A heterogeneous unmanned aerial vehicle cluster cooperative flexible scheduling method

Abstract

The application discloses a heterogeneous unmanned aerial vehicle cluster cooperative flexible scheduling method, and belongs to the field of unmanned system intelligent decision-making. First, an unmanned aerial vehicle comprehensive task allocation model is constructed; then, on the basis of the comprehensive task allocation model, a comprehensive task allocation method is constructed to solve the unmanned aerial vehicle-target matching problem; then, based on the solved unmanned aerial vehicle task alliance, an unmanned aerial vehicle flexible resource scheduling model is constructed to visualize the timing scheduling problem of the unmanned aerial vehicle alliance; finally, an unmanned aerial vehicle timing scheduling method is established to solve the unmanned aerial vehicle timing scheduling scheme. The application solves an efficient and feasible cooperative task planning scheme under the satisfaction of various restriction conditions, and solves the problem of when each unmanned aerial vehicle in the heterogeneous unmanned aerial vehicle cluster uses which resource to execute which task.

Description

Technical Field

[0001] This invention belongs to the field of intelligent decision-making for unmanned systems, and particularly relates to a flexible scheduling method for heterogeneous unmanned aerial vehicle (UAV) swarms. Background Technology

[0002] With the increasing intelligence of unmanned systems, the independent mission execution mode of a single UAV is no longer suitable for the complex and ever-changing battlefield environment. Swarm collaboration is becoming a key form of mission execution in future unmanned warfare. In swarm collaborative operations, each UAV typically carries a variety of heterogeneous resources with different functions and high coupling, such as reconnaissance sensors and electronic countermeasures equipment. The use of these resources often restricts and influences each other. For example, a UAV swarm member using reconnaissance resources may be detected by a target UAV, thus affecting the effectiveness of its strike mission; a swarm member carrying electronic countermeasures equipment can improve the effectiveness of its strike resources by suppressing and jamming targets. Therefore, answering the question of "which UAV uses which resources to perform which mission at what time"—that is, achieving optimal collaborative planning of resources across UAVs, time, and space—is crucial for ensuring efficient mission completion, avoiding resource conflicts, and maximizing overall mission effectiveness. Compared to traditional simple formation scheduling methods, swarm collaborative mission planning, by deeply integrating the resource capabilities of each UAV, enables parallel mission execution and dynamic resource allocation, significantly improving the swarm's mission adaptability, survivability, and strike effectiveness, and has extremely high application value and research significance. However, due to the complex coupling relationships between resources, the dynamically changing battlefield situation, and strict spatiotemporal constraints, how to find an efficient and feasible collaborative task planning scheme under various constraints has become a core challenge in the field of intelligent decision-making for unmanned aerial vehicle (UAV) swarms. This invention aims to solve the problem of "when, when, what resources, and what tasks each UAV uses to perform" in heterogeneous UAV swarms, and proposes a flexible scheduling method for heterogeneous UAV swarms. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a cooperative flexible scheduling method for heterogeneous UAV swarms, belonging to the field of intelligent decision-making for unmanned systems. First, a comprehensive task allocation model for UAVs is constructed. Then, based on this model, a comprehensive task allocation method is developed to solve the UAV-target matching problem. Next, based on the solved UAV task alliance, a flexible resource scheduling model for UAVs is constructed to visualize the time-series scheduling problem of the UAV alliance. Finally, a time-series scheduling method is established to solve for the UAV time-series scheduling scheme. This invention solves an efficient and feasible cooperative task planning scheme under various constraints, addressing the question of when, how, and what resources each UAV in a heterogeneous UAV swarm should perform which task.

[0004] A flexible scheduling method for heterogeneous UAV swarms, such as Figure 1As shown, the specific steps are as follows:

[0005] Step 1: Construct a comprehensive UAV task allocation model;

[0006] Step 2: Construct a comprehensive task allocation method based on the comprehensive task allocation model;

[0007] Step 3: Construct a drone alliance time-series scheduling model;

[0008] Step 4: Establish a drone alliance time-series scheduling method based on the dung beetle algorithm.

[0009] Furthermore, in step one, the construction process of the UAV integrated task allocation model is as follows:

[0010] The integrated task allocation model for unmanned aerial vehicles (UAVs) includes a detection allocation model, a suppression allocation model, and an attack allocation model; the attack allocation model is also known as the encirclement allocation model.

[0011] Step 1.1: Select the constituent elements of the UAV integrated task allocation model;

[0012] Positional elements, motion elements, target detection elements, and perception capability elements are selected as variables for the detection allocation model; positional elements, motion elements, target suppression elements, and electronic countermeasures capability elements are selected as variables for the suppression allocation model; positional elements, motion elements, and target attack elements are selected as variables for the attack allocation model.

[0013] Step 1.2: Construct the detection allocation model, suppression allocation model, and attack allocation model sequentially based on the analytic hierarchy process (AHP).

[0014] Step 1.2.1, Construct the probe allocation model:

[0015] The detection advantage of the i-th UAV over the j-th target (dec) ij When the i-th drone does not have detection capability, dec ij The value is 0; the detection advantage is dec. ij The calculation method is as follows:

[0016] dec ij =D(x) i ,x t j ,v i ,v t j D j ,d i )

[0017] Where, x i x is the position of our i-th UAV in the spatial coordinate system; t jIt is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j It is the velocity vector of the j-th enemy target; D j The detection value of the j-th target; d i D is the detection range of our i-th UAV. When the i-th UAV does not have detection capability, the detection range is 0; D(·) is the detection advantage calculation function.

[0018] The detection advantage matrix formed by m friendly drones and n enemy targets is Det[det] ij ] m×n The allocation matrix is in, Let represent the detection decision variable of the i-th UAV towards the j-th target, indicating whether the i-th UAV should conduct reconnaissance of the j-th target; the detection allocation model of the UAV swarm towards the target swarm is expressed as:

[0019]

[0020] The detection allocation model shows that the total detection benefit is maximized and each target is tracked and detected by at least one UAV;

[0021] Step 1.2.2, Construct the suppression allocation model:

[0022] The suppressive advantage of the i-th drone over the j-th target (jam) ij When the i-th drone does not have suppression capability, jam ij Take 0, suppress the advantage jam ij The calculation method is as follows:

[0023] jam ij =J(x) i ,x t j ,v i ,v t j J j ,j i )

[0024] Where, x i x is the position of our i-th UAV in the spatial coordinate system; t j It is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j J is the velocity vector of the j-th enemy target; j It is the suppressive value of the j-th target; j iJ is the suppression distance of our i-th drone. When the i-th drone does not have suppression capability, the suppression distance is 0; J(·) is the suppression advantage calculation function.

[0025] The suppression advantage matrix formed by m friendly drones and n enemy targets is Jam[jam] ij ] m×n The allocation matrix is in, Let represent the suppression decision variable of the i-th drone against the j-th target, indicating whether the i-th drone should suppress the j-th target; the suppression allocation model of the drone swarm against the target swarm is expressed as:

[0026]

[0027] The suppression allocation model shows that the total suppression benefit is maximized, and each target is suppressed by at least one drone.

[0028] Step 1.2.3, Construct the attack allocation model:

[0029] The attack advantage of the i-th drone against the j-th target is kil ij When the i-th drone does not have attack capabilities, kil ij Take 0, attack advantage kil ij The calculation method is as follows:

[0030] kil ij =K(x) i ,x t j ,v i ,v t j ,K j ,k i )

[0031] Where, x i x is the position of our i-th UAV in the spatial coordinate system; t j It is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j K is the velocity vector of the j-th enemy target; j It is the attack value of the j-th target; k i It is the attack capability of the i-th drone, k i It is related to the attack resources carried by the i-th drone; K(·) is the attack advantage calculation function;

[0032] The attack advantage matrix formed by m friendly drones and n enemy targets is Kil[kil ij ]m×n The allocation matrix is in, Let represent the decision variable for the i-th drone against the j-th target, indicating whether the i-th drone attacks the j-th target; the attack allocation model for the drone swarm against the target swarm is expressed as:

[0033]

[0034] The attack allocation model shows that the total attack benefit is maximized and each target is attacked by at least one drone.

[0035] Furthermore, in step two, the construction process of the comprehensive task allocation method is as follows:

[0036] Step 2.1, the alliance proposes:

[0037] The Hungarian algorithm is used to solve the matching relationship between the detection drone and the target, with the detection drone as the leader of the alliance;

[0038] Step 2.2, Consortium Bidding:

[0039] Once a drone with a different payload receives a proposal message, it becomes a potential alliance member if its resources are not exhausted and it does not belong to any existing alliance.

[0040] Drones carrying heterogeneous resources establish a suppression advantage based on step one. ij With attack advantage kil ij The calculation method analyzes the suppression and attack benefits of the drone against each enemy target unit; according to the alliance order from highest to lowest attack benefit, the resource information and cost information of the drone itself are fed back to the alliance leader through the communication network in the form of alliance formation bidding messages.

[0041] Step 2.3, Alliance Formation:

[0042] The alliance leader forms the alliance after receiving feedback from all potential alliance members.

[0043] The system iterates through all bidding requests, determines the number and models of friendly drones that are allied with the enemy target, and sends the alliance formation results message to all potential bidding members. Drones that are not selected will also receive feedback, informing them of their failure to be selected, so as to explore opportunities to form other alliances.

[0044] Furthermore, in step three, the construction process of the drone alliance time-series scheduling model is as follows:

[0045] Step 3.1, task time allocation;

[0046] Using the comprehensive task allocation method in step two, a drone attack coalition of n drones targeting m targets is generated; for the assault coalition targeting the j-th target, the drone swarm needs to dynamically decide on all targets within the fire launch time window [t]. start ,t end The weapon resource launch and timing coordination scheme within the [section]; assuming the involvement of soft-kill resources, the time required to complete the mission is one soft-kill mission cycle T. r Then, according to the soft kill process time [t] start ,t end Divide into N H =(t end -t start ) / T r The task time is divided into N; otherwise, the time of the hard kill attack process is divided into N. H =(t end -t start ) / T hard Each task moment, of which T hard For a hard kill mission cycle;

[0047] Step 3.2, fitting the difference in expected task completion probabilities;

[0048] The expected hit probability of the weapon is calculated using a rapid simulation method:

[0049] First, the expected hit scenarios of heterogeneous weapon resources under various enemy and friendly situations and different enemy behavior constraints were simulated offline using the Monte Carlo method. Then, through a large number of Monte Carlo simulation samples, the results of each attack event were recorded and statistically analyzed.

[0050] Then, based on the attack results, the expected hit probability under uncertain target information is calculated, and the expected hit probability data is stored according to the database structure to form a complete data knowledge base;

[0051] Finally, a real-time online interpolation algorithm is used to accurately fit the sample data of the data knowledge base;

[0052] Step 3.3: Establish a UAV alliance time-series scheduling model based on different cost-effectiveness ratio criteria;

[0053] During a drone attack, the joint hit probability P within a mission cycle is... target for:

[0054]

[0055] Where, x ijt x is the decision variable. ijt x ∈{0,1} ijt=1 indicates that drone i uses weapon resources to attack target j at time t. ijt =0 indicates that UAV i will not use any weapon resources against target j at time t; N A P represents the number of drones participating in the attack on target t; ijt This indicates the probability of a successful hit using weapon resources by the current drone;

[0056] When there are α types of drones and s of each type, j The total number of targets is N target At time t, our UAV i has h types of weapon resources, and the quantity of each type of weapon resource is n. w The total quantity of weapon resources is W; the alliance formation problem is described using a quaternion array, denoted as . in, and These are respectively a drone collection and a weapons resource collection; This is the probability matrix of a successful hit; This indicates the matrix for estimating the remaining flight time of enemy targets;

[0057] Based on the ratio of resources to targets in different scenarios, five different scheduling models are established. The five different scheduling models include the scheduling criterion of maximizing damage probability, the scheduling criterion of early damage, the scheduling criterion of damage probability threshold, the scheduling criterion of maximum cost-effectiveness ratio damage, and the scheduling criterion with maximum damage ratio constraint.

[0058] The scheduling criterion for maximizing the probability of damage:

[0059]

[0060] Wherein, J(x) ikt ) represents the total benefit of UAV timing scheduling; the scheduling criterion of maximizing the damage probability aims to maximize the overall damage probability of all weapon resources. When the ratio of the number of weapon resources to the number of targets is less than 6, the criterion of maximizing the damage probability is selected.

[0061] The scheduling criteria for the pre-damage criterion:

[0062]

[0063] in, Indicates the remaining flight time of the current enemy target;

[0064] The objective function of the early damage criterion is proportional to the combined damage probability and inversely proportional to the sum of the products of the number of fire units and the remaining flight time. When the ratio of damage weapon resources to targets is between 6 and 15, and there are targets with less than 1 minute of remaining flight time, the early damage criterion is adopted.

[0065] The scheduling criterion for the damage probability threshold criterion:

[0066]

[0067] The scheduling criterion based on the damage probability threshold aims to enhance the adaptability of resource allocation by optimizing the distribution and use of weapon resources, based on achieving the damage probability threshold. This aims to achieve a damage weapon resource-to-target ratio of 6–15, with each damage weapon resource possessing damage capability, i.e., p. ijt When all values ​​are higher than 0.75, a threshold-based damage criterion is adopted.

[0068] The scheduling criterion of the maximum cost-effectiveness damage criterion:

[0069]

[0070] Under the scheduling criterion based on the maximum cost-effectiveness damage criterion, it is expected that through scientific and reasonable allocation of firepower resources, when the ratio of damage weapon resources to targets is ≥15, the maximum damage ratio constraint will be selected.

[0071] The scheduling criterion with the maximum damage ratio constraint:

[0072]

[0073] The scheduling criterion with the maximum damage ratio constraint introduces the constraint of the actual number of launches of our weapons resources on the basis of the maximization of damage probability criterion. When there are other situations besides the four scheduling models of the maximization of damage probability criterion, the early damage criterion, the damage probability threshold criterion, and the maximum cost-effectiveness ratio damage criterion, the damage resources are sufficient, and the scheduling criterion with the maximum damage ratio constraint is selected.

[0074] Furthermore, in step four, the process of establishing a drone alliance time-series scheduling method based on the dung beetle algorithm is as follows:

[0075] Step 4.1, Chaos Initialization:

[0076] N dung beetle positions are randomly generated within the search space and chaotic initialization is performed. The initial solution is initialized using the Logistic Tent Chaotic Map (LTCM) algorithm, and the chaotic initialization equation is as follows:

[0077]

[0078] Where x0 is the initial solution randomly generated before chaotic initialization, x0' is the initial solution after chaotic initialization, and r is the chaotic initialization control parameter;

[0079] Depending on the update method, dung beetles include rolling dung beetles, hatching dung beetles, foraging dung beetles, and thieving dung beetles; before the drone alliance's time-series scheduling method is executed, the number of each type of dung beetle is set according to an equal distribution ratio;

[0080] Step 4.2, establish the standard dung beetle optimization algorithm update process:

[0081] Dung beetles update their behavior through activities such as rolling a ball, dancing, reproducing, foraging, and stealing. Each dung beetle operator selects an update method based on its own circumstances.

[0082] The rolling ball behavior:

[0083] In the absence of obstacles, the rolling dung beetle uses sunlight for navigation to keep its dung ball rolling in a straight line. The intensity of light affects the dung beetle's path, and the position update method for the rolling behavior is as follows:

[0084] x s 0 (t+1)=x s 0 (t)+η×ξ×x s 0 (t+1)+b×Δx

[0085] Δx=|x s (t)-X W |

[0086] Where, x s 0 (t) represents the position of the s-th rolling dung beetle in the rolling ball strategy at time t; x s 0 (t+1) represents the position of the dung beetle s in the rolling ball strategy at time t+1; η is the natural coefficient, η=1 indicates no deviation, η=-1 indicates deviation from the original direction; X W This represents the worst-case position globally; Δx simulates the change in light intensity, with a larger Δx indicating weaker illumination; ξ and b are both constants, ξ = 0.5, b = 0.5;

[0087] The dancing behavior:

[0088] When encountering obstacles, the dung beetle uses dancing to reorient itself and find a new route. The location update method for this dancing behavior is as follows:

[0089] x s 1 (t+1)=x s 1 (t)+tan(θ)x s 1 (t)-x s1 (t-1)

[0090] Where, x s 1 (t) represents the position of the s-th dancing dung beetle at time t; x s 1 (t+1) represents the position of the dancing dung beetle s at time t+1; x s 1 (t-1) represents the position of the dancing dung beetle s at time t-1; θ is the dancing coefficient, θ~U(0,π);

[0091] The reproductive behavior:

[0092] The dung beetle rolls its dung ball to a safe place suitable for laying its eggs, thus providing a suitable habitat for its offspring.

[0093] Lb * =max(X * ×(1-R α ),Lb)

[0094] Ub * =min(X) * ×(1+R α ),Ub)

[0095] Among them, Lb * Ub * These represent the lower and upper boundaries of the spawning area, respectively; X * R represents the current local optimum; α This represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively.

[0096] The area where spawning occurs is dynamically adjusted with the number of iterations:

[0097] B a (λ+1)=X * +b1×(B a (λ)-Lb * )+b2×(B a (λ)-Ub * )

[0098] Among them, B a (t) represents the position information of the a-th egg in the λ-th iteration; b1 and b2 are constants, b1 = 0.4, b2 = 0.6;

[0099] The foraging behavior described:

[0100] Dung beetles emerge from the ground in search of food. Their optimal foraging area is dynamically updated.

[0101] Lb b =max(X b ×(1-R),Lb)

[0102] Ub b =min(X) b ×(1+R),Ub)

[0103]

[0104] Among them, X b Lb represents the globally optimal position. b Ub b These represent the lower and upper bounds of the optimal foraging area, respectively; T is the maximum computation time, set manually; the position update method for foraging dung beetles is as follows:

[0105] x s 3 (t+1)=x s 3 (t)+C1×(x s 3 (t)-Lb b )+C2×(x s 3 (t)-Ub b )x s 3 (t) represents the position of the s-th foraging dung beetle at time t; x s 3 (t+1) represents the position of the s-th foraging dung beetle at time t+1;

[0106] The theft:

[0107] The dung beetle steals dung balls, and its position is updated as follows:

[0108] x s 4 (t+1)=X b +S1×g×{|x 4 s (t)-X * |+|x s 4 (t)-X b |}

[0109] Where g is a random vector of size that follows a normal distribution; S1 is a constant; x s 4 (t) represents the position of the s-th dung beetle at time t; x s 4 (t+1) represents the position of the s-th dung beetle at time t+1;

[0110] Step 4.3, Optimize the update mechanism:

[0111] 4.3.1 Dynamically Adjusted Dung Beetle Update Method:

[0112] Introducing a dynamic adjustment function P s,λ This allows the rolling dung beetle to dynamically adjust the probability of rolling ball updates versus dancing updates based on its own fitness value; the dynamic adjustment function P s,λ The definition is as follows:

[0113]

[0114] Among them, P s,λ P represents the dynamic adjustment probability of regional foraging and guided foraging for the s-th dung beetle in the λ-th iteration; min and P max X is a constant, representing the minimum and maximum probabilities preset for the rolling dung beetle update method, respectively; s,λ F(X) represents the position of the s-th dung beetle during the λ-th iteration; s,λ F represents the fitness value of the position of the s-th dung beetle; mid，λ and F max，λ Let F be the minimum and maximum fitness values ​​of all *Dung Beetle* populations at the λ-th iteration; mid，λ The median fitness value of the current dung beetle population; the dynamic adjustment probability P s,λ The range of values ​​for is:

[0115]

[0116] The s-th rolling dung beetle in the λ-th iteration uses P s,λ The probability of rolling ball updates is P. s,λ The probability of dancing is updated;

[0117] 4.3.2 Dung Beetle Update Mechanism Based on Simulated Annealing:

[0118] The Metropolis criterion is introduced to optimize the dung beetle's update strategy: by accepting a new state that is worse than the current state, the diversity of the algorithm's search process is ensured, thus preventing the dung beetle from prematurely converging to a local optimum; the probability p of the dung beetle accepting a worse position is... st as follows:

[0119]

[0120] Where It is the current iteration number; MaxIt is the maximum iteration number; x s 4 (t) represents the position of the s-th dung beetle at time t; xs 4 (t+1) represents the position that the dung beetle will update according to its original stealing behavior; F(·) is the fitness function for calculating the position; e is the base of the natural logarithm.

[0121] The beneficial effects of this invention are as follows:

[0122] This invention achieves optimal collaborative planning of resources across UAVs, time, and space dimensions. Based on the complex coupling relationships between resources, the dynamically changing battlefield situation, and strict spatiotemporal constraints, it deeply integrates the resource capabilities of each UAV to achieve parallel execution of tasks and dynamic allocation of resources, significantly improving the task adaptability, survivability, and strike effectiveness of the cluster. It effectively answers the question of "when, when, what resources, and what tasks each UAV uses to perform" in heterogeneous UAV clusters. Attached Figure Description

[0123] Figure 1 This is an overall flowchart of the present invention;

[0124] Figure 2 This is a task allocation situation diagram for the present invention;

[0125] Figure 3 This is a Gantt chart of the scheduling scheme of the present invention;

[0126] Figure 4 The fitness curve of the drone alliance timing scheduling method for the implementation case. Detailed Implementation

[0127] Example:

[0128] A total of 13 drones coordinated an attack on three targets. The target coordinates are: Target 1: (25.0, 60.0), Target 2: (50.0, 70.0), Target 3: (75.0, 60.0); the drone coordinates are: Drone U1: (14.8, 70.3), Drone U2: (23.8, 49.1), Drone U3: (28.7, 65.4), Drone U4: (62.8, 74.9), Drone U5: ( The task allocation results are shown in the following diagram: U6: (64.6, 71.9), U7: (78.5, 53.8), U8: (77.8, 66.2), U9: (71.6, 69.7), U10: (63.2, 62.0), U11: (25.2, 68.6), U12: (40.3, 51.4), U13: (27.6, 52.0). Figure 2 As shown, U6, U8, and U13 are performing reconnaissance missions, U2, U4, U11, and U10 are performing suppression missions, and the remaining UAVs are performing attack missions.

[0129] Based on the established mission alliance, the scheduling problem of each UAV is planned, and the scheduling result is as follows: Figure 3 As shown, drone U10 executes the task in 10 seconds, drone U11 in 12.2 seconds, drone U4 in 15 seconds, drone U2 in 16.3 seconds, drone U12 in 31.7 seconds, drone U7 in 46.1 seconds, drone U9 in 46.9 seconds, drone U1 in 51.4 seconds, drone U3 in 67.8 seconds, and drone U5 in 94.3 seconds; the fitness curves of the algorithm execution are as follows. Figure 4 As shown, the optimal fitness curve of the algorithm reaches its maximum value after 40 iterations, and the average optimal fitness curve reaches its peak value after 62 iterations, with the final optimal fitness being 9.4669.

Claims

1. A flexible scheduling method for heterogeneous unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: Step 1: Construct a comprehensive UAV task allocation model; Step 2: Construct a comprehensive task allocation method based on the comprehensive task allocation model; Step 3: Construct a drone alliance time-series scheduling model; Step 4: Establish a drone alliance time-series scheduling method based on the dung beetle algorithm.

2. The heterogeneous UAV swarm collaborative flexible scheduling method according to claim 1, characterized in that, In step one, the construction process of the UAV integrated task allocation model is as follows: The integrated task allocation model for unmanned aerial vehicles (UAVs) includes a detection allocation model, a suppression allocation model, and an attack allocation model; the attack allocation model is also known as the encirclement allocation model. Step 1.1: Select the constituent elements of the UAV integrated task allocation model; Positional elements, motion elements, target detection elements, and perception capability elements are selected as variables for the detection allocation model; positional elements, motion elements, target suppression elements, and electronic countermeasures capability elements are selected as variables for the suppression allocation model; positional elements, motion elements, and target attack elements are selected as variables for the attack allocation model. Step 1.2: Construct the detection allocation model, suppression allocation model, and attack allocation model sequentially based on the analytic hierarchy process (AHP). Step 1.2.1, Construct the probe allocation model: The detection advantage of the i-th UAV over the j-th target (dec) ij When the i-th drone does not have detection capability, dec ij The value is 0; the detection advantage is dec. ij The calculation method is as follows: dec ij =D(x i ,x t j ,v i ,v t j ,D j ,d i ) Where, x i x is the position of our i-th UAV in the spatial coordinate system; t j It is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j It is the velocity vector of the j-th enemy target; D j The detection value of the j-th target; d i D is the detection range of our i-th UAV. When the i-th UAV does not have detection capability, the detection range is 0; D(·) is the detection advantage calculation function. The detection advantage matrix formed by m friendly drones and n enemy targets is Det[det] ij ] m×n The allocation matrix is in, Let represent the detection decision variable for the i-th UAV towards the j-th target, indicating whether the i-th UAV should conduct reconnaissance of the j-th target; the detection allocation model for the UAV swarm towards the target swarm is expressed as: The detection allocation model shows that the total detection benefit is maximized and each target is tracked and detected by at least one UAV. Step 1.2.2, Construct the suppression allocation model: The suppressive advantage of the i-th drone over the j-th target (jam) ij When the i-th drone does not have suppression capability, jam ij Take 0, suppress the advantage jam ij The calculation method is as follows: jam ij =J(x i ,x t j ,v i ,v t j ,J j ,j i ) Where, x i x is the position of our i-th UAV in the spatial coordinate system; t j It is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j J is the velocity vector of the j-th enemy target; j It is the suppressive value of the j-th target; j i J is the suppression distance of our i-th drone. When the i-th drone does not have suppression capability, the suppression distance is 0; J(·) is the suppression advantage calculation function. The suppression advantage matrix formed by m friendly drones and n enemy targets is Jam[jam] ij ] m×n The allocation matrix is in, Let represent the suppression decision variable of the i-th drone against the j-th target, indicating whether the i-th drone should suppress the j-th target; the suppression allocation model of the drone swarm against the target swarm is expressed as: The suppression allocation model shows that the total suppression benefit is maximized, and each target is suppressed by at least one drone. Step 1.2.3, Construct the attack allocation model: The attack advantage of the i-th drone against the j-th target is kil ij When the i-th drone does not have attack capabilities, kil ij Take 0, attack advantage kil ij The calculation method is as follows: kilo ij =K(x i ,x t j ,in i ,in t j ,K j ,to i ) Where, x i x is the position of our i-th UAV in the spatial coordinate system; t j It is the position of the j-th enemy target in the spatial coordinate system; v i The velocity vector of our i-th UAV; v t j K is the velocity vector of the j-th enemy target; j It is the attack value of the j-th target; k i It is the attack capability of the i-th drone, k i It is related to the attack resources carried by the i-th drone; K(·) is the attack advantage calculation function; The attack advantage matrix formed by m friendly drones and n enemy targets is Kil[kil ij ] m×n The allocation matrix is in, Let represent the decision variable for the i-th drone against the j-th target, indicating whether the i-th drone attacks the j-th target; the attack allocation model for the drone swarm against the target swarm is expressed as: The attack allocation model shows that the total attack benefit is maximized and each target is attacked by at least one drone.

3. The heterogeneous UAV swarm collaborative flexible scheduling method according to claim 1, characterized in that, In step two, the construction process of the comprehensive task allocation method is as follows: Step 2.1, the alliance proposes: The Hungarian algorithm is used to solve the matching relationship between the detection drone and the target, with the detection drone as the leader of the alliance; Step 2.2, Consortium Bidding: Once a drone with a different payload receives a proposal message, it becomes a potential alliance member if its resources are not exhausted and it does not belong to any existing alliance. Drones carrying heterogeneous resources establish a suppression advantage based on step one. ij With attack advantage kil ij The calculation method analyzes the suppression and attack benefits of the drone against each enemy target unit; according to the alliance order from highest to lowest attack benefit, the resource information and cost information of the drone itself are fed back to the alliance leader through the communication network in the form of alliance formation bidding messages. Step 2.3, Alliance Formation: The alliance leader forms the alliance after receiving feedback from all potential alliance members. The system iterates through all bidding requests, determines the number and models of friendly drones that are allied with the enemy target, and sends the alliance formation results message to all potential bidding members. Drones that are not selected will also receive feedback, informing them of their failure to be selected, so as to explore opportunities to form other alliances.

4. The heterogeneous UAV swarm collaborative flexible scheduling method according to claim 1, characterized in that, In step three, the construction process of the drone alliance time-series scheduling model is as follows: Step 3.1, task time allocation; Using the comprehensive task allocation method in step two, a drone attack coalition of n drones targeting m targets is generated; for the assault coalition targeting the j-th target, the drone swarm needs to dynamically decide on all targets within the fire launch time window [t]. start ,t end [Includes] a weapon resource launch and timing coordination scheme; Assuming the involvement of soft-kill resources, the time required to complete the mission is one soft-kill mission cycle T. r Then, according to the soft kill process time [t] start ,t end Divide into N H =(t end -t start ) / T r The task time is divided into N; otherwise, the hard kill attack process time is divided into N. H =(t end -t start ) / T hard Each task moment, of which T hard For a hard kill mission cycle; Step 3.2, fitting the difference in expected task completion probabilities; The expected hit probability of the weapon is calculated using a rapid simulation method: First, the expected hit scenarios of heterogeneous weapon resources under various enemy and friendly situations and different enemy behavior constraints were simulated offline using the Monte Carlo method. Then, through a large number of Monte Carlo simulation samples, the results of each attack event were recorded and statistically analyzed. Then, based on the attack results, the expected hit probability under uncertain target information is calculated, and the expected hit probability data is stored according to the database structure to form a complete data knowledge base; Finally, a real-time online interpolation algorithm is used to accurately fit the sample data of the data knowledge base; Step 3.3: Establish a UAV alliance time-series scheduling model based on different cost-effectiveness ratio criteria; During a drone attack, the joint hit probability P within a mission cycle is... target for: Where, x ijt x is the decision variable. ijt x ∈{0,1} ijt =1 indicates that drone i uses weapon resources to attack target j at time t. ijt =0 indicates that UAV i will not use any weapon resources against target j at time t; N A P represents the number of drones participating in the attack on target t; ijt This indicates the probability of a successful hit using weapon resources by the current drone; When there are α types of drones, and s are the number of drones of each type. j The total number of targets is N target At time t, our UAV i has h types of weapon resources, and the quantity of each type of weapon resource is n. w The total quantity of weapon resources is W; the alliance formation problem is described using a quaternion array, denoted as W. in, and These are respectively a drone collection and a weapons resource collection; This is the probability matrix of a successful hit; This indicates the matrix for estimating the remaining flight time of enemy targets; Based on the ratio of resources to targets in different scenarios, five different scheduling models are established. The five different scheduling models include the scheduling criterion of maximizing damage probability, the scheduling criterion of early damage, the scheduling criterion of damage probability threshold, the scheduling criterion of maximum cost-effectiveness ratio damage, and the scheduling criterion with maximum damage ratio constraint. The scheduling criterion for maximizing the probability of damage: Wherein, J(x) ikt ) represents the total benefit of UAV timing scheduling; the scheduling criterion of maximizing the damage probability aims to maximize the overall damage probability of all weapon resources. When the ratio of the number of weapon resources to the number of targets is less than 6, the criterion of maximizing the damage probability is selected. The scheduling criteria for the pre-damage criterion: in, Indicates the remaining flight time of the current enemy target; The objective function of the early damage criterion is proportional to the combined damage probability and inversely proportional to the sum of the products of the number of fire units and the remaining flight time. When the ratio of damage weapon resources to targets is between 6 and 15, and there are targets with less than 1 minute of remaining flight time, the early damage criterion is adopted. The scheduling criterion for the damage probability threshold criterion: The scheduling criterion based on the damage probability threshold aims to enhance the adaptability of resource allocation by optimizing the distribution and use of weapon resources, based on achieving the damage probability threshold. This aims to achieve a damage weapon resource-to-target ratio of 6–15, with each damage weapon resource possessing damage capability, i.e., p. ijt When all values ​​are higher than 0.75, a threshold-based damage criterion is adopted. The scheduling criterion of the maximum cost-effectiveness damage criterion: Under the scheduling criterion based on the maximum cost-effectiveness damage criterion, it is expected that through scientific and reasonable allocation of firepower resources, when the ratio of damage weapon resources to targets is ≥15, the maximum damage ratio constraint will be selected. The scheduling criterion with the maximum damage ratio constraint: The scheduling criterion with the maximum damage ratio constraint introduces the constraint of the actual number of launches of our weapons resources on the basis of the maximization of damage probability criterion. When there are other situations besides the four scheduling models of the maximization of damage probability criterion, the early damage criterion, the damage probability threshold criterion, and the maximum cost-effectiveness ratio damage criterion, the damage resources are sufficient, and the scheduling criterion with the maximum damage ratio constraint is selected.

5. The heterogeneous unmanned aerial vehicle (UAV) swarm collaborative flexible scheduling method according to claim 1, characterized in that, In step four, the process of establishing a drone alliance time-series scheduling method based on the dung beetle algorithm is as follows: Step 4.1, Chaos Initialization: N dung beetle positions are randomly generated within the search space and chaotic initialization is performed; the initial solution is obtained using the Logistic Stir-Tent Chaotic Initialization Algorithm, and the chaotic initialization equation is as follows: Where x0 is the initial solution randomly generated before chaotic initialization, x0' is the initial solution after chaotic initialization, and r is the chaotic initialization control parameter; Depending on the update method, dung beetles include rolling dung beetles, hatching dung beetles, foraging dung beetles, and thieving dung beetles; before the drone alliance's time-series scheduling method is executed, the number of each type of dung beetle is set according to an equal distribution ratio; Step 4.2, establish the standard dung beetle optimization algorithm update process: Dung beetles update their behavior through activities such as rolling a ball, dancing, reproducing, foraging, and stealing. Each dung beetle operator selects an update method based on its own circumstances. The rolling ball behavior: In the absence of obstacles, the rolling dung beetle uses sunlight for navigation to keep its dung ball rolling in a straight line. The intensity of light affects the dung beetle's path, and the position update method for the rolling behavior is as follows: x s 0 (t+1)=x s 0 (t)+η×ξ×x s 0 (t+1)+b×Δx Δx=|x s (t)-X W | Where, x s 0 (t) represents the position of the s-th rolling dung beetle in the rolling ball strategy at time t; x s 0 (t+1) represents the position of the dung beetle s in the rolling ball strategy at time t+1; η is the natural coefficient, η=1 indicates no deviation, η=-1 indicates deviation from the original direction; X W This represents the worst-case position globally; Δx simulates the change in light intensity, with a larger Δx indicating weaker illumination; ξ and b are both constants, ξ = 0.5, b = 0.5; The dancing behavior: When encountering obstacles, the dung beetle uses dancing to reorient itself and find a new route. The location update method for this dancing behavior is as follows: x s 1 (t+1)=x s 1 (t)+tan(θ)|x s 1 (t)-x s 1 (t-1)| Where, x s 1 (t) represents the position of the s-th dancing dung beetle at time t; x s 1 (t+1) represents the position of the dancing dung beetle s at time t+1; x s 1 (t-1) represents the position of the dancing dung beetle s at time t-1; θ is the dancing coefficient, θ~U(0,π); The reproductive behavior: The dung beetle rolls its dung ball to a safe place suitable for laying its eggs, thus providing a suitable habitat for its offspring. Lb * =max(X * ×(1-R α ),Lb) Ub * =min(X * ×(1+R α ),Ub) Among them, Lb * Ub * These represent the lower and upper boundaries of the spawning area, respectively; X * R represents the current local optimum; α This represents the maximum number of iterations; Lb and Ub represent the lower and upper bounds of the optimization problem, respectively. The area where spawning occurs is dynamically adjusted with the number of iterations: B a (λ+1)=X * +b1×(B a (λ)-Lb * )+b2×(B a (λ)-Ub * ) Among them, B a (t) represents the position information of the a-th egg in the λ-th iteration; b1 and b2 are both constants, b1 = 0.4, b2 = 0.6; The foraging behavior described: Dung beetles emerge from the ground in search of food. Their optimal foraging area is dynamically updated. Lb b =max(X b ×(1-R),Lb) Ub b =min(X b ×(1+R),Ub) Among them, X b Lb represents the globally optimal position. b Ub b These represent the lower and upper bounds of the optimal foraging area, respectively; T is the maximum computation time, set manually; the position update method for foraging dung beetles is as follows: x s 3 (t+1)=x s 3 (t)+C1×(x s 3 (t)-Lb b )+C2×(x s 3 (t)-Ub b ) x s 3 (t) represents the position of the s-th foraging dung beetle at time t; x s 3 (t+1) represents the position of the s-th foraging dung beetle at time t+1; The theft: The dung beetle steals dung balls, and its position is updated as follows: x s 4 (t+1)=X b +S1×g×{|x 4 s (t)-X * |+|x s 4 (t)-X b |} Where g is a random vector of size that follows a normal distribution; S1 is a constant; x s 4 (t) represents the position of the s-th dung beetle at time t; x s 4 (t+1) represents the position of the s-th dung beetle at time t+1; Step 4.3, Optimize the update mechanism: 4.3.1 Dynamically Adjusted Dung Beetle Update Method: Introducing a dynamic adjustment function P s,λ This allows the rolling dung beetle to dynamically adjust the probability of rolling ball updates versus dancing updates based on its own fitness value; the dynamic adjustment function P s,λ The definition is as follows: Among them, P s,λ P represents the dynamic adjustment probability of regional foraging and guided foraging for the s-th dung beetle in the λ-th iteration; min and P max X is a constant, representing the minimum and maximum probabilities preset for the rolling dung beetle update method, respectively; s,λ F(X) represents the position of the s-th dung beetle during the λ-th iteration; s,λ F represents the fitness value of the position of the s-th dung beetle; mid,λ and F max,λ Let F be the minimum and maximum fitness values ​​of all *Dung Beetle* populations at the λ-th iteration; mid,λ The median fitness value of the current dung beetle population; the dynamic adjustment probability P s,λ The range of values ​​for is: The s-th rolling dung beetle in the λ-th iteration uses P s,λ The probability of rolling ball updates is P. s,λ The probability of dancing is updated; 4.3.2 Dung Beetle Update Mechanism Based on Simulated Annealing: The Metropolis criterion is introduced to optimize the dung beetle's update strategy: by accepting a new state that is worse than the current state, the diversity of the algorithm's search process is ensured, thus preventing the dung beetle from prematurely converging to a local optimum; the probability p of the dung beetle accepting a worse position is... st as follows: Where It is the current iteration number; MaxIt is the maximum iteration number; x s 4 (t) represents the position of the s-th dung beetle at time t; x s 4 (t+1) represents the position that the dung beetle will update according to its original stealing behavior; F(·) is the fitness function for calculating the position; e is the base of the natural logarithm.

6. An electronic device, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 5.

Images