest arrival time in the attack formation to reach the attack location. If the current time is the attack time, the waypoint planned for that drone is z. j attack target q i If the attack time has not yet arrived at the current location, the waypoint is calculated using the following formula:
[0085]
[0086] Where, z′ jx , z′ jy , z′ jz Let z be the coordinates of the j-th UAV at the next moment. jx , z jy , z jz Let z be the coordinate at the current time. t Let be the coordinates of the target point, and ΔT be the time length of one cycle.
[0087] S43, Tracking Task Path Planning
[0088] The tracking path planning method is the same as above, but the target waypoint is no longer the location where the attack is launched, but the location of the tracking target.
[0089] Step 5: Update the aerial combat scenario map
[0090] S51, Update drone cluster status
[0091] Based on broadcast information, the drone status is updated. If a drone has neither a tracking nor an attack mission, it executes a reconnaissance mission. If a reconnaissance drone detects an enemy target cluster, it updates to a tracking mission status. If the tracking drone's attack resources meet the attack mission resource threshold, it updates to an attack mission status; otherwise, it remains in the tracking mission status, serving only as a local center for the attack mission without participating in the attack itself, until the mission ends and it updates to a reconnaissance mission status. If an attack drone successfully completes its attack mission, it updates to a reconnaissance mission status. If the attack fails, it checks whether the attack resources meet the attack requirements. If they do, the attack continues; otherwise, the mission is canceled and reconnaissance is restarted.
[0092] S52, Update the status of known enemy target clusters.
[0093] For targets discovered for the first time, update the target information and organize an attack; for targets that are not attacked, record the information and increase the threat level so that they can be used as a reference when organizing the next attack mission.
[0094] S53, Track whether the target broadcast by the drone is in the middle of a strike mission.
[0095] If the target in the area is under attack, the reconnaissance drone does not need to go to the area and mark it on the map.
[0096] Step Six: Reassignment of Drone Swarm Operational Tasks
[0097] S61, No strike pre-formation formed
[0098] If the drone strike resources in the target area cannot meet the minimum threshold requirements or the time window constraints, tracking will be cancelled and the strike mission will not be carried out for the time being.
[0099] S62, unable to form an effective strike formation
[0100] The handling method is the same as above: cancel tracking, stop broadcasting target cluster information, and re-enter the reconnaissance mission state.
[0101] S63, Changes in Enemy Cluster Strike Capability
[0102] If, during tracking or strike operations, the strike cannot be completed within the specified time, and enemy reinforcements arrive; or if enemy reinforcements arrive earlier than expected, our tracking and strike drones will cancel their missions and adopt a high-maneuverability escape strategy to evade the enemy's encirclement and attack. The control law model for this high-maneuverability escape strategy is as follows:
[0103]
[0104]
[0105]
[0106] Among them, a u Indicates direct control force, k pv Indicates the control coefficient. To accelerate growth, For turning acceleration, This refers to the acceleration during ascent.
[0107] S64. Output task allocation results
[0108] Determine whether the current time t has arrived or whether all enemy drones have been eliminated. If so, output the total value of the targets acquired and the remaining resources for each drone; otherwise, update the time and continue with the tactical strike integrated task allocation.
[0109] The advantages and effects of the UAV swarm task dynamic allocation method based on the coyote pack prey selection behavior of the present invention are as follows: 1. The present invention provides a distributed architecture task allocation method, which improves system robustness while having high scalability and high reliability; 2. The present invention considers the high real-time and strong dynamic nature of combat scenarios and the characteristic of coyotes adjusting their hunting behavior when unexpected events occur, and proposes a dynamic task allocation method that adjusts the task status in real time according to environmental changes, and has good universality; 3. The present invention considers the severe environment of UAV swarm aerial combat and, based on the entire process of coyote hunting behavior in the wild, makes this method applicable to the tactical reconnaissance and strike integrated combat environment, integrating reconnaissance, tracking, and strike, and systematizing the entire process of UAV swarm aerial game, which has strong adaptability to severe combat environments; 4. The present invention considers the potential cooperative relationship of the enemy swarm and, based on the effective time window constraint of coyote pack hunting, reduces the possibility of enemy swarm support and counter-encirclement, and significantly improves the combat capability of UAV swarm in aerial combat environment. Fifth, this invention takes into account the dynamic real-time situation of the combat environment and, based on the prey selection behavior of coyotes, prioritizes the attack of targets in a vulnerable position by the drone swarm, so as to avoid them from suppressing our firepower in subsequent operations under favorable circumstances, thereby effectively improving the combat effectiveness of the drone swarm. Attached Figure Description
[0110] Figure 1 This invention relates to a dynamic task allocation method for drone swarms that mimics the prey selection behavior of coyote packs. (Mapping diagram) Figure 2 Flowchart of the UAV swarm task dynamic allocation method based on the coyote pack prey selection behavior of this invention
[0111] Figure 3 Example of an aerial combat situation at the initial moment
[0112] Figure 4 Example 30-minute aerial combat situation map
[0113] Figure 5 Example 60-minute aerial combat situation map
[0114] Figure 6 Example 90-minute aerial combat situation map
[0115] Figure 7 Example 120-minute aerial combat situation map
[0116] Figure 8 Example 150-minute aerial combat situation map
[0117] Figure 9 Aerial combat situation diagram at the end of the example air combat
[0118] Figure 10 Example of obtaining a comparison curve of the total value of the targets struck
[0119] Figure 11 Example: Curve showing the total resources used in drone swarm attacks as a function of simulation time. Detailed Implementation
[0120] The effectiveness of the task allocation method proposed in this invention is verified through a specific example below, illustrating the specific steps of the task allocation process. In this example, our own UAV z i (i = 1, 2, ..., 10) Ten enemy drones, q j Fifteen drones (i = 1, 2, ..., 15) engage in integrated tactical strike aerial combat within a given 3D grid environment of drone swarm air combat. The air combat scenario is 30 × 30 × 1 km in size. 3 The simulation environment for this example is configured with an Intel i7-12900 processor and 16GB of RAM. The software used is MATLAB 2020a.
[0121] The specific practical steps of the drone swarm task dynamic allocation method that mimics the prey selection behavior of coyotes are as follows:
[0122] Step 1: Initialize the drone swarm
[0123] The drone swarm status information is initialized, including randomly generated positions in the air combat scenario, randomly distributed speeds between [50, 350] km / h, initial drone strike resources of 35, resource consumption for three types of enemy drones of 30, 20, and 10 respectively, and three-axis Euler angles randomly distributed within [-π, π]. The simulation maximum time T is set. max =180min.
[0124] UAV mission state initialization. If no enemy target is detected, the UAV is in reconnaissance mission state. If an enemy UAV cluster is detected under the initial conditions, the next simulation iteration will update to tracking state and broadcast target information. The probability of detecting the target at the waypoint where the target is located during the reconnaissance mission can be obtained according to equation (7), and the cooperative reconnaissance capability gain coefficient μ = 0.3.
[0125] Step Two: Execute the target selection task based on the coyote pack behavior mechanism.
[0126] The initial target drones are divided into three categories, five of each, with an initial value of The values are 100, 90, and 80 respectively, and decrease over time according to equation (12), where the value decay coefficient α = 0.75.
[0127] Upon target discovery, the attack range is first determined based on prey selection behavior, i.e., the range of the enemy drone swarm. The specific determination principle is based on coyote prey selection behavior formulas (8), (9), (15), and (17), where the capability gain coefficient η = 0.3 and g is the gravitational acceleration of 9.8 m / s². 2 Then, the reconnaissance drone switched to tracking mission status and broadcast relevant information about enemy targets, summoning other drones to carry out strike missions.
[0128] Step 3: Generate strike formations based on the effective time window constraints of coyote hunting
[0129] First, based on the coyote strike resource allocation mechanism, according to the target strike resource threshold E d Select UAVs that meet the conditions and add them to the pre-formation queue. Then, traverse the queue to generate all feasible solutions. Calculate the total strike resources of the feasible formation according to equation (19) and consider the total strike resource threshold equation (20), where ρ = 0.1. Add the formations that meet the conditions to the pre-strike formation. Further consider the time window constraint and determine the final formation that meets the requirements according to equation (23), where the speed gain coefficient ζ = 0.5. The remaining UAVs are unmarked and enter the reconnaissance mission state.
[0130] Step 4: The drone swarm executes strike / tracking missions.
[0131] After determining the strike formation, the resource consumption of the strike drone is calculated according to equation (24), and the probability of the strike drone successfully striking the target is determined according to equation (9). The trajectory planning of the tracking mission and the strike mission is based on equation (26), where one iteration period ΔT = 1 min.
[0132] Step 5: Update the aerial combat scenario map
[0133] After completing a round of mission iterations, the drone swarm mission status is updated. Based on broadcast information, if a drone has neither a tracking nor an attack mission, it executes a reconnaissance mission. If a reconnaissance drone detects an enemy target swarm, it updates to a tracking mission status. If the tracking drone's attack resources meet the attack mission resource threshold, it updates to an attack mission status; otherwise, it remains in the tracking mission status, acting only as a local center for the attack mission without participating in the attack itself, until the mission ends and it updates back to a reconnaissance mission status. If an attack drone successfully completes its attack mission, it updates to a reconnaissance mission status. If the attack round fails, it checks whether the attack resources meet the attack requirements. If they do, the attack continues; otherwise, the mission is canceled and reconnaissance is restarted. Additionally, information about enemy-detected drone swarms is broadcast, marked on the map, and the threat level is updated.
[0134] Step Six: Reassignment of Drone Swarm Operational Tasks
[0135] Finally, if the strike mission is completed, the value acquired from the strike target is recorded. If the strike fails but the strike resources still meet the requirements, the strike mission continues; otherwise, the reconnaissance mission is reassigned. If enemy reinforcements arrive during the strike, escape is carried out according to equation (27), and then the reconnaissance mission is reassigned. When the simulation time reaches the total simulation time requirement, the simulation ends and the simulation results are output, such as the change curve of the total value of the strike target and the remaining strike resources of each UAV, etc.
[0136] The simulation results of this example will be explained next. Figures 3 to 9 To record the operational situation at key moments during the air combat, the entire simulation lasted 180 minutes, with data recorded every 30 minutes. Our ten UAVs, based on their battlefield environment, performed reconnaissance, tracking, or strike missions, as well as high-maneuver escapes after being counter-encircled by the enemy swarm. The target swarm updated its position over time; its movement strategy was difficult to detect, and it possessed certain counter-reconnaissance capabilities, making reconnaissance missions challenging. As the mission progressed, the target was gradually discovered in the environment. The operational situation map shows that initially, our forces were outnumbered, but as the air combat continued, we gradually gained the upper hand, moving from weakness to local advantage, from local advantage to overall advantage, and from confronting the enemy swarm to complete encirclement. Figures 10 to 11 Simulation results demonstrate that this invention offers significant advantages in optimizing formation strike capabilities and countering potential enemy coordination. Due to the differences in strike capabilities and remaining strike resources among various UAVs, an optimal task allocation strategy is needed to maximize the strike capability against enemy swarms while minimizing the potential damage to friendly swarms. The simulation results show that friendly swarms still possess sufficient capability to form effective formations and deliver efficient strikes against the enemy in the later stages of air combat. Formation-based tactical strike integration strategies effectively curb the possibility of enemy coordinated operations. Furthermore, by exploiting enemy unpreparedness based on topology and energy-altitude ratio, the invention eliminates enemy UAVs in vulnerable scenarios at a relatively low cost, preventing them from inflicting fatal blows on friendly forces from a high-altitude position in subsequent operations, thus improving mission execution efficiency.
Claims
1. A method for dynamic task allocation in a drone swarm based on coyote pack prey selection behavior, characterized in that: The implementation steps of this method are as follows: Step 1: Initialize the conditions for aerial combat in UAV swarms, specifically including: initializing the nonlinear kinematic and dynamic models of the UAVs; initializing the swarm combat mission; initializing the aerial combat map; Step Two: Perform a target selection task mimicking coyote pack behavior, specifically including: S21, UAV swarm reconnaissance combat environment; when no target is detected, all UAVs are in reconnaissance mode; when a target is detected, proceed to S22; S22. Predetermine the attack range based on prey selection behavior; map the coyote pack to the drone swarm, take the target swarm as the prey population, and based on the behavioral characteristics of coyotes in cooperative hunting, consider the part of the target swarm that is a connected graph as a local whole, and take it as the target of a strike mission; regularize and sort to determine the pre-attack range, broadcast the pre-attack range and enemy drone type information by discovering the target drones of the enemy swarm, take into account the geometric distance of our drones, and negotiate the arrival time and attack initiation time. S23, Reconnaissance drone broadcasts pre-selection results; Step 3: Generate an attack formation based on the effective time window constraint of coyote hunting, specifically including: S31. Statistics on drone strike capabilities in areas related to attack range; S32. Generate strike pre-formation based on the coyote strike resource allocation mechanism; S33. Generate formal strike formations based on the effective time window constraints of coyote hunting; Step 4: The drone swarm executes strike / tracking missions; Step 5: Update the aerial combat scenario map; Step Six: Reassignment of UAV swarm combat missions, specifically including: failure to form a pre-strike formation; inability to form an effective strike formation; changes in the enemy's swarm strike capabilities; output of mission assignment results; The specific process of step S22, which predetermines the attack range based on prey selection behavior, is as follows: This study maps coyote packs to drone swarms, treating the target swarm as the prey population. Based on the cooperative hunting behavior of coyotes, it considers the connected components of the target swarm as a local whole, serving as the target of a single strike mission. The behavior of any two individuals within the swarm directly or indirectly influences each other, thus achieving overall swarm cooperation. Geometric distance is used as one of the conditions for determining the range of interacting neighbors, including variables related to inter-individual interactivity. : (1) in, Indicates the location of the target. The maximum interaction geometric distance; when the geometric distance between targets is less than If they are neighbors, they are neighbors; otherwise, no interaction occurs, meaning that the interactivity is symmetrical. However, when there is a one-sided connection, it is still necessary to further determine the topological distance to avoid accumulating errors and causing the task allocation to repeatedly fail. Unlike geometric distance, and considering the coyote pack's behavior of encircling and killing some prey in hunting, the topological distance from the attacker's perspective during a predator-hunting scenario is defined as follows: For an individual To group other individuals in the cluster according to their relationship with individual The distance between them is numbered, that is, the distance from the individual. The most recent individual is numbered 1, the second most recent individual is numbered 2, and so on; therefore, there is an interactive variable. : (2) in, The topological distance between the two targets is... Maximum topological distance; the pre-attack range is determined through interactive variables, and the target risk is assessed through ranking to determine the priority allocation trend of strike resources in encirclement strikes; the target risk is normalized by relative ranking; in addition, the mechanical energy of the UAV during flight, including potential energy and kinetic energy, is considered: (3) in, This reflects the mechanical energy per unit mass of the enemy fighter jet, and is based on the drone. With subgroups The definition of drone energy-to-altitude ratio: (4) Construct a mechanical energy evaluation function: (5) In heterogeneous UAV swarms, considering the differences in combat resources among different UAV models, a kill rate matrix is introduced: (6) in, This indicates the kill rate of the Z-type fighter against the Q-type target.
2. The method for dynamic task allocation of UAV swarms based on the coyote pack prey selection behavior described in claim 1, characterized in that: The specific process of generating the formal strike formation based on the effective time window constraint of coyote hunting in step S33 is as follows: Considering the average geometric distance and average topological distance of the enemy cluster: (7) (8) in, The average geometric distance, The average topological distance is used. Considering the segmentation criteria of enemy cluster subgroups and the determination of pre-attack range, the calculation method for the potential enemy reinforcement time window is as follows: (9) in, To estimate the minimum time for potential reinforcements, For observing the airspeed of a single enemy drone, To estimate the velocity gain coefficient; if the formation is feasible, the estimated time to strike the target If the time constraint is met, the strike mission is completed and the enemy is safely evacuated before reinforcements arrive.
3. The method for dynamic task allocation of UAV swarms based on the coyote pack prey selection behavior described in claim 2, characterized in that: If multiple feasible formations satisfy the time window constraint, then take... With minimum arrival time The formation carried out this strike mission.