Multi-uav cooperative task allocation method and device, equipment and medium

Abstract

The application relates to a multi-unmanned aerial vehicle cooperative task allocation method and device, equipment and medium, wherein the method comprises the following steps: jointly encoding a mapping relationship between a task and an unmanned aerial vehicle and a task execution sequence on each unmanned aerial vehicle to generate an initial population; performing plume diffusion, chemotactic climbing and cluster elite attraction processing on a current population of the initial population in sequence to generate a clustered optimized population; performing local disturbance fine tuning according to a current optimal solution in the clustered optimized population to obtain a fine-tuned optimal solution, performing weighted fusion according to an elite solution to obtain a fused new solution, and dynamically adjusting a search radius; calculating the fitness of each individual in the current population in the current iteration, updating a global optimal solution according to the fitness, and judging whether an iteration termination condition is reached; if not, returning to perform a global exploration operation again; and if yes, generating a target optimal task allocation scheme and a flight path of each unmanned aerial vehicle. The application improves the allocation efficiency and accuracy of multi-unmanned aerial vehicle cooperative tasks.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) collaborative control technology, and in particular to a method, apparatus, device and medium for multi-UAV collaborative task allocation. Background Technology

[0002] With the rapid growth of demand for urban instant delivery, e-commerce instant delivery, and urgent medical supplies, multi-rotor or vertical take-off and landing fixed-wing drones are regarded as an important technological path to solve the "last mile" delivery problem in cities due to their advantages such as high mobility, ability to avoid congestion, and direct delivery to the last mile.

[0003] Existing delivery scheduling and trajectory planning technologies have significant limitations: centralized integer programming methods, such as mixed-integer linear programming, experience exponentially increasing computational complexity as the task scale expands, making it difficult to meet real-time decision-making requirements; traditional metaheuristic algorithms, such as genetic algorithms and particle swarm optimization, while computationally efficient, generally suffer from premature convergence and insufficient population diversity, making it difficult to obtain high-quality solutions under complex constraints. More critically, existing methods lack effective dynamic event response mechanisms. When encountering unexpected situations such as temporary no-fly zones, sudden order insertions, or equipment failures, they cannot quickly complete trajectory replanning, severely impacting the reliability and safety of task execution. The unique characteristics of urban delivery scenarios further exacerbate these technical challenges: multiple objectives such as delivery time, energy consumption, and safety risks need to be optimized simultaneously; multiple constraints such as time windows, load limits, battery life, and noise control must be met; and the system must have the ability to respond to dynamic environmental changes in real time. These technical bottlenecks severely restrict the large-scale application of drones in urban logistics. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, device, and medium for allocating collaborative tasks among multiple unmanned aerial vehicles (UAVs) to improve the efficiency and accuracy of such allocation.

[0005] To address the aforementioned technical problems, embodiments of this application provide a method for multi-UAV collaborative task allocation, including:

[0006] The mapping relationship between tasks and drones and the task execution order on each drone are jointly encoded to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector;

[0007] The current population of the initial population is subjected to plume diffusion, chemotactic climbing and clustering elite attraction processes in sequence to perform global exploration operations on the current population and generate a clustered optimized population.

[0008] Local perturbation fine-tuning is performed on the current best solution in the clustered population after optimization to obtain the fine-tuned best solution. Weighted fusion is performed on the input elite solutions to obtain the fused new solution. The search radius is dynamically adjusted according to the current iteration number and the preset iteration number.

[0009] Calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution based on the fitness, and determine whether the iteration termination condition has been met. If not, return and re-execute the global exploration operation.

[0010] If the iteration termination condition is met, the target optimal task allocation scheme and the flight paths of each UAV are generated.

[0011] To address the aforementioned technical problems, embodiments of this application provide a multi-UAV collaborative task allocation device, comprising:

[0012] The population initialization module is used to jointly encode the mapping relationship between tasks and UAVs and the task execution order on each UAV to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector;

[0013] The global exploration module is used to sequentially perform plume diffusion, chemotactic climbing, and clustering elite attraction processes on the current population of the initial population in order to perform global exploration operations on the current population and generate a clustered optimized population.

[0014] The local optimization module is used to perform local perturbation fine-tuning based on the current best solution in the clustered population to obtain the fine-tuned best solution, and to perform weighted fusion based on the input elite solutions to obtain a new fused solution, and to dynamically adjust the search radius based on the current iteration number and the preset iteration number.

[0015] The optimal solution update module is used to calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution according to the fitness, and determine whether the iteration termination condition has been met. If not, it returns to re-execute the global exploration operation.

[0016] The allocation scheme generation module is used to generate the target optimal task allocation scheme and the flight paths of each UAV when the iteration termination condition is met.

[0017] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is to provide an electronic device, including one or more processors; and a memory for storing one or more programs, so that the one or more processors implement the multi-UAV collaborative task allocation method described in any one of the above-mentioned methods.

[0018] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the multi-UAV cooperative task allocation method described above.

[0019] This invention provides a method, apparatus, device, and medium for multi-UAV collaborative task allocation. The method includes: jointly encoding the mapping relationship between tasks and UAVs and the task execution order on each UAV to generate an initial population containing N individuals, where each individual in the initial population is a candidate solution vector; sequentially performing plume diffusion, chemotactic climbing, and elite clustering attraction processing on the current population of the initial population to perform a global exploration operation, generating a clustered optimized population; performing local perturbation fine-tuning based on the current optimal solution in the clustered optimized population to obtain a fine-tuned optimal solution, and performing weighted fusion based on the input elite solutions to obtain a fused new solution, and dynamically adjusting the search radius based on the current iteration number and a preset iteration number; calculating the fitness of each individual in the current population in the current iteration, updating the global optimal solution based on the fitness, and determining whether the iteration termination condition has been met; if not, returning to re-execute the global exploration operation; if the iteration termination condition is met, generating a target optimal task allocation scheme and the flight paths of each UAV. This invention generates an initial population through joint encoding, dynamically adjusts the search process by combining global exploration and local optimization strategies, effectively balancing global search and local development capabilities. At the same time, it introduces a dynamic event response mechanism to quickly generate replanned tracks, which helps improve the allocation efficiency and accuracy of multi-UAV collaborative tasks. Attached Figure Description

[0020] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the multi-UAV collaborative task allocation system provided in this application;

[0022] Figure 2 This is a flowchart illustrating the implementation of the multi-UAV collaborative task allocation method provided in this application embodiment;

[0023] Figure 3 This is a flowchart illustrating the implementation of the first sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0024] Figure 4This is a flowchart illustrating the implementation of the second sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0025] Figure 5 This is a flowchart illustrating the implementation of the third sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0026] Figure 6 This is a flowchart illustrating the implementation of the fourth sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0027] Figure 7 This is a flowchart illustrating the implementation of the fifth sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0028] Figure 8 This is a flowchart illustrating the implementation of the sixth sub-process in the multi-UAV collaborative task allocation method provided in this application embodiment;

[0029] Figure 9 This is a schematic diagram of a multi-UAV collaborative task allocation device provided in an embodiment of this application;

[0030] Figure 10 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0032] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0034] like Figure 1 , Figure 1 This application provides a schematic diagram of a multi-UAV collaborative task allocation system. The system includes a ground control center, an environmental perception subsystem, a HEVCO (Hydrothermal Extremophile Vent Colony Optimization, a swarm intelligence optimization framework based on the extreme bacterial behavior of deep-sea black smokers) optimization engine, a UAV swarm, and a communication and monitoring network. The ground control center has a built-in order management and visualization interface, supporting the import of real-time orders (delivery points) from e-commerce / pharmacy / catering platforms, setting timeliness levels and priorities, and displaying UAV locations, remaining battery power, and task status in real-time on an electronic sand table. The environmental perception subsystem integrates new order insertions, cancellations, and expedited order event flows from urban low-altitude air traffic control APIs (temporary no-fly / height-restricted zones), millimeter-wave weather radar and ground meteorological stations (gust / rain risk areas), and smart transportation / high-rise hoisting early warning (dynamic obstacle) platforms. The continuous output of the environmental perception subsystem provides the HEVCO optimization engine with real-time trajectory replanning. The HEVCO optimization engine is deployed on edge cloud servers along the urban area to implement the multi-UAV collaborative task allocation method of this application. A drone swarm is a cluster of multiple drones, such as eight eVTOL logistics drones. Each drone has a 2kg payload, a flight endurance of 35–45 minutes, a cruising speed of 15 m / s (peak 22 m / s), and is equipped with a drop-out cargo container, ADS-BOut, and obstacle avoidance radar. The communication and monitoring network uses a 5G + millimeter-wave 60GHz converged link, with end-to-end data latency ≤40ms, meeting the requirements for city-level real-time replanning and RTH (Return-to-Home) control. These units are interconnected via a 5GHz wireless link, with data latency not exceeding 50ms, satisfying real-time trajectory replanning requirements.

[0035] This application proposes a multi-UAV cooperative task allocation method, including: jointly encoding the mapping relationship between tasks and UAVs and the task execution order to generate an initial population; sequentially performing plume diffusion, chemotactic climbing, and clustering elite attraction on the population to complete global exploration; performing local perturbation fine-tuning and fusion with elite solutions based on the current optimal solution to dynamically adjust the search radius; updating the global optimal solution through fitness calculation until the termination condition is met, and finally outputting the optimal allocation scheme and track.

[0036] Among these, joint encoding refers to integrating task allocation relationships and execution order into a unified solution vector, which can be implemented using a binary-integer hybrid encoding method. This encoding method avoids the cooperative efficiency loss caused by solution space splitting in traditional methods. Feather diffusion refers to globally perturbing the population using the Lévy flight distribution, which can be achieved by generating random step sizes following the Lévy distribution; its long-tail characteristic helps to escape local optima. Chemotaxis and climbing refers to fine-tuning the task order based on gradient estimation, which can be achieved by approximating the gradient direction using the finite difference method and adjusting the task arrangement along the fitness-increasing direction. Clustering elite attraction refers to identifying high-quality solution regions through clustering algorithms, such as using the K-means algorithm to divide the population into several clusters and using the cluster centers to guide the population's evolutionary direction. Dynamic search radius adjustment refers to exponentially decaying the perturbation amplitude according to the iteration progress, for example, setting the initial radius to 20% of the solution space dimension and a decay coefficient of 0.95 per generation.

[0037] Specifically, in the initial population generation phase, a complete solution space representation is established through joint encoding. In the global exploration phase, a large-scale random perturbation is first introduced through plume diffusion to break the homogeneity of the population; then, chemotactic climbing is used to finely adjust the task order, improving local development capabilities; finally, high-quality solution regions are identified through clustering, and cluster centers are used to replace inferior individuals to maintain population diversity. In the local optimization phase, Gaussian perturbations are applied to the current optimal solution to prevent premature convergence, while historical elite solutions are integrated to retain high-quality genes. The dynamically adjusted search radius maintains a large exploration range in the early stages of iteration, gradually shrinking to focus on high-quality regions in later stages. The fitness function comprehensively considers task completion time, energy consumption coefficient, and risk weight, driving the population to evolve towards a multi-objective equilibrium solution.

[0038] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0039] Please see Figure 2 , Figure 2 This paper illustrates a specific implementation of a multi-UAV collaborative task allocation method.

[0040] It should be noted that if substantially the same result is obtained, the method of this invention is not based on... Figure 1 Limited to the order of the processes shown, this method includes the following steps:

[0041] S1: Jointly encode the mapping relationship between tasks and drones and the task execution order on each drone to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector.

[0042] Before implementing step S1, all drones are by default parked at the same city distribution center (the take-off and landing field near the ground control center) before scheduling begins. The coordinates of this assembly point are marked as: h = (x h ,y h ,zh The reasons are as follows: the urban delivery aircraft BA adopts a replaceable battery design, and centralized take-off and landing facilitate unified battery replacement, loading and aircraft inspection; urban airspace is limited, and scattered deployment will increase conflicts in temporary no-fly zones and ground safety hazards; unified scheduling can make better use of facilities such as automated sorting and smart cabinets in the warehouse.

[0043] Specifically, the basic unit of urban instant delivery is taken as a delivery task (corresponding to a pickup order). For ease of algorithm processing, each task is abstracted into a six-tuple: T j =(id) j ,p j ,(e j ,l j ),w j ,s j ,pr j ), where id j This represents a unique identifier used for integration with the platform's order number; p j =(x j ,y j ,z j (e) represents spatial coordinates, obtained from RTK point clouds or GIS; j ,l j () represents the delivery window. For example, a full-day delivery window is used for regular express delivery, and it can be uniformly set to [0, T]. max ];w j For order weight (kg), used for load constraints; s j The delivery service duration (s) includes placement, signing, and resetting; this application sets the delivery service duration to a fixed 3 seconds; pr j Priority is assigned (urgent orders > regular orders); it can be mapped to a penalty coefficient or a ranking weight. In this embodiment, the mapping relationship between tasks and drones and the task execution order on each drone are jointly encoded to generate an initial population containing N individuals, where each individual in the initial population is a candidate solution vector.

[0044] Please see Figure 3 , Figure 3 A specific implementation of step S1 is shown below:

[0045] S11: Encode the mapping relationship between the tasks and the drones and the execution order of the tasks on each drone to generate a mapping vector and a task order vector. S12: Concatenate the task order vector to the mapping vector to generate the solution vector in the initial population. S13: Initialize auxiliary variables, including the history matrix, fitness vector, and the global optimal solution.

[0046] Specifically, in multi-drone collaborative scenarios, decision-making essentially involves two layers: "task allocation" and "execution order." To complement the swarm intelligence optimization framework, "who executes" and "execution order" are encoded separately. This means encoding the mapping relationship between tasks and drones, as well as the task execution order on each drone, to generate a mapping vector and a task order vector. The mapping vector is A = [a1, a2, ..., a...]. 120 ],a j ∈{1,…,K}, representing task T j U-type drone aj Execution; Task sequence vector U-shaped drone m The task access order. The task order vector is concatenated to the mapping vector to generate the solution vector X = [A||P] in the initial population. (1) ||…||P (m) ]. Historical Matrix To record the state of the population in each generation, the fitness vector F = [f1, f2, ..., f...]. N [: Stores the objective function value of the current population individual. Globally optimal individual] Records the individual with the best fitness in the current population. Global best fitness value.

[0047] S2: Perform plume diffusion, chemotactic climbing and clustering elite attraction processes sequentially on the current population of the initial population to conduct a global exploration operation on the current population and generate a clustered optimized population.

[0048] Please see Figure 4 , Figure 4 A specific implementation of step S2 is shown below:

[0049] S21: Use the Levy flight distribution to globally perturb the current population of the initial population to perform plume diffusion on the current population and generate a perturbed population.

[0050] Specifically, based on the Lévy flight mechanism, long-distance perturbation operations simulate the microbial community's leapfrog exploration using plumes. This involves each individual in the current population... The update is as follows:

[0051]

[0052] Where α∈(0.01,0.2) represents the step size control factor; X represents a Lévy distribution random vector, commonly represented by λ = 1.5; bestThis is the optimal individual in the current iteration. In this embodiment, the entire task of the same geographic cluster is migrated to a random UAV with a probability p1 = 0.4, a step size coefficient α = 0.05, and a stability index λ = 1.5.

[0053] S22: A preset gradient estimation algorithm is used to locally fine-tune the task execution order in the perturbed population, so as to perform chemotaxis and climb the perturbed population and generate a locally fine-tuned population.

[0054] Specifically, chemotaxis is a behavior in which extremophiles depend on chemical signals (such as Fe) 2+ This application describes a microscale displacement mechanism implemented using H2S (Hyper-Sensitive-Sensitive) methods. By estimating the gradient direction of the objective function, it achieves a local fine-grained search similar to a "sensing-movement" mechanism. The update formula is as follows:

[0055]

[0056] Where: η∈(0.001,0.01) represents the chemotactic learning rate; This is the gradient estimate of the fitness function at the current position. Since the true objective function is usually non-differentiable or complex, gradient estimation can be performed using a stochastic perturbation method (such as SPSA).

[0057]

[0058] in, Let be the solution vector of the i-th individual in the t-th generation, δ be the amplitude of the SPSA random perturbation (gradient estimation supplement), and f(.) be the fitness function after penalty. In the embodiments of this application, 2-Opt / 3-Opt is performed on the selected UAV sequence with probability p2 = 0.35, and the local gradient consists of the range increment and the risk increment, with a step size η = 0.01.

[0059] S23: Cluster the locally fine-tuned population and update the locally fine-tuned population according to the candidate solutions corresponding to the cluster centers after clustering, so as to perform elite attraction processing on the locally fine-tuned population and generate the cluster-optimized population.

[0060] Specifically, under high temperature and high pressure environments, extremophiles often form biofilm-like aggregates to improve metabolic efficiency and synergistic adaptation. This application biomimeticly implements an elite aggregation mechanism, namely, selecting multiple local centers C through K-means clustering. k And attract individuals within the corresponding cluster to move closer to it:

[0061]

[0062] Among them, C k For individual X iThe center point of the cluster; γ∈(0.1,0.3) is the cluster attraction strength.

[0063] The embodiments of this application enhance the orderliness of the search direction while maintaining a certain degree of inter-individual diversity, serving as an important bridge connecting local development and global exploration. In one specific embodiment, K-means is executed every 10 generations, replacing the worst individual in the population with the cluster center solution, thus strengthening the "tower-platform" mapping.

[0064] The Lévy flight distribution refers to a random walk pattern with heavy-tailed characteristics. It can be implemented using a random number generator based on the Lévy distribution, enhancing the population's global exploration capability in the solution space by introducing long-step jumps and avoiding the search blind spots caused by traditional uniform perturbations. The pre-estimated gradient algorithm guides local optimization of the solution by approximating the gradient direction, enabling fine-tuning in the perturbed solution space and improving solution quality. Clustering divides the population into multiple subgroups to extract distribution features, and can be implemented using K-means or hierarchical clustering algorithms. The population is updated using candidate solutions at cluster centers, maintaining population diversity and guiding the search direction.

[0065] Specifically, the global perturbation phase generates long-step jump perturbations using the Lévy flight distribution, enabling the population to quickly cover a wide area of ​​the solution space and avoid getting trapped in local optima. The local fine-tuning phase uses gradient information to finely adjust the task execution order, preserving the diversity brought by the perturbation while improving the feasibility of the solution. The clustering phase divides the population into multiple subgroups and calculates cluster centers. By replacing the worst individuals with cluster centers, elite attraction is achieved, maintaining population diversity while focusing the search on high-quality regions. These three phases form a closed-loop mechanism of diffusion-optimization-focusing, enhancing global exploration while maintaining the convergence of the solution.

[0066] Please see Figure 5 , Figure 5 A specific implementation of step S2 is shown below:

[0067] S231: When the preset number of iterations is reached, the K-means clustering algorithm is used to cluster the locally fine-tuned population to obtain the cluster centers.

[0068] S232: Update each individual in the locally fine-tuned population according to the cluster center to generate multiple updated individuals.

[0069] S233: Determine the optimal solution for the cluster center based on the updated individuals, and determine the worst individual in the population after the local fine-tuning.

[0070] S234: Replace the worst individual in the locally fine-tuned population with the optimal solution at the cluster center to generate the cluster-optimized population.

[0071] K-means clustering is a data partitioning method that divides a population into multiple subgroups. Specifically, it uses Euclidean distance to measure the similarity between individuals and iteratively calculates the intra-cluster mean to update the cluster centers, identifying potentially high-quality regions in the solution space. The cluster center is the geometric center of all individuals in each subgroup, obtained by calculating the arithmetic mean of the coordinates in each dimension within the subgroup, representing the solution distribution characteristics of that region. The update calculation refers to the process of adjusting the individual positions based on the cluster centers, using linear interpolation or random offset to move individuals towards the cluster centers to enhance local search capabilities. The optimal solution at each cluster center is the individual with the highest fitness among the candidate solutions corresponding to each cluster center, determined by traversing all cluster centers and comparing their objective function values, guiding the direction of population evolution. The worst individual is the individual with the lowest fitness ranking in the current population, identified through a sorting and filtering mechanism, triggering the replacement of inferior solutions.

[0072] Specifically, when the algorithm reaches a preset iteration node, a clustering analysis process is triggered. First, the K-means algorithm is used to divide the locally optimized population into several subgroups, with the center point of each subgroup reflecting the solution distribution characteristics of that region. Then, a position update operation is performed on each individual, moving it towards its respective cluster center to enhance the targeting of the local search. After fitness evaluation, the updated individuals are selected as elite candidates corresponding to the cluster centers. Simultaneously, a ranking mechanism identifies the individual with the lowest fitness in the current population as the elimination target. Finally, by replacing the worst solution with an elite solution, a stepwise improvement in solution quality is achieved while maintaining the same population size. This process maintains population diversity through periodic clustering analysis, guides the search direction using cluster centers, and avoids ineffective searches by combining an elite replacement mechanism, forming a dynamically balanced optimization process.

[0073] S3: Perform local perturbation fine-tuning based on the current best solution in the clustered population to obtain the fine-tuned best solution, and perform weighted fusion based on the input elite solutions to obtain the fused new solution, and dynamically adjust the search radius based on the current iteration number and the preset iteration number.

[0074] Specifically, the above steps constitute the exploration phase, aiming to achieve a broad global search, while the embodiments of this application constitute the exploration phase, which focuses on a fine-grained search of the current optimal region. This application simulates the mechanism of "retention and fermentation-structural reconstruction-local perturbation" in the vicinity of an environmentally stable region by extreme bacterial communities to accurately exploit local extrema, thereby accelerating convergence and improving solution quality. In the embodiments of this application, local perturbation fine-tuning is performed based on the current optimal solution in the clustered population to obtain the fine-tuned optimal solution, and weighted fusion is performed based on the input elite solutions to obtain a new fused solution. Furthermore, the search radius is dynamically adjusted based on the current iteration number and the preset iteration number.

[0075] Please see Figure 6 , Figure 6 A specific implementation of step S3 is shown below:

[0076] S31: Based on the Gaussian distribution, the current optimal solution in the clustered population is locally perturbed, and after every few iterations, a reverse interference point is introduced to fine-tune the optimal solution after local perturbation, thereby generating the fine-tuned optimal solution.

[0077] Specifically, during the development phase, the population revolves around the globally optimal individual X. best A local perturbation is performed around the solution to find a potential better solution. The local perturbation operation on the previous optimal solution is as follows:

[0078]

[0079] Where: δ∈[0.01,0.1] represents the perturbation intensity, which decreases dynamically with iteration; This represents a Gaussian distribution with a mean of 0 and a covariance of an identity matrix. To prevent overfitting to the current optimal point and falling into a local extremum trap, a set of reverse perturbation points is introduced every certain number of generations (e.g., every 10 generations) to fine-tune the locally perturbed optimal solution, generating a fine-tuned optimal solution. The fine-tuning method for the reverse perturbation points is as follows:

[0080]

[0081] in, This refers to an individual in the optimal solution after fine-tuning.

[0082] S32: Perform weighted fusion on the multiple elite solutions input to obtain the new fused solution.

[0083] Specifically, in the later stages of development, the microbial community exhibits a clear trend of focused contraction, tending to form compressed encirclements around multiple local optima. This application simulates this phenomenon by introducing a multi-elite set. And regenerate individuals within its convex combination space:

[0084]

[0085] Wherein, weight w j It can be obtained by fitness normalization, or it can be used in the following form:

[0086]

[0087] Where ∈ represents a positive number to avoid division by zero, and f(·) is the fitness function. This application can be viewed as simulating the symbiotic adsorption behavior of bacterial communities on multiple nutrient hotspots, improving convergence stability and expanding the local search radius.

[0088] S33: Dynamically adjust the search radius based on the current iteration number and the preset iteration number.

[0089] Specifically, as the iteration progresses, the search range will gradually shrink. The shrinkage radius r(t) is controlled as follows:

[0090]

[0091] Where, r init Let T be the initial perturbation range, and t represent the current iteration number. max β represents the maximum number of iterations; β∈[1,3] is the contraction curvature control factor. This application ensures that the algorithm has the ability to converge gradually from the global to the local level, which conforms to the evolutionary paradigm of "diffusion first, then concentration" of most biomimetic optimization methods. During the development phase, by introducing a three-in-one mechanism of perturbation fine-tuning, multi-elite weighting, and contraction control, not only is the local search accuracy improved, but early convergence and elite trap problems are also avoided.

[0092] Among them, Gaussian distribution perturbation refers to generating a random offset centered on the current optimal solution that follows a normal distribution. Specifically, it can be implemented using Gaussian noise with a standard deviation of 10%-30% of the current search radius, used for fine-grained searching within the neighborhood of the optimal solution. Reverse perturbation points refer to generating offsets in the opposite direction to the current perturbation within a preset period. Dynamically adjusting the search radius refers to exponentially decaying the perturbation amplitude according to the iteration progress, used to balance global exploration and local development.

[0093] S4: Calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution based on the fitness, and determine whether the iteration termination condition has been met. If not, return and re-execute the global exploration operation.

[0094] Specifically, regarding the optimization objective and constraints, the multi-objective fitness function in this embodiment is:

[0095] And w1+w2+w3=1,

[0096] Among them, C m For distance / speed and operational stop, the value of U represents the drone. m The total time to complete all tasks and return home, E m =κD m +ηΔh m Indicates equivalent energy consumption; This represents the cumulative risk value. Furthermore, this example involves six hard constraints, as shown in Table 1 below:

[0097]

[0098] Table 1

[0099] Based on the above constraints, a unified formula is constructed, as follows:

[0100]

[0101] This application evaluates individual populations based on overall completion time, energy consumption, risk, and task rewards, and updates the global optimal solution. Specifically, when the preset maximum number of iterations is reached or the convergence condition is met, the iteration termination condition is determined to have been met.

[0102] S5: If the iteration termination condition is met, generate the target optimal task allocation scheme and the flight paths of each UAV.

[0103] Specifically, if the iteration termination condition is met, the target optimal task allocation scheme and the flight paths of each UAV are generated. The target optimal task allocation scheme and the flight paths of each UAV are then distributed to the UAV cluster for task allocation.

[0104] Please see Figure 7 , Figure 7 A specific implementation method following step S5 is shown below:

[0105] S5A: Performs tasks according to the optimal task allocation scheme for the target and the flight paths of each UAV, and detects dynamic events in real time during the task execution process.

[0106] Specifically, every 20 iterations or when a dynamic event is detected, the bottom 20% of individuals in fitness ranking can be randomly reset to restore diversity. During task execution, dynamic events are detected in real time, including the addition of no-fly zones, movement of strong wind risk zones, tower alarm re-inspection, and drone failure events.

[0107] S5B: When a dynamic event is detected, the affected track segment is located as a local window, and random window reset and local iterative optimization are performed based on the local window to generate a replanned track.

[0108] Dynamic events refer to environmental changes during mission execution, such as temporary no-fly zones, sudden order insertions, or sudden drops in onboard battery power. These can be detected in real time through sensor data fusion and airspace management systems to trigger the trajectory replanning mechanism. A local window refers to a continuous trajectory segment directly affected by dynamic events. Locating this window narrows the optimization scope and avoids redundant calculations of the global path. A random reset window involves randomly adjusting waypoints within the local window according to a preset ratio. This can be achieved by generating new waypoint coordinates using uniform or Gaussian distributions, introducing diversity while preserving the original path structure. A scaling parameter reduces the step size range for plume diffusion and chemotactic climb operations, typically set to 0.1 to 0.3 times the original parameter, enabling refined searching within the local window. The minimum increment algorithm adjusts the path with the goal of minimizing the trajectory length increment, while the Bézier curve generates a smooth curve trajectory using control points. The combined application of these two algorithms ensures the continuity and flyability of the replanned trajectory. The fitness increment threshold refers to the upper limit of the allowed changes in path quality. It can be dynamically set through historical optimization data to control the extent of changes in the replanning scheme.

[0109] Specifically, when the UAV formation is performing a mission, the airspace monitoring system continuously collects meteorological data and air traffic control instructions. If a temporary no-fly zone is detected, the segments of the affected UAVs' current flight paths that intersect with the no-fly zone are marked as local windows. 30% of the waypoints within this window are randomly repositioned, generating new local segments containing the disturbed waypoints. A plume diffusion operation is performed on this segment using scaling parameters, exploring the feasible solution space through small-scale random walks, followed by gradient ascent optimization using chemotactic ramping. The optimized segment connects adjacent waypoints using a minimum increment algorithm, and sharp angles at path transitions are eliminated using third-order Bézier curves. Finally, the fitness increment of the new flight path relative to the original scheme is calculated. If the increment does not exceed a preset threshold, only the local flight paths of the affected UAVs are updated, while the remaining UAVs maintain their original flight plans.

[0110] Please see Figure 8 , Figure 8 A specific implementation of step S5B is shown below:

[0111] S5B1: When a dynamic event is detected, the affected track segment is located as the local window.

[0112] Specifically, locate the affected drone flight path segments, with no more than 10 waypoints, and use the affected flight path segments as local windows.

[0113] S5B2: Randomly reset the waypoints in the local window according to a preset ratio to obtain the reset local segment.

[0114] Specifically, 50% of waypoints within a local window are randomly reset to a safe and feasible region.

[0115] S5B3: Based on the reset local segment, use scaled parameter segments to perform plume diffusion and chemotactic climbing, so as to perform local iterative optimization on the reset local segment and generate an optimized local segment.

[0116] Specifically, using the scaling parameter α loc =0.02,η loc =0.005 Performed plume diffusion and chemotactic climbing 25 times.

[0117] S5B4: The minimum distance increment algorithm and Bézier curves are used to process the optimized local flight segment to generate a smoothed flight track.

[0118] Specifically, the minimum distance algorithm is used to bypass obstacles, and the trajectory is smoothed with a fifth-order Bézier curve to ensure that the constraints (C5, C6) are met.

[0119] S5B5: Calculate the fitness increment of the changed part in the smoothed track. If the fitness increment does not exceed a preset threshold, the smoothed track is used as the replanned track.

[0120] The local window refers to a segment of a track composed of consecutive waypoints affected by a dynamic event. Specifically, the boundary of the affected area can be determined by the event trigger location and a preset safety radius, and the original track within this area is used as the optimization range. This feature restricts the optimization range to a local area, avoiding the computational resource consumption caused by global path reconstruction. Preset proportional random reset refers to randomly selecting waypoints within the local window for position reset at a fixed ratio. For example, selecting 30% of waypoints in the affected segment for coordinate perturbation preserves the original path framework while introducing diversity. Scaled parameter plume diffusion refers to proportionally reducing the step size factor in the original optimization algorithm, for example, adjusting the diffusion radius to 1 / 5 of its original value, performing Lévy flight distribution perturbation within a confined space, balancing local search efficiency and convergence speed. Chemitropic climbing uses gradient estimation methods to fine-tune the reset waypoints, for example, calculating the fitness change direction using the finite difference method and adjusting the waypoint coordinates along the gradient ascent direction. The minimum increment algorithm is used to calculate the total distance increment between the adjusted segment and the original track, prioritizing the feasible adjustment scheme with the smallest increment. Bézier curve processing smooths out abrupt changes in heading by inserting control points, eliminating safety hazards caused by sharp turns. The fitness increment threshold determination mechanism compares the magnitude of change in the objective function value before and after adjustment; for example, it accepts the new solution when the change does not exceed 5%, preventing excessive deviation from the global optimum.

[0121] Specifically, when a drone swarm encounters a temporary no-fly zone or equipment alarm during a delivery mission, the flight path segments of the affected drones are quickly located as local windows. Within these windows, some waypoints are randomly selected for coordinate reset; for example, the latitude and longitude coordinates of three out of ten waypoints are randomly adjusted to generate new candidate solutions for local flight segments. Subsequently, a plume diffusion operation with scaled parameters is used to perturb the reset waypoints with a small step size, avoiding jumping out of the local optimum. During the chemotactic ramping phase, the influence of waypoint movement on task completion time, energy consumption, and risk indicators is calculated to gradually optimize the waypoint positions. The optimized local flight segment uses a minimum increment algorithm to calculate the connection point with the original flight path, ensuring that the total path increment is minimized. At the same time, Bézier curves are used to eliminate abrupt changes in heading angle, generating a smooth trajectory that conforms to the drone's maneuverability. Finally, based on a fitness increment threshold, the flight path is updated only when the optimization effect meets preset requirements, maintaining the efficiency of multi-drone collaboration.

[0122] Please refer to Figure 9 As a response to the above Figure 1 The implementation of the method shown in this application provides an embodiment of a multi-UAV collaborative task allocation device, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0123] like Figure 9 As shown, the multi-UAV collaborative task allocation device in this embodiment includes: a population initialization module 61, a global exploration module 62, a global exploration module 63, an optimal solution update module 64, and an allocation scheme generation module 65, wherein:

[0124] Population initialization module 61 is used to jointly encode the mapping relationship between tasks and UAVs and the task execution order on each UAV to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector;

[0125] The global exploration module 62 is used to sequentially perform plume diffusion, chemotactic climbing and clustering elite attraction processes on the current population of the initial population, so as to perform global exploration operations on the current population and generate a clustered optimized population.

[0126] The local optimization module 63 is used to perform local perturbation fine-tuning based on the current best solution in the clustered population after optimization, to obtain the fine-tuned best solution, and to perform weighted fusion based on the input elite solutions to obtain a new fused solution, and to dynamically adjust the search radius based on the current iteration number and the preset iteration number.

[0127] The optimal solution update module 64 is used to calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution according to the fitness, and determine whether the iteration termination condition has been met. If not, it returns to re-execute the global exploration operation.

[0128] The allocation scheme generation module 65 is used to generate the target optimal task allocation scheme and the flight paths of each UAV when the iteration termination condition is met.

[0129] Furthermore, the allocation scheme generation module 65 also includes:

[0130] The dynamic event detection module is used to perform tasks according to the target optimal task allocation scheme and the flight paths of each UAV, and to detect dynamic events in real time during the task execution process.

[0131] The track replanning module is used to locate the affected track segment as a local window when a dynamic event is detected, and to perform random window reset and local iterative optimization based on the local window to generate a replanned track.

[0132] Furthermore, the trajectory replanning module includes:

[0133] A track segment localization unit is used to locate the affected track segment as the local window when a dynamic event is detected.

[0134] A waypoint reset unit is used to randomly reset the waypoints in the local window according to a preset ratio to obtain a reset local flight segment.

[0135] A local iterative optimization unit is used to perform local iterative optimization on the reset local segments based on the reset local segments using scaled-down parameter segments for plume diffusion and chemotactic climbing, thereby generating optimized local segments.

[0136] The trajectory smoothing unit is used to process the optimized local segment using the minimum distance increment algorithm and Bézier curves to generate a smoothed trajectory.

[0137] The fitness increment calculation unit is used to calculate the fitness increment of the changed part in the smoothed track. If the fitness increment does not exceed a preset threshold, the smoothed track is used as the replanned track.

[0138] Furthermore, the population initialization module 61 includes:

[0139] The plume diffusion unit is used to globally perturb the current population of the initial population using the Levy flight distribution, so as to perform plume diffusion on the current population and generate a perturbed population.

[0140] A chemotactic climbing unit is used to locally fine-tune the task execution order in the perturbed population using a preset gradient estimation algorithm, so as to perform chemotactic climbing on the perturbed population and generate a locally fine-tuned population.

[0141] The clustering optimization unit is used to cluster the locally fine-tuned population and update the locally fine-tuned population according to the candidate solutions corresponding to the cluster centers after clustering, so as to perform elite attraction processing on the locally fine-tuned population and generate the cluster-optimized population.

[0142] Furthermore, the clustering optimization unit includes:

[0143] The cluster center generation unit is used to cluster the locally fine-tuned population using the K-means clustering algorithm when a preset number of iterations is reached, thereby obtaining cluster centers.

[0144] An individual update unit is used to perform update calculations on each individual in the locally fine-tuned population according to the cluster center, generating multiple updated individuals;

[0145] The optimal solution determination unit is used to determine the optimal solution of the cluster center based on the updated individuals, and to determine the worst individual in the locally fine-tuned population.

[0146] An individual replacement unit is used to replace the worst individual in the locally fine-tuned population with the optimal solution at the cluster center, thereby generating the cluster-optimized population.

[0147] Furthermore, the local optimization module 63 includes:

[0148] The local perturbation unit is used to locally perturb the current optimal solution in the clustered population based on Gaussian distribution, and to introduce a reverse perturbation point after every few iterations to fine-tune the optimal solution after local perturbation, thereby generating the fine-tuned optimal solution.

[0149] A weighted fusion unit is used to perform weighted fusion on multiple input elite solutions to obtain a new fused solution;

[0150] The search radius adjustment unit is used to dynamically adjust the search radius based on the current iteration number and the preset iteration number.

[0151] Furthermore, the population initialization module 61 includes:

[0152] The encoding unit is used to encode the mapping relationship between the tasks and the drones and the execution order of the tasks on each drone, and generate a mapping vector and a task order vector.

[0153] A solution vector generation unit is used to concatenate the task order vector after the mapping vector to generate the solution vector in the initial population.

[0154] An auxiliary variable initialization unit is used to initialize auxiliary variables, wherein the auxiliary variables include the history matrix, the fitness vector, and the global optimal solution.

[0155] To address the aforementioned technical problems, embodiments of this application also provide an electronic device. Please refer to [link / reference needed] for details. Figure 10 , Figure 10 This is a basic structural block diagram of the electronic device in this embodiment.

[0156] Electronic device 7 includes a memory 71, a processor 72, and a network interface 73 that are interconnected via a system bus. It should be noted that... Figure 10 Only an electronic device 7 with three components—memory 71, processor 72, and network interface 73—is shown. However, it should be understood that implementing all shown components is not required; more or fewer components can be implemented alternatively. Those skilled in the art will understand that the electronic device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0157] Electronic devices can be computing devices such as desktop computers, laptops, PDAs, and cloud servers. Electronic devices can interact with users through methods such as keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0158] The memory 71 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 71 may be an internal storage unit of the electronic device 7, such as the hard disk or memory of the electronic device 7. In other embodiments, the memory 71 may also be an external storage device of the electronic device 7, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 7. Of course, the memory 71 may also include both internal storage units and external storage devices of the electronic device 7. In this embodiment, the memory 71 is typically used to store the operating system and various application software installed on the electronic device 7, such as the program code of a multi-UAV collaborative task allocation method. In addition, the memory 71 may also be used to temporarily store various types of data that have been output or will be output.

[0159] In some embodiments, processor 72 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. Processor 72 is typically used to control the overall operation of electronic device 7. In this embodiment, processor 72 is used to run program code stored in memory 71 or process data, for example, to run the program code of the above-described multi-UAV cooperative task allocation method to implement various embodiments of the multi-UAV cooperative task allocation method.

[0160] The network interface 73 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the electronic device 7 and other electronic devices.

[0161] This application also provides another embodiment, namely, a computer-readable storage medium storing a computer program that can be executed by at least one processor to cause the at least one processor to perform the steps of the multi-UAV cooperative task allocation method described above.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0163] Obviously, the embodiments described above are merely some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the scope of this application. This application can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of protection of this application.

Claims

1. A method for multi-UAV collaborative task allocation, characterized in that, include: The mapping relationship between tasks and drones and the task execution order on each drone are jointly encoded to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector; The current population of the initial population is subjected to plume diffusion, chemotactic climbing and clustering elite attraction processes in sequence to perform global exploration operations on the current population and generate a clustered optimized population. Local perturbation fine-tuning is performed on the current best solution in the clustered population after optimization to obtain the fine-tuned best solution. Weighted fusion is performed on the input elite solutions to obtain the fused new solution. The search radius is dynamically adjusted according to the current iteration number and the preset iteration number. Calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution based on the fitness, and determine whether the iteration termination condition has been met. If not, return and re-execute the global exploration operation. If the iteration termination condition is met, the target optimal task allocation scheme and the flight paths of each UAV are generated. The process of sequentially performing plume diffusion, chemotactic ascent, and elite clustering on the current population of the initial population to conduct a global exploration operation on the current population and generate a clustered optimized population includes: The Lévy flight distribution is used to globally perturb the current population of the initial population to induce plume diffusion in the current population and generate a perturbed population. A preset gradient estimation algorithm is used to locally fine-tune the task execution order in the perturbed population in order to perform chemotaxis and climb the slope of the perturbed population, thereby generating a locally fine-tuned population. The locally fine-tuned population is clustered, and the locally fine-tuned population is updated according to the candidate solutions corresponding to the cluster centers to perform elite attraction processing on the locally fine-tuned population and generate the cluster-optimized population. Clustering the locally fine-tuned population and updating the population based on the candidate solutions corresponding to the cluster centers to perform elite attraction processing on the locally fine-tuned population, generating the cluster-optimized population, including: When the preset number of iterations is reached, the K-means clustering algorithm is used to cluster the locally fine-tuned population to obtain the cluster centers; Each individual in the locally fine-tuned population is updated according to the cluster center to generate multiple updated individuals; The optimal solution for the cluster center is determined based on the updated individuals, and the worst individual in the population after the local fine-tuning is also determined. The worst individual in the locally fine-tuned population is replaced by the optimal solution at the cluster center to generate the cluster-optimized population.

2. The multi-UAV collaborative task allocation method according to claim 1, characterized in that, If the iteration termination condition is met, after generating the target optimal task allocation scheme and the flight paths of each UAV, the method further includes: The task is executed according to the target optimal task allocation scheme and the flight paths of each UAV, and dynamic events are detected in real time during the task execution process; When a dynamic event is detected, the affected track segment is located as a local window, and random window reset and local iterative optimization are performed based on the local window to generate a replanned track.

3. The multi-UAV collaborative task allocation method according to claim 2, characterized in that, When a dynamic event is detected, the affected track segment is located as a local window, and random window reset and local iterative optimization are performed based on the local window to generate a replanned track, including: When a dynamic event is detected, the affected track segment is located as the local window; The waypoints in the local window are randomly reset according to a preset ratio to obtain the reset local flight segment; Based on the reset local flight segment, the plume diffusion and chemotactic climbing are carried out using the scaled parameter segment to perform local iterative optimization of the reset local flight segment and generate the optimized local flight segment. The optimized local flight segment is processed using the minimum increment algorithm and Bézier curves to generate a smoothed flight track. Calculate the fitness increment of the changed portion in the smoothed trajectory. If the fitness increment does not exceed a preset threshold, the smoothed trajectory is used as the replanned trajectory.

4. The multi-UAV collaborative task allocation method according to claim 1, characterized in that, The process of performing local perturbation fine-tuning based on the current optimal solution in the clustered population to obtain the fine-tuned optimal solution, performing weighted fusion based on the input elite solutions to obtain a new fused solution, and dynamically adjusting the search radius based on the current iteration number and the preset iteration number includes: The current optimal solution in the clustered population is locally perturbed based on the Gaussian distribution, and after every few iterations, a reverse interference point is introduced to fine-tune the optimal solution after local perturbation, thereby generating the fine-tuned optimal solution. The multiple elite solutions are weighted and fused to obtain the new fused solution; The search radius is dynamically adjusted based on the current iteration number and the preset iteration number.

5. The multi-UAV cooperative task allocation method according to any one of claims 1 to 4, characterized in that, The process of jointly encoding the mapping relationship between tasks and drones and the task execution order on each drone to generate an initial population containing N individuals includes: The mapping relationship between the tasks and the drones and the execution order of the tasks on each drone are encoded to generate a mapping vector and a task order vector; The task order vector is concatenated to the mapping vector to generate the solution vector in the initial population; Initialize auxiliary variables, which include the history matrix, fitness vector, and the global optimal solution.

6. A multi-UAV collaborative task allocation device, characterized in that, include: The population initialization module is used to jointly encode the mapping relationship between tasks and UAVs and the task execution order on each UAV to generate an initial population containing N individuals, wherein each individual in the initial population is a candidate solution vector; The global exploration module is used to sequentially perform plume diffusion, chemotactic climbing, and clustering elite attraction processes on the current population of the initial population in order to perform global exploration operations on the current population and generate a clustered optimized population. The local optimization module is used to perform local perturbation fine-tuning based on the current best solution in the clustered population to obtain the fine-tuned best solution, and to perform weighted fusion based on the input elite solutions to obtain a new fused solution, and to dynamically adjust the search radius based on the current iteration number and the preset iteration number. The optimal solution update module is used to calculate the fitness of each individual in the current population in the current iteration, update the global optimal solution according to the fitness, and determine whether the iteration termination condition has been met. If not, it returns to re-execute the global exploration operation. The allocation scheme generation module is used to generate the target optimal task allocation scheme and the flight paths of each UAV when the iteration termination condition is met. The global exploration module includes: The plume diffusion unit is used to globally perturb the current population of the initial population using the Levy flight distribution, so as to perform plume diffusion on the current population and generate a perturbed population. A chemotactic climbing unit is used to locally fine-tune the task execution order in the perturbed population using a preset gradient estimation algorithm, so as to perform chemotactic climbing on the perturbed population and generate a locally fine-tuned population. The clustering optimization unit is used to cluster the locally fine-tuned population and update the locally fine-tuned population according to the candidate solutions corresponding to the cluster centers after clustering, so as to perform elite attraction processing on the locally fine-tuned population and generate the cluster-optimized population. The clustering optimization unit includes: The cluster center generation unit is used to cluster the locally fine-tuned population using the K-means clustering algorithm when a preset number of iterations is reached, thereby obtaining cluster centers. An individual update unit is used to perform update calculations on each individual in the locally fine-tuned population according to the cluster center, generating multiple updated individuals; The optimal solution determination unit is used to determine the optimal solution of the cluster center based on the updated individuals, and to determine the worst individual in the locally fine-tuned population. An individual replacement unit is used to replace the worst individual in the locally fine-tuned population with the optimal solution at the cluster center, thereby generating the cluster-optimized population.

7. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the multi-UAV cooperative task allocation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the multi-UAV cooperative task allocation method as described in any one of claims 1 to 5.

Images