Multi-unmanned aerial vehicle cooperative task allocation method with self-evolution search capability

Abstract

The invention discloses a multi-unmanned aerial vehicle cooperative task allocation method with a self-evolution search capability, belongs to the technical field of emergency rescue, and aims to solve the problems that the multi-unmanned aerial vehicle cooperative task allocation method is easy to fall into local optimum and is weak in constraint processing capability. Comprising the steps that task modeling and parameter initialization are carried out, and an emergency rescue task is converted into a computable mathematical model; decoding the particles and executing deadlock detection and repair, and converting continuous solution vectors in particle swarm optimization into executable task allocation schemes meeting all rescue scene constraints; calculating a fitness value and a self-adaptive inertia weight of each particle to realize personalized search for different particles; convergence stagnation is monitored, intelligent variable neighborhood search and disturbance are triggered, and double-adaptive capacity of depth and neighborhood selection is achieved; updating the particles by using a self-adaptive inertia weight; and repeating the steps until the termination condition is met. According to the method, the solving quality and stability of a multi-unmanned aerial vehicle cooperative task allocation scheme under complex constraints can be remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of emergency rescue technology, and relates to UAV mission planning and scheduling, and particularly to a dynamic intelligent task allocation method for multi-UAV collaborative operations, specifically a multi-UAV collaborative task allocation method with self-evolving search capability. Background Technology

[0002] Multi-UAV collaborative task allocation is a key technological issue in modern emergency rescue. Its goal is to plan the optimal task execution sequence for multiple UAVs operating collaboratively, maximizing overall efficiency while meeting various complex constraints and dynamic task change scenarios. The core of this research is an NP-hard combinatorial optimization problem, and developing intelligent and efficient task allocation methods is crucial for solving such problems.

[0003] Particle Swarm Optimization (PSO) algorithms are widely used to solve multi-UAV cooperative task allocation problems due to their ease of implementation and fast convergence. However, multi-UAV cooperative task allocation is a discrete and strongly constrained optimization problem, and applying PSO algorithms to solve it faces many inherent drawbacks that are difficult to overcome, such as: difficulty in ensuring feasibility, susceptibility to getting trapped in local optima, and rigidity of the search strategy.

[0004] Patent CN202210635458.9 explored ways to overcome the aforementioned shortcomings, with its core idea being to mix PSO with other local search algorithms. However, it lacks real-time feedback and learning capabilities for the search process. Patent CN202410708609.8 adopted an approach to improve the internal mechanism of the PSO algorithm, implementing a "group-based" inertial weight adjustment strategy to enhance the solution quality and robustness of the PSO algorithm. However, it ignores the differences among "individuals" within the "population" and heavily relies on preset adjustment rules, leading to premature convergence. In summary, existing technical solutions have not effectively solved the adaptability problem of local search, nor have they achieved truly differentiated search based on the individual particle states, thus limiting the improvement of algorithm performance. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-UAV collaborative task allocation method with self-evolving search capabilities, solving the problems of weak local search adaptability and difficulty in achieving differentiated search in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A multi-UAV cooperative task allocation method with self-evolving search capability includes the following steps:

[0008] Step S1: Task Modeling and Algorithm Parameter Initialization. This step aims to transform the rescue mission requirements into a computable mathematical model. First, a mathematical model is constructed for the multi-UAV collaborative task allocation problem, including decision variables, objective function, and constraints.

[0009] Decision variables: using binary variables This indicates a multi-drone collaborative task allocation decision. If the task... Assigned to drones ,but Otherwise, it is 0.

[0010] Objective function (fitness function): Minimize the overall cost function Its formula is:

[0011]

[0012] in, It is the total number of drones participating in the mission. For the first The total flight distance cost for a drone to perform all tasks within its assigned sequence. For the first The time deviation cost of a drone performing its mission. For the first The total rescue benefits obtained by using drones to complete missions. The weighting coefficients are used to weight different physical quantities through dimensional conversion. In specific implementations, they are set to fixed values ​​to balance the three indicators.

[0013] Constraints include single task allocation constraints, multi-task collaboration constraints, and UAV capability constraints. It is a collection of single rescue missions. It is a collection of multiple rescue missions. It is the total number of tasks. It is a drone Maximum task load limit A collection of drones.

[0014] The single task assignment constraint means that each single rescue task must be assigned to one and only one drone.

[0015]

[0016] Multi-task coordination constraints represent each multi-rescue mission Must be The drones are used in coordinated operations.

[0017]

[0018] The drone capability constraint means that the number of tasks each drone can perform does not exceed its maximum payload limit.

[0019]

[0020] Subsequently, the parameters of the hybrid particle swarm optimization algorithm are initialized, including: PSO parameters (population size, maximum number of iterations, learning factor), adaptive weight parameters (inertia weight upper and lower bounds), A-VNS parameters (neighborhood structure set, search depth upper and lower bounds, historical success and call counters), perturbation parameters (basic perturbation probability and scale, enhancement coefficient), and convergence monitoring parameters (stagnation threshold).

[0021] Step S2: Particle Decoding and Feasibility Assurance. This step is responsible for transforming the continuous solution vectors of the PSO into an executable task allocation scheme that satisfies all rescue constraints. First, the continuous position vectors of each particle are decoded into an initial task allocation scheme containing task allocation decisions and task execution order using decoding rules. After decoding, the initial task allocation scheme is subjected to constraint correction, deadlock detection, and deadlock repair to obtain a feasible task allocation scheme. Constraint correction first corrects the task allocation constraints and UAV capability constraints. Deadlock detection and repair are crucial to ensuring the executability of the scheme, and the specific process is as follows:

[0022] Constructing a directed graph :vertex For all multi-rescue missions, there is a directed edge collection This represents the execution order relationship between tasks, if there are directed edges. This indicates that there is a drone that needs to perform a task first. Execute the task again .

[0023] Loop detection: A depth-first search algorithm is used to detect the loop. The loop in.

[0024] To fix a deadlock: If a cycle is detected, select a directed edge from the cycle. Find all that are executed first. Execute again The drones exchanged in their mission sequence and The location is determined to eliminate the loop. This process is repeated until the diagram is drawn. There are no loops in the solution, thus yielding a feasible solution without deadlock.

[0025] Step S3: Fitness Evaluation and Inertia Weight Adaptation. For the feasible solution processed in Step S2, calculate its comprehensive cost function according to the mathematical model defined in Step S1. The value, and use this value as the particle. fitness value . The smaller the value, the better the task allocation scheme. Subsequently, an adaptive inertia weight is calculated independently for each particle to achieve personalized search. Preferably, the adaptive inertia weight calculation formula is:

[0026]

[0027] in, For particles Adaptive inertia weights, and These are the preset upper and lower limits for the inertia weight, respectively. For particles The current fitness value, and These are the best and worst fitness values ​​of the population in the current iteration, respectively. This mechanism allows poorly performing particles to receive a larger inertia weight for global exploration, while well-performing particles receive a smaller inertia weight for local refinement.

[0028] Step S4: Intelligent Adaptive Variable Neighborhood Search (A-VNS) and Perturbation. This occurs when the fitness value of the global optimum continuously reaches a preset stagnation threshold a certain number of times. When the algorithm is determined to be trapped in a local optimum, the intelligent adaptive variable neighborhood search and perturbation module is triggered. This module first performs intelligent variable neighborhood search, and then performs population perturbation, as follows:

[0029] S4-1: Intelligent Variable Neighborhood Search (A-VNS). This part is one of the core innovations of this invention, and its "intelligence" is reflected in the dual adaptive mechanism.

[0030] S4-1-1: Adaptive Adjustment of Search Depth: Its purpose is to dynamically allocate computational resources based on the convergence state of the population. First, calculate the current population diversity index. For example, all particles and the globally optimal particle. The average Euclidean distance is calculated and normalized.

[0031]

[0032] in, Indicates population size, Represents the dimension of the particle. Indicates the first The particle in the first Dimensional component values, Represents the globally optimal particle In the Dimensional component values, The maximum diagonal length of the search space (normalization factor) is used to... The value is constrained to be within the interval [0, 1].

[0033] Then, the depth of this variable neighborhood search is dynamically adjusted according to the following optimal formula. :

[0034]

[0035] in, This represents the minimum number of neighborhood searches allowed when the algorithm gets stuck in a local optimum. This represents the maximum number of iterations allowed for neighborhood search when the algorithm gets stuck in a local optimum. This mechanism enables automatic depth mining when population diversity is low, meaning the algorithm may get stuck in a local optimum. When the population diversity is high, a shallow search is performed to save computational resources.

[0036] S4-1-2: Adaptive Selection of Neighborhood Structure Based on Historical Performance: The aim is to allow the algorithm to learn from historical experience and prioritize more effective search strategies. First, calculate the performance of each neighborhood structure... Historical success rate of operations such as swapping, insertion, and reversal P hist j = SuccessCount [j] / CallCount [j] ,in SuccessCount [j] It is the first A neighborhood structure optimization success counter records the number of times the operation successfully finds a better solution; CallCount [j] It is the first A total call counter for each neighborhood structure is used to record the total number of times the operation is selected by the algorithm. Then, a roulette wheel selection algorithm is employed to prioritize neighborhood structures with high historical success rates. This serves as the starting operation for this variable neighborhood search. Preferably, to ensure fair exploration, if this is the first time A-VNS is triggered, a neighborhood structure is randomly selected.

[0037] S4-1-3: VNS Search and Learning: Using the Current Globally Optimal Solution For the initial solution, the neighborhood structure selected in S4-1-2 is used. and the search depth determined in S4-1-1 Perform a local search. If a better solution is found during the search... Then, it is used to update the global optimal solution, reset the stall counter, and update the call counter and success counter of the corresponding neighborhood structure, thus forming a complete "experience learning" closed loop.

[0038] S4-2: Serialization Perturbation Guarantee. After the A-VNS search, regardless of whether a better solution is found, a global perturbation is immediately performed. This mechanism aims to provide the population with additional stochastic search momentum to avoid the algorithm getting stuck in small depressions near the optimal solution. For each particle in the population, a dynamically enhanced perturbation probability is applied. Apply a random Gaussian perturbation vector This pushes its position into a new region of the search space. The formula for calculating the perturbation probability is defined as:

[0039]

[0040] in, This refers to the actual probability value of a particle performing a position perturbation operation in the current iteration. This is an initial constant preset by the algorithm, used to maintain the basic exploration ability of the population. In this embodiment, the value is set to 0.1. This is the perturbation enhancement factor, used to linearly amplify the perturbation probability when the A-VNS mechanism is triggered, thereby strengthening the algorithm's ability to escape local optima. In this embodiment, this enhancement factor is set to 2.0. The 1.0 in the formula represents... The maximum upper limit of the value ensures that the probability value after enhanced calculation is always within the range of [the maximum possible value]. [0,1] Within the effective closed interval, the logical rigor of the mathematical model is guaranteed.

[0041] Step S5: Particle Update. Use the adaptive inertia weights calculated in Step S3. The velocity and position of each particle are updated according to the standard PSO formula.

[0042]

[0043] in, and They represent the first The particle in the first The second iteration and the first The velocity vector at the next iteration and They represent the first The particle in the first The second iteration and the first The position vector at the next iteration. It is the first The adaptive inertial weight of each particle is used to balance global exploration and local exploitation capabilities; , It is the learning factor, usually set to a fixed positive number, used to adjust the weight of particles learning towards individual and group optimality; It is a random number in the range [0,1], which increases the randomness of the exploration process. Indicates the first The optimal position of a historical individual found by each particle. This represents the globally optimal position found by the entire particle swarm. Boundary adjustments are then made to the updated velocity and position.

[0044] Step S6: Loop and Termination. Repeat steps S2 to S5 until the maximum number of iterations is reached. Finally, output the task allocation scheme corresponding to the globally optimal position.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) The solution is highly feasible and reliable. By introducing a deadlock detection and repair algorithm based on directed graphs, the deadlock problem in multi-machine collaborative tasks is completely solved, ensuring that each generated task allocation scheme is executable.

[0047] (2) Strong global optimization capability and solution quality. Through the collaborative mechanism of adaptive inertial weights, intelligent A-VNS (deep adaptation + neighborhood selection adaptation) and serial perturbation, the algorithm has a strong ability to escape local optima and can stably find a better global combat plan.

[0048] (3) The algorithm has high intelligence and adaptability. The algorithm has decision-making ability similar to "experience", can autonomously adapt to task scenarios of different scales and complexities, and has strong robustness. Attached Figure Description

[0049] Figure 1 This is a flowchart of the multi-UAV collaborative task allocation method with self-evolving search capability of the present invention.

[0050] Figure 2 This is a flowchart for deadlock detection and repair in multiple rescue missions.

[0051] Figure 3 This is a flowchart of the workflow of the Intelligent Variable Neighborhood Search (A-VNS) and Disturbance Coordination Module.

[0052] Figure 4 This is a schematic diagram of a multi-drone collaborative multi-target rescue mission allocation scheme. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and a specific embodiment.

[0054] Example: Planning of multi-drone collaborative rescue missions, the specific process is as follows: Figure 1 As shown.

[0055] 1. Scene settings and parameter configuration

[0056] This embodiment assumes: 5 heterogeneous drones 20 ground targets To carry out coordinated rescue missions. Among them, the target... For high-value targets, at least two drones are needed to complete a coordinated rescue mission. The remaining 13 are single rescue missions. The maximum payload for each drone... Q max =[7 7 6 5 6] The drones took off from different locations, resulting in differences in their speed and rescue capabilities.

[0057] The algorithm parameters in this embodiment are set as follows: the PSO parameter is set to the population size. Maximum number of iterations Learning factor The adaptive weight parameters are set to... 0.4, 0.9; A-VNS parameters are set to neighborhood structure set. Minimum search depth 5. Maximum search depth 25; The perturbation parameter is set as the basic perturbation probability. 0.1, Basic disturbance scale Disturbance enhancement coefficient The convergence monitoring parameter is set to the stagnation threshold. .

[0058] 2. Detailed description of the algorithm execution process

[0059] The algorithm in this embodiment is executed according to steps S1 to S6 as described in the invention.

[0060] In step S1 (initialization), the objective function is established based on the scenario. The mathematical model was constructed, and the positions and velocities of 50 particles were randomly initialized. According to the encoding scheme, the position of each particle is a dimensionless matrix. The vector.

[0061] In step S2 (decoding and feasibility assurance), for each particle, its 27-dimensional position vector is decoded into a preliminary task allocation scheme. Subsequently, the algorithm automatically performs constraint corrections, such as correcting task allocations that exceed the drone's capability limits. Then, referring to... Figure 2Deadlock detection and repair are performed by constructing a task dependency directed graph and eliminating loops within it, resulting in a completely feasible deadlock-free task allocation scheme.

[0062] After step S3 (fitness evaluation), the algorithm enters the main loop. In the early stages of iteration, the algorithm uses an adaptive inertia weight mechanism to allow high-performing particles to focus on local development, while lower-performing particles actively explore new regions.

[0063] When the algorithm iterates to approximately generation 120, because the global optimal solution has not improved for 15 consecutive generations, step S4 (intelligent adaptive variable neighborhood search and perturbation, module details are as follows) is triggered. Figure 3 (As shown).

[0064] In the Intelligent Variable Neighborhood Search (A-VNS) part of step S4, the algorithm first performs depth adaptive adjustment. The system calculates the current population diversity. The value has been reduced to a lower level, such as 0.15, and a larger VNS search depth is dynamically calculated based on this. twenty two.

[0065] Next, the algorithm performs adaptive neighborhood structure selection. Based on historical records, the "insertion" operation is selected first because it has the highest historical success rate, and is used as the starting neighborhood structure for this search.

[0066] Subsequently, starting from the current optimal solution, the algorithm performs a VNS search of depth 22 using the "insertion" neighborhood and successfully finds a better candidate solution, improving the fitness value by approximately 3%. The system immediately updates the global optimal solution, resets the stall counter, and updates the success and call counts of the "insertion" operation.

[0067] After A-VNS is completed, the algorithm continues to perform serialization perturbation, applying a global perturbation to all particles in the population to further stimulate population vitality.

[0068] After step S5 (particle update), the algorithm continues to iterate until it converges after reaching the maximum number of iterations (1500) in step S6 (loop and termination). The final output task allocation scheme clearly provides the rescue mission sequence for each drone and ensures that high-value targets are all rescued collaboratively. See [link to relevant documentation]. Figure 4 .

[0069] 3. Comparison of Implementation Results

[0070] To quantitatively verify the effectiveness of this invention, under the same scenario, the algorithm of this invention (HPSO-AVNS) was compared with the standard particle swarm optimization (PSO) algorithm and a hybrid particle swarm optimization algorithm without adaptive capability (PSO-VNS). The results of 10 independent runs are shown in Table 1 below:

[0071] Table 1 Algorithm Performance Comparison

[0072]

[0073] Note: (1) Calculation method for percentage improvement: (Comparison algorithm value - Algorithm value of this invention) / Comparison algorithm value × 100%;

[0074] (2) Success rate refers to the percentage of times a better solution than PSO-VNS is found in 10 runs.

[0075] As shown in the table above, the algorithm of this invention (HPSO-AVNS) is significantly better than the comparative algorithm in terms of solution quality (i.e., average final fitness value) and stability (i.e., success rate).

[0076] The comparison data in this embodiment and Table 1 fully demonstrate that the technical solution (HPSO-AVNS) provided by the present invention can be effectively applied to complex emergency rescue scenarios. It is significantly superior to the existing technologies (PSO and PSO-VNS) in terms of solution quality and stability, providing reliable technical support for emergency rescue decision-making.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-UAV cooperative task allocation method with self-evolving search capability, characterized in that, Includes the following steps: Step S1: Model the multi-UAV cooperative task allocation problem and initialize the parameters of the hybrid particle swarm optimization algorithm; Step S2: Decode the continuous position vector of each particle in the particle swarm into an initial task allocation scheme, and perform constraint correction and deadlock detection and repair on the initial task allocation scheme to obtain a feasible task allocation scheme. Step S3: Calculate the fitness value of each of the feasible task allocation schemes, and calculate the independent adaptive inertia weight for each particle based on the fitness value; Step S4: Monitor the fitness value of the global optimal solution. If the number of consecutive times it does not improve reaches a preset stagnation threshold, the intelligent variable neighborhood search and perturbation module is triggered. The module first performs intelligent variable neighborhood search and then performs population perturbation. Step S5: Update the velocity and position of all particles using the adaptive inertial weights; Step S6: Repeat steps S2 to S5 until the termination condition is met, and output the task allocation scheme corresponding to the global optimal solution.

2. The multi-UAV cooperative task allocation method with self-evolving search capability according to claim 1, characterized in that, Step S1 is as follows: First, a mathematical model is constructed for the multi-UAV collaborative task allocation problem, including: Decision variables: using binary variables This indicates a multi-drone collaborative task allocation decision; if the task... Assigned to drones ,but Otherwise, it is 0; Objective function: Minimize the overall cost function Its formula is: ； in, It is the total number of drones participating in the mission. For the first The total flight distance cost for a drone to perform all tasks within its assigned sequence. For the first The time deviation cost of a drone performing its mission. For the first The total rescue benefits obtained by using drones to complete missions. These are weighting coefficients, which allow different physical quantities to be weighted through dimensional transformation. In specific implementations, they are set to fixed values ​​to balance the three indicators. Constraints include single-task allocation constraints, multi-task coordination constraints, and UAV capability constraints. It is a collection of single rescue missions. It is a collection of multiple rescue missions. It is the total number of tasks. It is a drone Maximum task load limit For drones; The single task assignment constraint means that each single rescue task must be assigned to one and only one drone; ； Multi-task coordination constraints represent each multi-rescue mission Must be Coordinated execution using drones; ； The drone capability constraint means that the number of missions each drone can perform does not exceed its maximum payload limit; ； Subsequently, the parameters of the hybrid particle swarm optimization algorithm are initialized, including: PSO parameters: population size, maximum number of iterations, learning factor; adaptive weight parameters: upper and lower bounds of the weights; A-VNS parameters: neighborhood structure set, upper and lower bounds of the search depth, historical success and call counters; perturbation parameters: basic perturbation probability and scale, enhancement coefficient; convergence monitoring parameters: stagnation threshold.

3. The multi-UAV cooperative task allocation method with self-evolving search capability according to claim 1, characterized in that, Step S2 is as follows: First, the continuous position vector of each particle is decoded into an initial task allocation scheme containing task allocation decisions and task execution order through decoding rules; After decoding, the initial task allocation scheme is subjected to constraint correction, deadlock detection and deadlock repair to obtain a feasible task allocation scheme; the constraint correction first corrects the task allocation constraints and the UAV capability constraints.

4. A multi-UAV cooperative task allocation method with self-evolving search capability according to claim 3, characterized in that, The specific process for deadlock detection and repair is as follows: Constructing a directed graph :vertex For all multi-rescue missions, there is a directed edge collection This represents the execution order relationship between tasks, if there are directed edges. This indicates that there is a drone that needs to perform a task first. Execute the task again ; Loop detection: A depth-first search algorithm is used to detect the loop. The loop in; To fix a deadlock: If a cycle is detected, select a directed edge from the cycle. Find all that are executed first. Execute again The drones exchanged in their mission sequence and The location is determined to eliminate the loop; this process is repeated until the diagram is complete. There are no loops in the solution, thus yielding a feasible solution without deadlock.

5. A multi-UAV cooperative task allocation method with self-evolving search capability according to claim 1, characterized in that, Step S3 is as follows: For the feasible solution processed in step S2, calculate its comprehensive cost function according to the mathematical model defined in step S1. The value, and use this value as the particle. fitness value Subsequently, an adaptive inertia weight is independently calculated for each particle to achieve personalized search; the formula for calculating the adaptive inertia weight is: ； in, For particles Adaptive inertia weights, and These are the preset upper and lower limits for the inertia weight, respectively. For particles The current fitness value, and These are the best and worst fitness values ​​of the population in the current iteration, respectively.

6. A multi-UAV cooperative task allocation method with self-evolving search capability according to claim 1, characterized in that, Step S4 is as follows: S4-1: Intelligent Variable Neighborhood Search (A-VNS); S4-1-1: Adaptive Adjustment of Search Depth: First, calculate the current population diversity index. ; Then, the depth of this variable neighborhood search is dynamically adjusted according to the following optimal formula. : ； in, This represents the minimum number of neighborhood searches allowed when the algorithm gets stuck in a local optimum. This represents the maximum number of iterations allowed to perform a neighborhood search when the algorithm gets stuck in a local optimum. S4-1-2: Adaptive selection of neighborhood structure based on historical performance: First, calculate the historical performance of each neighborhood structure. Historical success rate P hist j = SuccessCount [j] / CallCount [j] ,in SuccessCount [j] It is the first A neighborhood structure optimization success counter records the number of times the operation successfully finds a better solution; CallCount [j] It is the first A total call counter for each neighborhood structure is used to record the total number of times the operation is selected by the algorithm; subsequently, a roulette wheel selection method is used to prioritize neighborhood structures with high historical success rates. As the starting operation for this variable neighborhood search; preferably, to ensure fair exploration, if it is the first time A-VNS is triggered, a neighborhood structure is randomly selected; S4-1-3: VNS Search and Learning: Using the Current Globally Optimal Solution For the initial solution, the neighborhood structure selected in S4-1-2 is used. and the search depth determined in S4-1-1 Perform a local search; if a better solution is found during the search... Then, it is used to update the global optimal solution, reset the stall counter, and update the call counter and success counter of the corresponding neighborhood structure, thus forming a complete "experience learning" closed loop; S4-2: Serialization perturbation guarantee; after the A-VNS search, regardless of whether a better solution is found, a global perturbation is immediately performed; for each particle in the population, a dynamically enhanced perturbation probability is applied. Apply a random Gaussian perturbation vector This pushes its position into a new region of the search space as a supplementary safeguard mechanism to prevent the algorithm from getting stuck in small depressions near the optimal solution; among which, This refers to the actual probability value of a particle performing a position perturbation operation in the current iteration; These are preset initial constants; This is the perturbation amplification coefficient, used to linearly amplify the perturbation probability when the A-VNS mechanism is triggered; 1.0 in the formula indicates... The maximum upper limit of the value ensures that the probability value after enhanced calculation is always within the range of [the maximum possible value]. [0,1] Within the effective closed interval.

7. A multi-UAV cooperative task allocation method with self-evolving search capability according to claim 6, characterized in that, In step S4-1-1, The calculation uses all particles and the globally optimal particle. The average Euclidean distance is then normalized. ； in, Indicates population size, Represents the dimension of the particle. Indicates the first The particle in the first Dimensional component values, Represents the globally optimal particle In the Dimensional component values, Indicates the maximum diagonal length of the search space, used to... The value is constrained to be within the interval [0, 1].

8. A multi-UAV cooperative task allocation method with self-evolving search capability according to claim 1, characterized in that, Step S5 is as follows: Use the adaptive inertia weights calculated in step S3 The velocity and position of each particle are updated according to the PSO formula; ； in, and They represent the first The particle in the first The second iteration and the first The velocity vector at the next iteration and They represent the first The particle in the first The second iteration and the first The position vector at the next iteration. It is the first The adaptive inertial weight of each particle is used to balance global exploration and local exploitation capabilities; , It is the learning factor, usually set to a fixed positive number, used to adjust the weight of particles learning towards individual and group optimality; It is a random number in the range [0,1], which increases the randomness of the exploration process. Indicates the first The optimal position of a historical individual found by each particle. This represents the globally optimal position found by the entire particle swarm; and boundary processing is performed on the updated velocity and position.

Images