A multi-unmanned aerial vehicle cooperative task allocation method based on hierarchical dynamic partitioning and hybrid path planning and a computing device

Abstract

The application discloses a kind of based on layered dynamic partition and hybrid path planning multi-unmanned plane cooperative task allocation method, comprising: constructing layered dynamic partition model, task space is globally initially divided, and based on environmental dynamic change, local adaptive clustering and region boundary adjustment are carried out, and the task subset of each unmanned plane is generated;Establish multi-objective cooperative optimization model, to minimize the total flight distance of cluster and task completion time as target, combine task priority and the task subset is evenly distributed;Hybrid path planning algorithm is used to solve the task sequence after distribution, based on bionic optimization framework, initial flight path is generated by fusing manifold gradient descent strategy and topological difference evolution strategy;Introduce safety distance dynamic adjustment mechanism, and the initial flight path is robustly avoided and corrected in random obstacle environment, and the multi-unmanned plane cooperative task allocation and path planning result is output.The application realizes the efficient, safe and balanced operation of multi-unmanned plane cluster.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control and path planning technology, and in particular to a multi-UAV collaborative task allocation method and computing device based on hierarchical dynamic partitioning and hybrid path planning. Background Technology

[0002] In recent years, multi-UAV systems have been increasingly widely used in logistics delivery, emergency search and rescue, and military reconnaissance. In multi-UAV collaborative operation scenarios, task allocation and path planning are core technologies to ensure efficient cluster execution. As the application environment evolves from static to dynamic and complex scenarios, the scale of tasks expands and environmental complexity gradually increases. Existing technologies generally face technical bottlenecks when dealing with randomly distributed obstacles, dynamically changing task scales, and differentiated task priorities.

[0003] For example, balancing computational complexity and environmental adaptability is difficult. Traditional global optimization algorithms are prone to the curse of dimensionality when obstacle density increases, while local replanning strategies lead to computational redundancy, making it difficult to achieve a balance between real-time performance and planning quality. Load balancing and path efficiency are not well coordinated; most task allocation models only pursue the shortest global path without fully considering the variance constraints of the number of task points per drone, resulting in some drones in the cluster being overloaded and reducing the overall task completion rate. Dynamic obstacle avoidance is also weak; existing path planning methods mostly rely on fixed safety thresholds. When the number and location of obstacles change randomly, frequent global path replanning is easily triggered, significantly increasing response latency and energy consumption.

[0004] Therefore, there is an urgent need for a multi-UAV collaborative task allocation and path planning method that can balance computational efficiency, load balancing, and dynamic obstacle avoidance robustness. Summary of the Invention

[0005] This invention aims to provide a multi-UAV collaborative task allocation method (Hierarchical Dynamic Partitioning-Hybrid Path Planning Method, HDP-HPPM) and computing device based on hierarchical dynamic partitioning and hybrid path planning. It uses a hybrid path planning method based on manifold gradient descent and topological differential evolution to improve the artificial hummingbird algorithm, and a robust real-time obstacle avoidance method for UAV swarms based on a dynamic adjustment mechanism for safe distance. This enables the effective allocation of multi-UAV tasks in dynamic environments, thereby solving the technical problems of high computational complexity of task allocation, poor environmental adaptability, unbalanced swarm load, and frequent obstacle avoidance replanning in existing technologies under dynamic random obstacle environments.

[0006] According to one aspect of the present invention, a multi-UAV cooperative task allocation method based on hierarchical dynamic partitioning and hybrid path planning is provided, the method comprising: A hierarchical dynamic partitioning model is constructed to perform a global initial partitioning of the task space, and local adaptive clustering and region boundary adjustment are performed based on dynamic environmental changes to generate task subsets for each UAV. A multi-objective collaborative optimization model is established with the goal of minimizing the total flight distance of the cluster and the mission completion time, and the task subset is evenly allocated based on the task priority. A hybrid path planning algorithm is used to solve the assigned task sequence. The hybrid path planning algorithm is based on a biomimetic optimization framework and integrates manifold gradient descent strategy and topological differential evolution strategy to generate the initial flight path. A dynamic adjustment mechanism for safe distance is introduced to perform robust obstacle avoidance correction on the initial flight path in a random obstacle environment, and output the results of multi-UAV collaborative task allocation and path planning.

[0007] According to some embodiments, constructing the hierarchical dynamic partitioning model includes: The K-means algorithm is used to perform global coarse-grained partitioning of the task points to determine the initial number of partitions K. The formula for calculating the value of K is as follows:

[0008] in, The total number of task points. Density adjustment factor; Within the initial partition, an adaptive DBSCAN algorithm is used for local fine-grained clustering, with a neighborhood radius of [missing information]. Based on partition characteristics, the following is adaptively determined:

[0009] in, Indicates the equivalent radius of the partition. The standard deviation of the distance between neighboring points. This is a smoothing factor.

[0010] According to some embodiments, establishing a multi-objective collaborative optimization model includes: The cruise distance objective function is constructed by summing the Euclidean distances between consecutive task points on the assigned paths of each UAV, and a load balancing term is introduced to constrain the variance of the number of task points of each UAV. The priority of cruise efficiency and load balancing is coordinated by weighting coefficients to achieve balanced allocation of task points.

[0011] According to some embodiments, the method of using a hybrid path planning algorithm to solve the assigned task sequence includes: Initialize an artificial hummingbird population and map the task access sequence to individual location vectors; During the iterative search process, a local optimization phase based on manifold gradient descent and a global exploration phase based on topological differential evolution are executed alternately. The hybrid path planning algorithm updates the individual position using the following formula:

[0012] in, This represents the position vector of the i-th individual after the (t+1)-th iteration. Let represent the current position vector of the i-th individual at the t-th iteration; b is the guiding factor, used to control the weight of the influence of topological neighborhood information on individual updates; D is the step size adjustment coefficient, used to balance global exploration and local development. Let be the topological neighborhood set of the current solution, that is, the set of indices of all neighboring individuals that have a topological connection with individual i; | | is a topological neighborhood The number of individuals in the middle; j∈ This indicates traversing the topological neighborhood. Each neighboring individual j in the process; Let i be the position vector of the neighboring individual j at the t-th iteration; t is the current iteration number, and i is the index number of the current individual.

[0013] According to some embodiments, the local optimization stage based on manifold gradient descent includes: constructing a manifold constraint surface in the path solution space, calculating the gradient direction of the objective function with respect to individual positions, updating the step size along the opposite direction of the gradient, and generating a manifold gradient term to improve the local search stability and convergence speed of the algorithm in complex path optimization.

[0014] According to some embodiments, the global exploration phase based on topological differential evolution includes: constructing a topological adjacency network among individuals, selecting differential mutation vectors based on the topological adjacency network, performing crossover and selection operations on the differential mutation vectors, and generating topological differential evolution terms.

[0015] According to some embodiments, the balanced allocation of the task subset based on task priority includes: Different tasks are assigned priority weight coefficients, and a high-priority task response delay penalty term is set in the optimization model. By dynamically adjusting the weight of the task access sequence, the high-priority task is guaranteed to respond first in the path planning.

[0016] According to some embodiments, the dynamic adjustment mechanism for the safe distance includes: The system senses the number and location changes of obstacles in the environment in real time, and dynamically calculates the minimum safe distance threshold based on the obstacle distribution density and the drone's flight status. The formula for the safe distance threshold is:

[0017] in, As a baseline safety distance, Indicates the first The number of obstacles within the perception range of each drone. To effectively perceive area, The current velocity vector, parameters and These are the environmental density sensitivity and the velocity compensation coefficient, respectively. When a planned path node intrudes into the safety interval threshold range, a local path replanning or node offset correction is triggered.

[0018] According to another aspect of the present invention, a computing device is provided, including a processor; and a memory storing a computer program that, when executed by the processor, implements the method as described in any of the preceding claims.

[0019] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the method as described in any of the preceding claims.

[0020] The technical solutions provided by the embodiments of the present invention have the following beneficial effects: This invention addresses the computational bottleneck in ultra-large-scale tasks by constructing a hierarchical dynamic partitioning model, which decomposes a large-scale task space into multiple smaller sub-regions for parallel processing. This effectively solves the balance between computational complexity and environmental adaptability in task allocation within dynamic environments, reducing the computational load on high-dimensional search spaces. The proposed hybrid path planning method integrates manifold gradient descent (MG) and topological differential evolution (TD) strategies. Manifold gradient descent improves the stability and convergence speed of local searches, while topologically aware differential mutation enhances directional selectivity, overcoming the tendency of single algorithms to get trapped in local optima. This enables the algorithm to exhibit excellent convergence and path optimization capabilities in random obstacle environments. A robust obstacle avoidance strategy combining dynamic safety distance adjustment and hybrid planning is employed, which adapts to random changes in the number, location, and task scale of obstacles. By replacing global replanning with local path fine-tuning, it significantly reduces response latency and maintains task execution efficiency.

[0021] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] Figure 1This is a flowchart illustrating a method for multi-UAV collaborative task allocation based on hierarchical dynamic partitioning and hybrid path planning, as an example embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the hierarchical data flow of the hierarchical dynamic partitioning and hybrid path planning method in an example embodiment of the present invention.

[0024] Figure 3 This is a comparison chart of response times for two architectures, hierarchical partitioning and non-hierarchical partitioning, based on an example embodiment.

[0025] Figure 4 This is a comparison chart of task completion rates for two architectures, hierarchical partitioning and non-hierarchical partitioning, based on an example embodiment.

[0026] Figure 5 This is a global partition overview diagram of an embodiment of the present invention.

[0027] Figure 6 This is a schematic diagram of path planning and obstacle avoidance in a local area according to an embodiment of the present invention.

[0028] Figure 7 This is a block diagram of a computing device according to an exemplary embodiment of the present invention. Detailed Implementation

[0029] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the relevant invention and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0030] like Figure 1 As shown in the figure, this embodiment proposes a multi-UAV collaborative task allocation method based on hierarchical dynamic partitioning and hybrid path planning.

[0031] In S101, a hierarchical dynamic partitioning model is constructed to perform a global initial partitioning of the task space, and local adaptive clustering and region boundary adjustment are performed based on dynamic environmental changes to generate task subsets for each UAV.

[0032] According to some embodiments, in multi-UAV task allocation, reasonable task point partitioning is a prerequisite for efficient planning. Considering the random distribution of task points, a hierarchical dynamic partitioning model is constructed, and the K-means algorithm is used for initial partitioning at the global level.

[0033] Specifically, the K-means algorithm is used to perform global coarse-grained partitioning of the task points to determine the initial number of partitions K. The formula for calculating the value of K is shown in formula (1):

[0034] in, The total number of task points. This is a density adjustment factor.

[0035] Furthermore, in order to simultaneously consider both cruising distance and zone balance, the objective function is defined by formula (2):

[0036] in, It is the first The centroid coordinates of each partition It is a partition load balancing item. These are the corresponding weighting coefficients. The set of task points assigned to the i-th drone. Let (x, y) be the coordinate vector of the task point in the two-dimensional plane.

[0037] This embodiment uses the K-means algorithm to maintain overall cruise efficiency while avoiding excessive differences in partition size, thus laying the foundation for subsequent local task allocation.

[0038] Based on global partitioning, this embodiment also introduces an adaptive DBSCAN algorithm for local partitioning, taking into account the differences in task point density within different regions. Specifically, within the initial partition, the adaptive DBSCAN algorithm is used for local fine-grained clustering, with a neighborhood radius... Based on the partition characteristics, the following is determined adaptively using formula (3):

[0039] in, Indicates the equivalent radius of the partition. The standard deviation of the distance between neighboring points. This is a smoothing factor. The DBSCAN algorithm is used to narrow the neighborhood range in dense regions and expand the radius in sparse regions, thereby ensuring the rationality of clustering results under different partitions.

[0040] Furthermore, to improve the rationality of task allocation, a distance metric based on formula (4) is introduced:

[0041] in, The coordinates of the two task points are two-dimensional coordinates. It is the Euclidean distance between two points. For the Sigmoid term, the value is approximately 0.5 at close range (magnification by 1.5 times) and →1 at long range (magnification approaches 2 times).

[0042] This embodiment enhances the distinguishability between nearby task points, thereby improving the matching rationality between the UAV and the task. To ensure dynamic convergence of parameters, the adjustment process of the neighborhood radius is controlled by the differential equations of formulas (5) and (6):

[0043]

[0044] in, It is the search radius of the j-th neighborhood. The gradient learning rate (α=0.01) controls... The strength of the gradient's effect on the radius. It is the attenuation coefficient (β=0.005) to prevent the radius from growing indefinitely.

[0045] The fitness function is the compactness of each UAV partition. The mean, Let be the variance of all task points within the k-partition of the UAV to the centroid. This represents the spatial variance of the task points assigned to drone k.

[0046] Ultimately, local task allocation can adaptively adapt to the task distribution in different areas, achieving a reasonable division of the UAV task point set.

[0047] In S103, a multi-objective collaborative optimization model is established with the goal of minimizing the total flight distance of the cluster and the task completion time, and the task subset is evenly allocated in combination with the task priority.

[0048] According to some embodiments, establishing a multi-objective collaborative optimization model includes constructing a cruise distance objective function based on the sum of Euclidean distances between consecutive task points on each UAV's assigned path, introducing a load balancing term to constrain the variance of the number of task points for each UAV, and coordinating the priorities of cruise efficiency and load balancing through weight coefficients to achieve balanced allocation of task points. The balanced allocation of the task subset based on task priority includes assigning priority weight coefficients to different tasks, setting a high-priority task response delay penalty term in the optimization model, and ensuring the early response of high-priority tasks in path planning by dynamically adjusting the weights of the task access sequence.

[0049] Specifically, the path length of each UAV is defined as the sum of the Euclidean distances between consecutive task points. To ensure balanced load across all UAVs, a regional summation method is used instead of a global summation, as shown in formula (7):

[0050] in, This indicates the total number of drones in the cluster. It is the first The number of task points assigned to each drone. It is a drone The The coordinates of each task point are calculated. This objective function aims to minimize the overall flight distance of the UAVs, thereby improving task execution efficiency and reducing energy consumption. By independently calculating the path length of each UAV, the differences in individual UAV task loads can be reflected, providing a quantitative basis for task allocation and path optimization.

[0051] In multi-UAV inspection missions, the importance and urgency of different inspection points vary. High-priority inspection points typically require more frequent or timely inspections, while ordinary inspection points have relatively lower timeliness requirements. To effectively differentiate task priorities in path planning and task allocation, this embodiment introduces a timeliness optimization objective based on binary priority, and quantifies the timeliness value of task completion through an exponentially decaying utility function, see formula (8):

[0052] in, This represents the set of high-priority inspection tasks assigned to the k-th UAV. This represents a set of ordinary priority inspection tasks; and λ and μ are the completion times for the corresponding tasks performed by drone k; λ and μ are decay coefficients used to control the time sensitivity of different task types; γ is the weighting parameter for ordinary tasks, used to balance the contribution of high-priority tasks and ordinary tasks to the overall utility.

[0053] In S105, a hybrid path planning algorithm is used to solve the assigned task sequence. The hybrid path planning algorithm is based on a biomimetic optimization framework and integrates manifold gradient descent strategy and topological differential evolution strategy to generate the initial flight path.

[0054] According to some embodiments, the method of solving the assigned task sequence using a hybrid path planning algorithm includes: initializing an artificial hummingbird population (AHA algorithm, a biomimetic optimization algorithm based on hummingbird intelligent behavior), mapping the task access sequence to individual position vectors, and alternately executing a local optimization stage based on manifold gradient descent and a global exploration stage based on topological differential evolution during the iterative search process. The hybrid path planning algorithm updates the individual positions using the following formula (9):

[0055] in, This represents the position vector of the i-th individual after the (t+1)-th iteration. Let represent the current position vector of the i-th individual at the t-th iteration; b is the guiding factor, used to control the weight of the influence of topological neighborhood information on individual updates; D is the step size adjustment coefficient, used to balance global exploration and local development. Let be the topological neighborhood set of the current solution, that is, the set of indices of all neighboring individuals that have a topological connection with individual i; | | is a topological neighborhood The number of individuals in the middle; j∈ This indicates traversing the topological neighborhood. Each neighboring individual j in the process; Let i be the position vector of the neighboring individual j at the t-th iteration; t is the current iteration number, and i is the index number of the current individual.

[0056] This formula is integrated into the iterative framework of the standard AHA, introducing improved update rules. While maintaining the exploratory capability, it utilizes the neighborhood topology to guide the search direction, effectively improving the stability of the algorithm.

[0057] Specifically, the local optimization stage based on manifold gradient descent includes constructing a manifold constraint surface in the path solution space, calculating the gradient direction of the objective function with respect to the individual position, and updating the step size along the opposite direction of the gradient to improve the local search stability and convergence speed of the algorithm in complex path optimization. The global exploration stage based on topological differential evolution includes constructing a topological adjacency network between individuals, selecting differential mutation vectors based on the topological adjacency network, performing crossover and selection operations on the differential mutation vectors, and updating the individual position vectors.

[0058] For example, in the region search phase, the AHA algorithm develops paths through random perturbations. Since the perturbation direction is completely random, it cannot distinguish between high and low priority task regions, easily causing search direction deviations. Therefore, this embodiment proposes a priority manifold gradient optimization method to construct a task-driven Riemannian manifold space. The priority-sensitive metric tensor is defined using formula (10):

[0059] in, It is a 2×2 symmetric positive definite Riemannian metric tensor that defines the local anisotropic geometric structure; It is a rank-1 outer product matrix, defined along... The direction of stretching; As task priority weight, As a smoothing factor, to prevent A zero denominator in the time interval leads to numerical explosion. The metric tensor produces a curvature enhancement effect near high-priority task points, thereby guiding the search towards the critical task region.

[0060] The metric tensor defined by formula (10) The anisotropic geometry in the task space is characterized. The fitness gradient in equation (11) It is precisely based on The calculated Riemann gradient makes the search direction in The defined Riemannian manifold naturally slopes towards high-priority task regions, achieving a complete closed loop of metric structure-guided search:

[0061] The value of δ is set to 0.01 to prevent numerical overflow.

[0062] Furthermore, by solving the system of linear equations Having obtained the manifold gradient direction, the improved region search formula (12) is finally obtained as follows:

[0063] in, As a mixture coefficient, this embodiment retains the original random perturbation term to ensure the robustness of the algorithm, while introducing a normalized manifold gradient term to guide the search direction.

[0064] Furthermore, to improve planning efficiency and path quality, this embodiment proposes a hybrid path planning method. First, a greedy algorithm is used to quickly generate the initial paths for the UAVs. For a swarm of K UAVs, the path construction process for each UAV k follows the minimum path optimization objective, as shown in formula (13):

[0065] in, These are the coordinates of the next candidate mission point for drone k. It is the set of unassigned task points, and p is the two-dimensional coordinate variable of the candidate task points. These are the coordinates of the last point on the current path of drone k.

[0066] The initial path result is then input into the MG-TDAHA algorithm (Manifold Gradient Topological Differential Artificial Hummingbird Algorithm) for further optimization. To adapt to the solution characteristics of the MG-TDAHA algorithm, the two objective functions of priority response and obstacle avoidance robustness are expressed as follows (14):

[0067] in, and Control the weight allocation of the two items separately; Let be the center coordinates of obstacle / no-fly zone j, the j-th obstacle out of M obstacles; Let these be the candidate coordinates that drone k passes through on its path when it visits mission point i. The indicator function is used for hard obstacle avoidance determination. It is the basic safety distance threshold. It is the collision radius of drone k.

[0068] This embodiment utilizes a greedy strategy to quickly reduce initial computational complexity, and then improves path quality and safety through MG-TDAHA, thereby ensuring that the UAV swarm can maintain efficient and stable execution capabilities in an environment where mission and obstacle avoidance requirements change randomly.

[0069] In S107, a dynamic adjustment mechanism for safe distance is introduced to perform robust obstacle avoidance correction on the initial flight path under random obstacle environment, and output the results of multi-UAV cooperative task allocation and path planning.

[0070] According to some embodiments, an elastic safety distance mechanism is proposed to address the obstacle avoidance requirements of UAV swarms in random obstacle environments. This mechanism senses the changes in the number and position of obstacles in the environment in real time and dynamically calculates the minimum safety distance threshold based on the obstacle distribution density and the flight status of the UAVs. The safety distance threshold formula (15) is as follows:

[0071] in, As a baseline safety distance, Indicates the first The number of obstacles within the perception range of each drone. To effectively perceive area, The current velocity vector, parameters and These are the environmental density sensitivity and the velocity compensation coefficient, respectively.

[0072] The formula in this embodiment is used for automatic replanning after the algorithm path planning fails. When a planned path node intrudes into the safety interval threshold range, it triggers local path replanning or node offset correction.

[0073] Figure 2 This is a schematic diagram of the hierarchical data flow of the hierarchical dynamic partitioning and hybrid path planning method of the present invention.

[0074] See Figure 2This invention employs a three-tiered progressive architecture to achieve collaborative task allocation and path planning for multiple UAVs. According to an example embodiment, firstly, at the global task partitioning layer, the global hash of task points and obstacle distribution is input into a hierarchical clustering model, dynamically dividing the system into load-balanced sub-regions 1, 2, and 3 to reduce the complexity of the high-dimensional search space. Secondly, at the local scheduling and allocation layer, the task clusters in the sub-regions are mapped to individual nodes such as UAV 1 and UAV 2, and balanced allocation is achieved by combining priority weights and load constraints, generating a dedicated task list for each UAV. Thirdly, at the path planning layer, a hybrid path planning algorithm is initiated for the allocated task sequence, considering obstacles, ordinary priority task points, and high priority task points. Under the premise of meeting the dynamic safety interval threshold, the access order (such as the planned trajectories of nodes 1 to 5) is optimized, obstacles are avoided in real time, and a collision-free, low-latency collaborative flight path is output. The above three-tiered architecture has a clear data flow, combining hierarchical decoupling with feature coupling, effectively balancing the computational efficiency of large-scale task allocation with the robustness of planning in dynamic environments.

[0075] Furthermore, to verify the effectiveness and superiority of the HDP-HPPM method proposed in this invention, a series of simulation experiments were designed and compared with several classic algorithms. The experiments were run on a computer configured with an Intel Core i7-12700H CPU and 32GB RAM, using MATLAB R2021b as the software environment. The simulation scenario was a 1000m × 1000m task airspace, where the initial positions of task points, obstacles, and the UAV were randomly generated. To reduce the impact of randomness, each experiment was run independently 20 times, and the average result was taken. The task point size was set to {80, 500, 1000, 1500} to test the scalability of the algorithm in small to very large-scale task scenarios. Obstacle density was divided into three levels: low, medium, and high, with randomized positions and sizes to evaluate obstacle avoidance performance under different environmental complexities.

[0076] To comprehensively evaluate the overall performance of different path planning algorithms, this embodiment selects average flight distance, task completion time, obstacle avoidance rate, and response time as key indicators from four dimensions: path economy, task efficiency, safety, and timeliness.

[0077] According to some embodiments, the average flight distance reflects the system's performance in terms of path economy and energy consumption control. It is the average of the actual flight path lengths of all UAVs. The smaller the value, the shorter the planned path and the lower the energy consumption. The formula for the average flight distance is shown in (16):

[0078] in, The total number of drones, For the first Total flight distance of the drone This represents the number of mission points visited by the drone. Indicates the first position on its path The coordinates of each task point.

[0079] Task completion time refers to the time taken from the start of the task to the last UAV completing its assigned last task point. It is used to evaluate the overall efficiency and collaborative capability of a multi-UAV system in executing tasks. The smaller the task completion time value, the faster the task execution speed and the higher the system throughput. The formula for calculating the task completion time is shown in (17):

[0080] in, For the first The cruising speed of a drone is determined by the time taken by the drone that flies the longest distance.

[0081] Obstacle avoidance rate is a key indicator for evaluating the safety and robustness of an algorithm. It is calculated by counting the proportion of flight segments that do not collide with obstacles in the entire flight path. The higher the obstacle avoidance rate, the stronger the algorithm's obstacle avoidance ability in complex environments, and the higher the safety of the planned path. The calculation formula is as follows (18):

[0082] in, This represents the total number of flight segments in all drone flight paths that collide with obstacles. This represents the total number of flight segments along all drone flight paths.

[0083] Response time is used to measure the system's efficiency in planning high-priority tasks, and it characterizes the average computation time taken for all high-priority task points from the start of planning to the generation of feasible paths. The smaller the value, the faster the algorithm's planning response is on critical tasks. The formula for calculating response time is shown in (19):

[0084] in, The set of all high-priority task points. The planning completion time for high-priority task point j is the time from initiating the path / scheduling calculation for that task point to generating the planning result.

[0085] Furthermore, to verify the effectiveness of the proposed hierarchical partitioning framework in large-scale multi-UAV mission planning problems, this embodiment compares and analyzes the performance of hierarchical partitioning and non-hierarchical partitioning architectures in terms of computational efficiency and task completion rate. Experiments were conducted using different task sizes N=80, 500, 1000, and 1500 as examples, and were performed under the same experimental environment. The experimental results are as follows: Figure 3 , Figure 4 As shown in the figure, the blue solid line represents the hierarchical partitioning architecture, and the orange dashed line represents the non-hierarchical partitioning architecture.

[0086] According to the example embodiment, Figure 3 The graph shows a comparison of response times. From the perspective of computational response time, when the task size is small (N=80), the computation time of the two methods is similar, and the non-hierarchical partitioning architecture even has a certain advantage at low scales. However, as the task size increases, the computation time of the non-hierarchical partitioning architecture grows rapidly, exceeding 100 seconds when N=1500; while the growth trend of the hierarchical partitioning method is significantly moderated, requiring only about 6.8 seconds at the same scale. The shaded area in the graph indicates that when the number of tasks is large (greater than 500), the response time of the hierarchical partitioning architecture is significantly lower than that of the other architecture, showing a clear performance advantage. This demonstrates that the hierarchical partitioning architecture can effectively alleviate the computational bottleneck problem under large-scale tasks and has better scalability.

[0087] See Figure 4 Regarding task completion rate, the hierarchical partitioning architecture maintained a high level across different task sizes, consistently achieving a completion rate above 96%. In contrast, the non-hierarchical partitioning architecture experienced a significant performance decline as task size increased, with the completion rate gradually decreasing from nearly 100% at small scales to 58%. Particularly with large-scale tasks (N≥1000), the non-hierarchical architecture exhibited significant performance degradation, failing to guarantee the effective execution of most tasks. The shaded area in the figure represents the situation where, with a large number of tasks (greater than 500), the hierarchical partitioning architecture maintained a high completion rate (approximately above 95%), while the non-hierarchical partitioning architecture experienced a sharp drop in completion rate. In comparison, the hierarchical partitioning framework remained stable under the same conditions, fully demonstrating its robustness and reliability in complex task environments.

[0088] To further verify the role of different improvement modules in the overall algorithm performance, this embodiment designed an ablation experiment within a unified hierarchical partitioning framework. Specifically, four algorithm variants were constructed for comparative analysis: the baseline variant DGAHA, which does not introduce any additional modules and is used to provide performance comparison; DGAHA-TD, which introduces a topological difference (TD) module into the baseline algorithm to enhance global search capabilities; DGAHA-MG, which introduces a manifold gradient (MG) module into the baseline algorithm to improve local convergence accuracy; and DGAHA-TD-MG, which combines TD and MG modules to achieve synergistic optimization of global exploration and local development. By comparing the performance of different variants, the specific contributions of each module to partitioning and path planning effects can be clearly revealed. Experimental results confirm that introducing TD or MG alone can bring certain improvements to global exploration or local development, while introducing both simultaneously (i.e., MG-TDAHA) can achieve a balance between global exploration and local development, significantly outperforming the other three schemes, thus proving the technical value of the alternating execution mechanism described in this invention. The global partitioning overview diagram and the local area path planning and obstacle avoidance diagram of this invention are shown in [reference needed]. Figure 5 and Figure 6 .

[0089] like Figure 5 As shown, a hierarchical algorithm divides the original environment, which contains random obstacles and task points, into multiple non-overlapping sub-regions (Region 1 to Region 10), and marks high-priority task points (pentagrams) and obstacles (circular regions) to achieve global load balancing and environmental awareness. Figure 6 As shown, taking Region 2 as an example, this invention generates a smooth drone flight path (arrow line) in a local area and successfully avoids circular obstacles in the area, verifying the algorithm's local obstacle avoidance capability and path connectivity in complex environments.

[0090] Furthermore, to evaluate the performance of the proposed algorithm under different task scales, this embodiment will fix the number of obstacles Obs=9 and set the number of task points N=80, 500, 1000, 1500 for experiments.

[0091] Example 1: Path planning verification in a small-scale task scenario, N=80.

[0092] In Example 1, a small-scale drone cooperative mission environment is simulated. The total number of mission points N=80, the number of obstacles Obs=9, and the number of drones is 4.

[0093] Using the method described in this invention for region division and path planning, experimental results show that: the task region division is reasonable, the distribution of UAVs in each region is balanced, and high-priority task points can be covered first; the generated flight path is continuous and smooth with no risk of collision; the algorithm can quickly complete task allocation, and the computational pressure of region division and path planning is small, which verifies that this invention has good task coverage and priority response capabilities in basic task execution scenarios.

[0094] Example 2: Path planning verification in a medium-scale task scenario, N=500.

[0095] In Example 2, the task point density was increased to verify the processing pressure of the algorithm. The total number of task points N=500 and the number of obstacles Obs=9.

[0096] With this task scale, the density of task points increases, leading to greater complexity in region division and path planning. The experimental data obtained using the method described in this invention are shown in Table 1 below: Table 1. Planning Results (N=500) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region1 28 9.237866765 100 0.020508269 14.21719531 Region2 26 8.091388596 100 0.013561373 12.96617648 Region3 16 8.819366522 100 0.016022655 16.80987599 Region4 23 11.25031513 100 0.029094062 16.80098784 Region5 28 8.215124554 100 0.014761715 17.35651079 Region6 34 8.176237391 100 0.018707719 16.97606755 Region7 17 9.06906081 100 0.031529764 14.23128558 Region8 15 9.297993622 100 0.027923696 13.78688809 Region9 16 8.637085621 100 0.015681183 13.26002004 Region10 7 12.88698413 100 0.030047467 12.26818745 When N=500, the system was divided into 10 task areas, with the following drone numbers allocated to each area: 28, 26, 16, 23, 28, 34, 17, 15, 16, and 7. The drone distribution was relatively even, with an average path length ranging from approximately 8 to 13 km. Obstacle avoidance success rate remained at 100%, emergency response time was less than 0.03 seconds, and task completion time was between 12 and 17 minutes. Experimental results demonstrate that under medium-scale tasks, this invention still ensures reasonable area partitioning, effective path optimization, and rapid response to high-priority tasks.

[0097] Example 3: Path planning verification in a large-scale task scenario, N=1000.

[0098] In Example 3, the task complexity is further increased to verify the scalability of the algorithm. The total number of task points N=1000 and the number of obstacles Obs=9. The planning is performed using the method described in this invention, and the experimental data is shown in Table 2 below: Table 2 Planning Results (N=1000) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region1 17 8.687323147 100 0.016562412 11.85851956 Region2 23 9.068760261 98.91304348 0.019398174 14.16364816 Region3 15 10.32397502 100 0.010541283 12.81701504 Region4 17 8.932407574 100 0.013328571 11.43477484 Region5 22 8.838101455 100 0.020206369 13.6036073 Region6 25 8.647787218 100 0.021394679 15.02615371 Region7 18 7.51629386 100 0.005183308 10.46232494 Region8 15 9.390575771 100 0.011294281 11.66180276 Region9 19 7.909850632 100 0.007625293 13.43010779 Region10 16 8.578985045 98.95833333 0.020827223 11.15938914 Region11 16 9.026481292 100 0.027548236 12.70284525 In a large-scale task scenario with N=1000, the number of regions increased to 11, and the distribution of drones across regions remained relatively balanced. The average path length in each region ranged from approximately 7.51 to 10.33 km, the obstacle avoidance success rate remained above 99%, the emergency response time was kept below 0.03 seconds, and the task completion time remained within 15 minutes. Experimental results show that the algorithm maintained good regional balance and path optimization performance under large-scale tasks, ensuring both task coverage and priority response capabilities.

[0099] Example 4: Path planning verification in a large-scale task scenario, N=1500.

[0100] In this embodiment, an extremely high-density task environment is simulated. The total number of task points N = 1500, and the number of obstacles Obs = 9. The planning is performed using the method described in this invention, and the experimental data are shown in Table 3 below: Table 3 Planning Results (N=1500) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region1 23 7.167483109 98.55072464 0.010234112 8.710468885 Region2 18 7.346707017 100 0.0141486 9.186947337 Region3 26 6.92142355 100 0.008684508 7.738950906 Region4 19 7.430989346 100 0.011686993 12.38988002 Region5 27 7.090479756 100 0.016866421 9.615273273 Region6 22 7.626491351 100 0.014470407 9.130330411 Region7 20 7.341267548 100 0.017559034 16.59184159 Region8 19 6.631038351 100 0.012832537 10.0671614 Region9 22 6.734581498 100 0.014978162 15.93127908 Region10 21 7.031874206 100 0.013267184 8.270895598 Region11 32 7.066414059 100 0.01727718 13.46634932 Region12 27 8.00939989 100 0.017789941 17.06446841 Region13 29 6.784763833 100 0.014535348 10.79004095 In ultra-large-scale mission scenarios, the number of regions further increased, divided into 13 mission regions. The distribution of drones remained relatively balanced, with the average path length in each region ranging from 6.63 to 8.01 km. The obstacle avoidance success rate remained above 98%, the emergency response time was controlled between 0.01 and 0.03 seconds, and the mission completion time was within 17.1 minutes. Experimental results show that despite the significant increase in mission scale, the algorithm can still reasonably divide the mission regions, achieving a balance between global coverage and local path optimization, ensuring a relatively balanced number of drones in each region, and maintaining timely response to high-priority tasks.

[0101] In summary, as the task scale increases, the algorithm demonstrates good scalability and robustness in region partitioning and path planning. It can respond quickly to small-scale tasks, prioritizing high-priority task points; under medium to large-scale tasks, the allocation of drones in each region is reasonable, path optimization performance is stable, and task completion time shows a controllable growth trend with increasing task volume, fully demonstrating the applicability and effectiveness of the algorithm under multi-task, multi-region, and multi-constraint conditions.

[0102] Furthermore, an analysis of the impact of obstacle density on the performance of the present invention was conducted.

[0103] According to the example embodiment, the number of task points is fixed at N=1000, and obstacles are randomly arranged in the same planned airspace. For each obstacle configuration, the number of UAVs, average path length, obstacle avoidance success rate, emergency response time, and task completion time in each partition are recorded. To measure the level of obstacle density, we adopt the definition of area coverage, which is the sum of the areas occupied by obstacles in the planned airspace divided by the total planned area area as the obstacle density measure. The specific formula is shown in (20):

[0104] in, For the first The projected area of ​​each obstacle This represents the total mission airspace area. This formula can be used to quantify the extent to which obstacles occupy feasible space and to compare the complexity of different scenarios. For ease of interpretation of experimental results, this embodiment provides three density levels based on coverage: Low density: <0.10; Medium density: 0.10≤ ≤0.20; High density: >0.20.

[0105] In this experiment, the number of fixed task points N=1000 and Obs=4, meaning that 4 obstacles are randomly placed in the same planned airspace, corresponding to the obstacle area coverage rate. The value is approximately 0.05, which falls under the category of low-density scenarios. The experiment aims to examine the impact of the proposed method on partition balance, path continuity, and obstacle avoidance behavior under conditions of sparse obstacles.

[0106] Example 5: Path planning verification in low-density obstacle scenarios.

[0107] In this embodiment, the total number of task points is set to N=1000, and four obstacles are randomly placed within the same planned airspace, with an obstacle area coverage rate ρobs≈0.05. The planning was performed using the method described in this invention, and the experimental data are shown in Table 4 below: Table 4 Planning Results (Obs=4) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region1 20 8.595194888 100 0.011231662 11.4653592 Region2 19 11.14809428 100 0.020481575 15.10801976 Region3 16 9.593093036 100 0.0236112 11.2255937 Region4 9 8.285120497 100 0.012113254 10.36033289 Region5 18 8.312302834 100 0.014055582 10.72463823 Region6 17 9.93522011 100 0.023366418 16.14641264 Region7 23 9.941237326 100 0.014060966 13.8153515 Region8 15 9.231663051 100 0.021208935 12.25552512 Region9 22 7.924589063 100 0.01798699 12.85383229 Region10 16 9.872470577 100 0.019398929 11.86879091 Region11 13 9.72269711 100 0.020281432 15.35025659 Region12 20 8.659946384 100 0.022123656 10.01369558 Region13 16 9.922189646 100 0.016677827 15.05437225 The data in the table shows that the mission area was effectively divided into 13 sub-regions, with the number of UAVs in each sub-region ranging from 9 to 23, demonstrating good load balancing. The average flight distance in each region ranged from 7.9 km to 11.15 km; the obstacle avoidance success rate was 100%. The emergency response time was controlled between 0.011 s and 0.023 s; the mission completion time was controlled within 16 minutes. Experimental results show that in low-density obstacle environments, this invention has strong initial path generation capabilities, capable of generating feasible paths in one go, reducing the real-time recalculation overhead caused by obstacle avoidance, and maintaining the continuity and efficiency of mission execution.

[0108] Example 6: Path planning verification in a medium-density obstacle scenario.

[0109] In this embodiment, the total number of task points is set to N=1000, and 7 obstacles are randomly placed, with an obstacle area coverage rate ρobs≈0.12. The planning was performed using the method described in this invention, and the experimental data are shown in Table 5 below: Table 5 Planning Results (Obs=7) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region 1 13 11.00371687 100 0.025256197 19.0482635 Region 2 26 8.891122323 100 0.027453975 14.145164 Region 3 18 8.330194228 100 0.017064026 10.57607616 Region 4 14 9.576532731 100 0.024851636 14.99792921 Region 5 13 10.51385194 100 0.02282575 14.41831281 Region 6 15 7.835835536 100 0.007008309 14.31945276 Region 7 14 9.208594569 100 0.019522677 13.19049089 Region 8 24 7.820808796 100 0.016954999 9.840529944 Region 9 18 8.49472566 98.89 0.017962297 10.95324629 Region 10 15 9.188542535 100 0.015291151 12.63068258 Region 11 17 10.14520834 100 0.019308647 12.67835593 Region 12 20 8.457586116 100 0.017844119 12.23418793 The mission area was effectively divided into 12 sub-regions, with the number of UAVs in each region ranging from 13 to 26. The average flight distance in each region was within 11 km; the obstacle avoidance success rate remained above 98.89%. Emergency response time was controlled between 0.007 s and 0.027 s; mission completion time was controlled between 9.84 min and 19.05 min. Experimental results show that when medium-density environments cause narrowing of passageways or the appearance of local bottlenecks, this invention can avoid obstacles through effective local adjustments (such as detours), reducing the need for global replanning and thus maintaining efficient and stable execution performance.

[0110] Example 7: Path planning verification in a high-density obstacle scenario.

[0111] In this embodiment, the total number of task points N=1000, and the number of obstacles is increased to 16, corresponding to an obstacle area coverage rate ρobs≈0.28. The planning was performed using the method described in this invention, and the experimental data are shown in Table 6 below: Table 6 Planning Results (Obs=16) area Number of drones Flight distance (km) Obstacle avoidance rate (%) Emergency response time (s) Task completion time (min) Region 1 9 7.620177311 100 0.016580294 12.26678642 Region 2 14 8.914259418 100 0.017801896 14.54039273 Region 3 21 7.085837749 100 0.012227444 8.533512235 Region 4 14 7.233836836 100 0.004107811 10.00149372 Region 5 6 6.179549503 100 0.022063011 8.815478779 Region 6 14 7.08302928 98.23 0.012768186 8.330596824 Region 7 11 9.152664714 100 0.020655176 12.73138504 Region 8 19 7.567277977 100 0.020601032 10.92045783 Region 9 13 7.853128443 97.16 0.020085466 12.22117903 Region 10 10 8.638248202 100 0.026895888 11.93937159 Region 11 16 7.314160419 99.02 0.01489711 10.26235918 Region 12 14 6.427501066 100 0.013428293 10.56993928 Region 13 12 7.406068877 100 0.011398459 8.521335629 Region 14 18 7.534399416 100 0.008508342 9.429308796 Region 15 16 7.121783514 100 0.01475548 8.717961224 Unassigned Area 1 8.242588494 100 0 7.005495059 The task area was divided into 15 sub-regions, with the number of drones in each sub-region ranging from 6 to 21, resulting in good load balancing. Additionally, there was an unassigned area containing one task point, executed solely by a single drone to ensure complete task coverage across the entire area. The average flight distance in each region ranged from 6.2 km to 9.2 km; the obstacle avoidance success rate remained above 97%; and the number of path replanning attempts was 2. Emergency response times were consistently between 0.004 s and 0.026 s (milliseconds); and task completion times were controlled between 7 min and 15 min. Experimental results demonstrate that even in high-density obstacle environments, despite reduced feasible paths and more frequent path detours, this invention maintains a high obstacle avoidance success rate and real-time scheduling performance. When necessary, the algorithm can trigger a global replanning mechanism to ensure complete task success, proving its adaptability and robustness in extremely complex environments.

[0112] Based on some embodiments, the experimental data for different obstacle densities are summarized in Table 7 below: Table 7 Planning results under different obstacle densities Number of obstacles Coverage ρobs Obstacle avoidance success rate (%) Number of path replanning 2 0.02 100 0 3 0.03 100 0 4 0.05 100 0 7 0.12 99.95 0 8 0.15 99.88 1 9 0.18 99.75 1 13 0.22 99.21 1 14 0.25 98.82 2 16 0.28 98.59 2 As shown in Table 7 (Performance Summary under Different Obstacle Densities), the number of path replanning attempts increases with the increase in obstacle coverage. In low-density (Obs=4) and medium-density (Obs=7) scenarios, the algorithm mainly avoids obstacles through local adjustments, with zero replanning attempts. However, in high-density (Obs=16) scenarios, due to the reduction in feasible paths, the algorithm triggers two global replanning attempts to ensure task success. This indicates that the present invention can adaptively adjust the planning strategy according to environmental complexity, achieving a balance between efficiency and safety.

[0113] This invention changes the rigid mode of traditional fixed thresholds. In low-density obstacle scenarios, the algorithm can avoid obstacles in one go using the initial path, with zero replanning attempts. In high-density (Obs=16) scenarios, it can dynamically calculate the safety interval based on the obstacle density, triggering replanning only when necessary (two attempts). This adaptive mechanism ensures an obstacle avoidance success rate of over 98% while minimizing the time loss caused by frequent replanning.

[0114] In summary, the HDP-HPPM method of this invention achieves significant improvements in task allocation balance, path economy, priority responsiveness, and obstacle avoidance robustness, providing an efficient and reliable solution for collaborative task execution by multiple UAVs in large-scale, complex environments. Future research will further explore the application potential of this method in dynamic environments, energy-constrained environments, and heterogeneous UAV swarms.

[0115] Figure 7A block diagram of a computing device according to an exemplary embodiment of the present invention is shown.

[0116] like Figure 7 As shown, the computing device 30 includes a processor 12 and a memory 14. The computing device 30 may also include a bus 22, a network interface card 16, and an I / O interface 18. The processor 12, memory 14, network interface card 16, and I / O interface 18 can communicate with each other via the bus 22. The computing device of this invention can be applied to monitoring equipment such as drones.

[0117] Processor 12 may include one or more general-purpose CPUs (Central Processing Units), microprocessors, or application-specific integrated circuits, for executing relevant program instructions. According to some embodiments, computing device 30 may also include a high-performance display adapter (GPU) 20 for accelerating processor 12.

[0118] Memory 14 may include a machine system readable medium in the form of volatile memory, such as random access memory (RAM), read-only memory (ROM), and / or cache memory. Memory 14 is used to store one or more programs containing instructions, as well as data. Processor 12 may read the instructions stored in memory 14 to perform the methods described above according to embodiments of the present invention.

[0119] The computing device 30 can also communicate with one or more networks via the network interface card 16. The network interface card 16 can be a DPU smart network card.

[0120] Bus 22 can include address bus, data bus, control bus, etc. Bus 22 provides a path for exchanging information between components.

[0121] If the technical solution of this invention involves personal information processing, the relevant products have established a sound user authorization mechanism: before collecting, using, or sharing personal information, the obligation to inform is fulfilled in accordance with the law, and the individual's voluntary and explicit consent is obtained; if sensitive personal information is involved, the user's separate and explicit consent is further obtained. Specific measures include, but are not limited to: setting up prominent prompts in the information collection area, or clearly displaying the processing rules through electronic interfaces such as pop-ups, checkboxes, and active submissions. The processing rules include, but are not limited to, the processor, purpose, method, and information type, ensuring that the user voluntarily authorizes the process based on informed consent. All personal information processing activities strictly comply with national laws and regulations, especially the relevant provisions of the "Personal Information Protection Law of the People's Republic of China," to effectively protect the legitimate rights and interests of personal information subjects.

[0122] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0123] 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, and improvements 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 collaborative task allocation method based on hierarchical dynamic partitioning and hybrid path planning, characterized in that, include: A hierarchical dynamic partitioning model is constructed to perform a global initial partitioning of the task space, and local adaptive clustering and region boundary adjustment are performed based on dynamic environmental changes to generate task subsets for each UAV. A multi-objective collaborative optimization model is established with the goal of minimizing the total flight distance of the cluster and the mission completion time, and the task subset is evenly allocated based on the task priority. A hybrid path planning algorithm is used to solve the assigned task sequence. The hybrid path planning algorithm is based on a biomimetic optimization framework and integrates manifold gradient descent strategy and topological differential evolution strategy to generate the initial flight path. A dynamic adjustment mechanism for safe distance is introduced to perform robust obstacle avoidance correction on the initial flight path in a random obstacle environment, and output the results of multi-UAV collaborative task allocation and path planning.

2. The method according to claim 1, characterized in that, The construction of the hierarchical dynamic partitioning model includes: The K-means algorithm is used to perform global coarse-grained partitioning of the task points to determine the initial number of partitions K. The formula for calculating the value of K is as follows: in, The total number of task points. Density adjustment factor; Within the initial partition, an adaptive DBSCAN algorithm is used for local fine-grained clustering, with a neighborhood radius of... Based on partition characteristics, the following is adaptively determined: in, Indicates the equivalent radius of the partition. The standard deviation of the distance between neighboring points. This is a smoothing factor.

3. The method according to claim 1, characterized in that, The establishment of the multi-objective collaborative optimization model includes: The cruise distance objective function is constructed by summing the Euclidean distances between consecutive task points on the assigned paths of each UAV, and a load balancing term is introduced to constrain the variance of the number of task points of each UAV. The priority of cruise efficiency and load balancing is coordinated by weighting coefficients to achieve balanced allocation of task points.

4. The method according to claim 1, characterized in that, The process of solving the assigned task sequence using a hybrid path planning algorithm includes: Initialize an artificial hummingbird population and map the task access sequence to individual location vectors; During the iterative search process, a local optimization phase based on manifold gradient descent and a global exploration phase based on topological differential evolution are executed alternately. The hybrid path planning algorithm updates the individual position using the following formula: in, This represents the position vector of the i-th individual after the (t+1)-th iteration. Let represent the current position vector of the i-th individual at the t-th iteration; b is the guiding factor, used to control the weight of the influence of topological neighborhood information on individual updates; D is the step size adjustment coefficient, used to balance global exploration and local development. Let be the topological neighborhood set of the current solution, that is, the set of indices of all neighboring individuals that have a topological connection with individual i; | | is a topological neighborhood The number of individuals in the middle; j∈ This indicates traversing the topological neighborhood. Each neighboring individual j in the process; Let i be the position vector of the neighboring individual j at the t-th iteration; t is the current iteration number, and i is the index number of the current individual.

5. The method according to claim 4, characterized in that, The local optimization stage based on manifold gradient descent includes: constructing a manifold constraint surface in the path solution space, calculating the gradient direction of the objective function with respect to individual positions, updating the step size along the opposite direction of the gradient, and generating a manifold gradient term to improve the local search stability and convergence speed of the algorithm in complex path optimization.

6. The method according to claim 4, characterized in that, The global exploration phase based on topological differential evolution includes: constructing a topological adjacency network among individuals, selecting differential mutation vectors based on the topological adjacency network, performing crossover and selection operations on the differential mutation vectors, and generating topological differential evolution terms.

7. The method according to claim 1, characterized in that, The balanced allocation of the task subset based on task priority includes: Different tasks are assigned priority weight coefficients, and a high-priority task response delay penalty term is set in the optimization model. By dynamically adjusting the weight of the task access sequence, the high-priority task is guaranteed to respond first in the path planning.

8. The method according to claim 1, characterized in that, The dynamic adjustment mechanism for the safe distance includes: The system senses the number and location changes of obstacles in the environment in real time, and dynamically calculates the minimum safe distance threshold based on the obstacle distribution density and the drone's flight status. The formula for the safe distance threshold is: in, As a baseline safety distance, Indicates the first The number of obstacles within the perception range of each drone. To effectively perceive area, The current velocity vector, parameters and These are the environmental density sensitivity and the velocity compensation coefficient, respectively. When a planned path node intrudes into the safety interval threshold range, a local path replanning or node offset correction is triggered.

9. A computing device, characterized in that, include: processor; as well as A memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images