Multi-machine cooperative path planning method, system and device in city and storage medium

Abstract

The invention relates to the technical field of unmanned aerial vehicle path planning, in particular to a multi-aerial vehicle cooperative path planning method, system and device in a city and a storage medium. Comprising an information acquisition module, a path generation module, a clustering updating module and an optimal judgment module. According to the method, the initial position, the target position and the obstacle information of each unmanned aerial vehicle are acquired, an urban low-altitude airspace task scene is defined, the MP-GWO algorithm is utilized to generate a multi-aerial-vehicle cooperative initial path, differential penalty coefficients are set for different types of no-fly zones, a K-means clustering mechanism is embedded into a GWO main cycle, and the multi-aerial-vehicle cooperative initial path is obtained. According to the method, parallel sub-populations are divided according to a multi-target fitness vector, each sub-population independently executes an alpha, beta and delta guide updating strategy, a triple termination judgment mechanism integrating fixed iteration times, a convergence threshold value and elite stagnation is adopted, a current optimal planning path is output, efficient optimization of the path in a complex urban environment is achieved, and the path optimization efficiency is improved. The method is suitable for urban low-altitude airspace multi-unmanned aerial vehicle cooperative tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) path planning, and in particular to a method, system, device and storage medium for urban multi-machine collaborative path planning. Background Art

[0002] While single-drone technology has matured, limitations in sensing range, anti-interference capabilities, and computing power make it difficult to adapt to today's increasingly demanding requirements. Against this backdrop, multi-drone systems have emerged. Their advantages in field of view, payload capacity, and flight time have made them a research hotspot.

[0003] Path planning is a complex optimization problem. Traditional path planning algorithms primarily include graph search and sampling-based methods, such as Rapidly Expanding Random Trees (RRT), Artificial Potential Fields (APF), the A* algorithm, and the Dijkstra algorithm. However, these traditional methods struggle to solve path planning problems in complex environments. Since its introduction in 2014, the Gray Wolf Optimizer (GWO), a nature-inspired optimization algorithm that simulates the hunting behavior of gray wolf packs, has been widely used to solve various optimization problems, particularly in the field of collaborative optimization.

[0004] However, the GWO algorithm still has room for improvement, mainly manifested in problems such as premature convergence, long optimization time, and difficulty escaping local optimal solutions in complex environments. When applying the GWO algorithm to the collaborative path optimization of multiple UAVs, especially when avoiding conflicts between UAVs and static buildings, although the goal is to optimize the path as much as possible to approach the optimal or suboptimal solution, it still faces problems such as poor search path and long search time. These challenges mainly stem from the algorithm's insufficient local search capability in complex environments, which makes it difficult to effectively explore the global solution. In order to enable multiple UAVs to coordinate flight missions in complex environments, while taking into account global search capabilities and local development efficiency, we propose a multi-universal collaborative path planning method, system, equipment and storage medium within the city. Summary of the Invention

[0005] This invention aims to address the difficulty of traditional path planning algorithms in complex environments. By applying the GWO algorithm to the coordinated path optimization of multiple UAVs, particularly when avoiding conflicts between UAVs and static structures, while the goal is to optimize the path as close to the optimal or suboptimal solution as possible, it still faces challenges such as poor path search and excessive search time. These challenges primarily stem from the algorithm's limited local search capabilities in complex environments, making it difficult to effectively explore global solutions. To ensure that multiple UAVs can coordinate flight missions in complex environments, it is necessary to balance global search capabilities with local development efficiency.

[0006] To achieve the above objectives, one of the objectives is to provide a method for multi-machine collaborative path planning in an urban area, comprising the following steps:

[0007] S1. Obtain the initial position, target position, and obstacle information of each UAV, and use the grid method to model and define the urban low-altitude airspace mission scenario;

[0008] S2. Generate the initial multi-aircraft coordination path using the MP-GWO algorithm, construct a multi-objective fitness vector that includes energy, altitude, no-fly zone threat, spatial collision, and time coordination costs, and set differentiated penalty coefficients for different types of no-fly zones.

[0009] S3, embed K-means clustering into the optimization main loop, divide the subpopulation according to the fitness vector, and each subpopulation independently performs the nonlinear position update guided by α, β, and δ. After correcting the path feasibility, merge and iterate;

[0010] S4, adopt the triple judgment conditions of fixed number of iterations, convergence threshold and elite stagnation, and output the current optimal planning path when any condition is met.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] 1. This urban multi-drone collaborative path planning method, system, equipment, and storage medium utilizes an information acquisition module to obtain each drone's initial position, target position, and obstacle information. The path generation module utilizes the MP-GWO algorithm to generate the initial multi-drone collaborative path. A multi-objective fitness vector is constructed, encompassing energy, altitude, no-fly zone threat, spatial collision, and temporal coordination costs. Differentiated penalty coefficients are set for different types of no-fly zones. K-means clustering is embedded in the GWO main loop to form an MP-GWO multi-population parallel optimization framework. Fitness vector clustering enables dynamic subpopulation partitioning, enhancing global search capabilities and structural stability. Differentiated penalty coefficients are set for different types of no-fly zones, supporting dynamic adjustment of the temporal dimension, and improving the path planning's adaptability to complex urban environments.

[0013] 2. The clustering update module embeds K-means clustering into the optimization main loop and divides the subpopulations according to the fitness vector. Each subpopulation independently performs nonlinear position updates guided by α, β, and δ. After correcting the path feasibility, the iterations are merged. The optimal judgment module adopts the triple judgment conditions of fixed number of iterations, convergence threshold, and elite stagnation. When any condition is met, the current optimal planning path is output. The nonlinear convergence factor and dynamic weighting strategy are introduced to balance the algorithm's global exploration and local development capabilities, improve the limitations of the traditional GWO linear decrement, integrate fixed iterations, convergence threshold, and elite stagnation judgment, balance computational efficiency and optimization accuracy, and avoid premature convergence and infinite iterations.

[0014] As a further improvement of this technical solution, the specific steps of S2 generating the multi-machine collaborative initial path are as follows:

[0015] S2.1.1 Population Initialization: Given a flight space and mission objectives, generate several search individuals. Each search individual represents a complete set of UAV path planning solutions. Each individual is encoded as a combined vector of multiple UAV tracks, representing the sequence of track points of all UAVs from the starting point to the end point.

[0016] S2.1.2, Fitness evaluation: Perform path quality evaluation on each individual and construct a multi-objective fitness vector. Each individual corresponds to a set of multi-dimensional fitness features.

[0017] As a further improvement of this technical solution, the multi-objective fitness vector constructed in S2.1.2 includes the following cost indicators:

[0018] Energy consumption cost: calculated based on the total length of the path, and the length of each path segment is obtained by accumulating the Euclidean distance between waypoints;

[0019] Flight altitude cost: When the drone's flight altitude exceeds the preset allowable range, a height penalty cost is incurred;

[0020] No-fly zone threat cost: When a path crosses or approaches a no-fly zone, a penalty is generated based on the type of no-fly zone and the degree of proximity. Different types of no-fly zones, including but not limited to vegetation areas, built-up areas, large-scale activity areas, and government-controlled areas, are assigned different threat coefficients.

[0021] Spatial collision cost: Compare the actual flight distance between any two drones at the trajectory point with the safe distance. If the safe distance is exceeded, the cumulative number of collisions will result in a penalty.

[0022] Time coordination cost: Evaluate the time window range for each drone to reach the target point. If the actual arrival time of the drone exceeds the reasonable time range, the delay penalty cost will be increased.

[0023] The beneficial effect of adopting the above further improvements is that interrelated or conflicting goals (such as shortest path, lowest energy consumption, and highest security) are converted into different dimensions of vectors, avoiding the one-sidedness of single-objective optimization and dynamically balancing the goal priorities through weight coefficients (such as multiplying different weights by different dimensions of the vector).

[0024] As a further improvement of this technical solution, S2.1.2 constructs a multi-objective fitness vector, introduces regional judgment logic into the fitness function, superimposes corresponding penalty values ​​based on whether the path passes through the designated area, and supports dynamic adjustment of no-fly zone restriction rules based on time.

[0025] As a further improvement of this technical solution, the specific steps of using the K-means clustering algorithm in S3 to divide each group of paths are as follows:

[0026] S3.1.1. K-means clustering subpopulation division: Using the fitness vector as the clustering input feature, the K-means clustering algorithm is used to divide the search individuals into multiple subpopulations. Individuals in each subpopulation have similar performance, which facilitates efficient collaborative optimization in the local search space. This clustering process is performed dynamically to adapt to changes in individual status.

[0027] S3.1.2. Selecting guide individuals based on fitness values: Select α, β, and δ as guide individuals. Within each subpopulation, α is the individual with the best fitness in the current subpopulation, β is the second best, and δ is the third best.

[0028] S3.1.3. Position Update: For each search individual, update its path combination vector within its subpopulation, calculate the distance between the individual and individuals α, β, and δ, and use the three guide individuals to calculate candidate positions based on the principle of the gray wolf optimization algorithm to guide the position update of the remaining search individuals.

[0029] S3.1.4, Boundary and feasibility correction: perform boundary detection and track feasibility correction on the updated individual position;

[0030] S3.1.5. Subpopulation merging and iterative updating: Merge all updated subpopulations into a new global population, perform K-means clustering again, and enter the next round of iteration.

[0031] As a further improvement of the present technical solution, when updating the position in S3.1.3, a nonlinear convergence factor is introduced, and parameter a is used as a control parameter to affect the value of the algorithm convergence factor A, thereby indirectly affecting the iterative update of the position of a single gray wolf.

[0032] The beneficial effect of adopting the above further improvements is that the dynamic weight can be adjusted in real time according to the fitness value of the current iteration, so that the influence of high-quality solutions accounts for a higher proportion. The dynamic weight can indirectly control the update step size. When the environment changes drastically (such as the appearance of new obstacles), the dynamic weight ratio is increased to make the position update more aggressive; when approaching the optimal solution, the dynamic weight is reduced, and the trajectory is smoothed by static averaging to prevent oscillation.

[0033] As a further improvement of this technical solution, when S4 outputs the optimal planned path, its judgment conditions include the following:

[0034] Maximum number of iterations: Set a fixed upper limit and terminate the calculation when the number of iterations is reached to ensure that the calculation time is controllable;

[0035] Convergence threshold judgment: If the improvement of the global optimal fitness for several consecutive generations is less than the threshold (such as 1e-6), the algorithm is considered to have converged and terminated early;

[0036] Member elite stagnation judgment: integrated with the KO algorithm, if the optimal solution has not been updated for more than N generations, it is considered to have fallen into a local optimum, which can trigger termination or reinitialization.

[0037] The beneficial effect of these improvements is that, for some scenarios (such as urban environments with rapidly changing dynamic obstacles), excessive pursuit of a theoretically optimal solution can cause path planning to lag behind environmental changes. The triple judgment criteria strikes a balance between optimization accuracy and real-time performance, outputting a "satisfactory solution" for the current environment rather than a "perfect solution."

[0038] A second object of the present invention is to provide a multi-machine collaborative path planning system within an city, comprising any one of the multi-machine collaborative path planning methods within an city described above, including an information collection module, a path generation module, a clustering update module, and an optimal determination module;

[0039] The information acquisition module obtains the initial position, target position and obstacle information of each drone, and uses the grid method to model and define the urban low-altitude airspace mission scenario;

[0040] The path generation module uses the MP-GWO algorithm to generate the initial path for multi-aircraft cooperation, constructs a multi-objective fitness vector that includes energy, altitude, no-fly zone threat, spatial collision and time cooperation costs, and sets differentiated penalty coefficients for different types of no-fly zones;

[0041] The cluster update module embeds K-means clustering into the optimization main loop, divides the subpopulation according to the fitness vector, and each subpopulation independently performs the nonlinear position update guided by α, β, and δ, and merges and iterates after correcting the path feasibility;

[0042] The optimal decision module adopts the triple decision conditions of fixed number of iterations, convergence threshold and elite stagnation, and outputs the current optimal planning path when any one of the conditions is met.

[0043] The third purpose is that the multi-machine collaborative path planning device in the city includes a memory and a processor, and the memory and processor are applied to any of the above-mentioned multi-machine collaborative path planning systems in the city when executed.

[0044] The fourth purpose is that the computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements any of the above-mentioned multi-machine collaborative path planning systems within the city.

[0045] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the overall flow chart of the present invention;

[0047] Figure 2 This is the S1 flow chart of the present invention;

[0048] Figure 3 This is the S2 flow chart of the present invention;

[0049] Figure 4 This is the S3 flow chart of the present invention;

[0050] Figure 5 It is a module flow chart of the present invention;

[0051] Figure 6 It is a flowchart of the method of the present invention.

[0052] The meaning of each number in the figure is:

[0053] 100, information collection module; 200, path generation module; 300, clustering update module; 400, optimal determination module. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] Currently, traditional path planning algorithms struggle to solve path planning problems in complex environments. When applying the GWO algorithm to collaborative path optimization for multiple drones, particularly when avoiding conflicts between drones and static structures, while the goal is to optimize the path as close to the optimal or suboptimal solution as possible, it still faces challenges such as poor search paths and excessive search times. These challenges stem primarily from the algorithm's limited local search capabilities in complex environments, making it difficult to effectively explore global solutions. To ensure that multiple drones can collaborate on missions in complex environments, it is necessary to balance global search capabilities with local development efficiency.

[0056] Therefore, the present invention proposes that the initial position, target position and obstacle information of each drone are obtained through an information acquisition module, the path generation module generates the initial path for multi-drone collaboration, and the GWO algorithm is used to randomly generate the initial path for each drone to the target position. Each group of paths represents a wolf. The clustering update module uses the K-means clustering algorithm to divide each group of paths to obtain K sub-populations with similar internal structures and consistent goals, and executes the update strategy. The optimal judgment module adopts the triple judgment conditions of fixed number of iterations, convergence threshold and elite stagnation, and outputs the current optimal planned path when any condition is met.

[0057] The details are as follows:

[0058] like Figure 1 As shown, one of the purposes of the present invention is to provide a method for multi-machine collaborative path planning in an urban area, comprising the following steps:

[0059] S1. Obtain the initial position, target position, and obstacle information of each UAV, and use the grid method to model and define the urban low-altitude airspace mission scenario;

[0060] S2. Generate the initial multi-aircraft coordination path using the MP-GWO algorithm, construct a multi-objective fitness vector that includes energy, altitude, no-fly zone threat, spatial collision, and time coordination costs, and set differentiated penalty coefficients for different types of no-fly zones.

[0061] S3, embed K-means clustering into the optimization main loop, divide the subpopulation according to the fitness vector, and each subpopulation independently performs the nonlinear position update guided by α, β, and δ. After correcting the path feasibility, merge and iterate;

[0062] S4, adopt the triple judgment conditions of fixed number of iterations, convergence threshold and elite stagnation, and output the current optimal planning path when any condition is met.

[0063] like Figure 2 As shown, the specific steps for S1 to obtain the initial position, target position and obstacle information of each drone are as follows:

[0064] S1.1.1. Use the grid method to model the overall environment, obstacles, and no-fly zones, and initialize the parameters;

[0065] S1.1.2. The mission scenario is that multiple drones fly in the low-altitude airspace of the city, crossing multiple static obstacles of different types to reach their respective target points.

[0066] There are four types of static obstacles (collectively referred to as no-fly zones below): vegetation, buildings, large-scale events, and government controls. Different types of no-fly zones are represented by different parameters.

[0067] Based on the urban low-altitude airspace scenario, define the modeling area (e.g., a three-dimensional space of 500m long × 500m wide × 100m high), clearly identify any obstacles on the boundary (e.g., fences at the city boundary or building exterior walls), and select a grid resolution (e.g., 10m × 10m × 10m). Balance computational efficiency and accuracy. A resolution that is too small (e.g., 1m) will increase the data volume, while a resolution that is too large (e.g., 50m) may overlook small obstacles.

[0068] Avoid excessive concentration of multiple drones' initial positions to reduce the risk of collisions during takeoff. For example, select multiple non-overlapping grids at the edge of the environment to ensure that the initial positions of adjacent drones are at least two grids apart. Determine the target point based on the drone mission type (such as logistics distribution or inspection). For example, if the target point for a delivery mission is the coordinates of a certain urban community (300m, 250m, 0m), the target points of multiple drones should not be too concentrated. For example, the projections of the target points of N drones on the horizontal plane should be distributed in different areas to reduce path intersection conflicts.

[0069] like Figure 3 As shown, the specific steps for S2 to generate the initial path for multi-machine collaboration are as follows:

[0070] S2.1.1 Population Initialization: Given a flight space and mission objectives, generate several search individuals. Each search individual represents a complete set of UAV path planning solutions. Each individual is encoded as a combined vector of multiple UAV tracks, representing the sequence of track points of all UAVs from the starting point to the end point.

[0071] S2.1.2, Fitness evaluation: Perform path quality evaluation on each individual and construct a multi-objective fitness vector. Each individual corresponds to a set of multi-dimensional fitness features.

[0072] The path of each drone consists of an ordered sequence of track point coordinates. The path sequences of all drones are spliced ​​into individual codes by number. For each drone, track points are randomly sampled in the free space from the starting point to the end point, ensuring that the spacing between adjacent points is reasonable (for example, greater than 2 times the grid resolution to avoid overcrowding of the path) and that all points do not cross obstacles or no-fly zones.

[0073] Each track point must be located in free space (grid value 0) and maintain a safe distance from obstacles and no-fly zones (e.g., greater than 1 grid). The paths of different drones must avoid collisions in time and space, meaning they cannot be in the same grid or adjacent grids at the same time (e.g., horizontal distance greater than 10m, height difference less than 5m).

[0074] When the initial population is generated, diversity is improved by increasing the random sampling ratio to avoid the limitation of the solution space caused by heuristic generation. When evaluating fitness, additional rewards are given to individuals with high diversity (such as adding a "diversity dimension") to prevent early convergence.

[0075] In order to better construct the multi-objective fitness vector, S2.1.2 constructs the multi-objective fitness vector including the following cost indicators:

[0076] Energy consumption cost: calculated based on the total length of the path, and the length of each path segment is obtained by accumulating the Euclidean distance between waypoints;

[0077] Flight altitude cost: When the drone's flight altitude exceeds the preset allowable range, a height penalty cost is incurred;

[0078] No-fly zone threat cost: When a path crosses or approaches a no-fly zone, a penalty is generated based on the type of no-fly zone and the degree of proximity. Different types of no-fly zones, including but not limited to vegetation areas, built-up areas, large-scale activity areas, and government-controlled areas, are assigned different threat coefficients.

[0079] Spatial collision cost: Compare the actual flight distance between any two drones at the trajectory point with the safe distance. If the safe distance is exceeded, the cumulative number of collisions will result in a penalty.

[0080] Time coordination cost: Evaluate the time window range for each drone to reach the target point. If the actual arrival time of the drone exceeds the reasonable time range, the delay penalty cost will be increased;

[0081] The cost formula of the multi-objective fitness vector is as follows:

[0082]

[0083] Where C is the total cost, n is the number of cost items, C Ei is the energy cost consumed by the i-th UAV, C Hi is the flight height cost of the i-th UAV, C Di is the cost of the i-th drone being threatened by the no-fly zone, C Si is the spatial collision cost of the i-th UAV, C Ti is the time coordination cost of the i-th UAV, w1 is the energy cost C consumed Ei The weight of w2 is the flight height cost C Hi The weight of w3 is the cost C of the no-fly zone threat. Di The weight of w4 is the spatial collision cost C Si The weight of w5 is the time collaboration cost C Ti The weight of .

[0084] Find a balance between the five types of costs: energy, altitude, no-fly zones, collisions, and time. For example, reducing energy costs may require low-altitude flight (increasing altitude costs), but low altitude may also approach obstacles (increasing collision costs). It is necessary to find the global optimum through weight adjustment and combine multi-objective optimization algorithms. Under the premise of meeting mission constraints (such as no-fly zones must not be entered, that is, the cost of drones being threatened by no-fly zones is enforced), iteratively search for the drone trajectory, altitude, and time coordination strategy that minimizes the total cost C, and ultimately achieve the multi-objective optimization of "safety, efficiency, and coordination", which not only ensures that drones do not enter no-fly zones and do not collide (safety), but also consume less energy and coordinate on time (efficiency), and flexibly adapt to different mission requirements through weights.

[0085] When constructing a multi-objective fitness vector, the above-mentioned cost indicators need to be considered comprehensively. Since different mission scenarios attach different importance to each cost indicator, it is necessary to set a corresponding weight coefficient for each cost indicator and obtain the final fitness value through weighted summation to comprehensively evaluate the pros and cons of the UAV path planning scheme. In different tasks, the weight coefficient should be flexibly adjusted according to actual needs. For example, logistics distribution tasks may pay more attention to energy consumption costs, while military reconnaissance missions may be more sensitive to the threat costs of no-fly zones.

[0086] In order to better set multi-level penalty strategies, S2.1.2 constructs a multi-objective fitness vector and introduces regional judgment logic into the fitness function. It superimposes the corresponding penalty value based on whether the path crosses the specified area and supports dynamic adjustment of no-fly zone restriction rules based on time.

[0087] Establish a mapping relationship between the no-fly zone type and the basic threat coefficient to form a standardized penalty matrix:

[0088] No-fly type Basic threat factor Typical scenarios Government-controlled areas 10 Government buildings, military bases Building area 8 High-rise buildings and bridges High-density activity area 6 Commercial Center vegetation area 4 Parks and forests School (daytime) 7 Primary and secondary schools, universities School (night) 3 Primary and secondary schools, universities Residential Community 5 Over residential areas

[0089] Map the no-fly zone boundary coordinates to a three-dimensional grid space, mark the no-fly zone core area (value 1) and the buffer zone (value 0.5), convert the track point coordinates to a grid index, and query the corresponding grid value:

[0090] A grid value of 1 means crossing the no-fly zone directly, and the penalty value is set to infinity (the path is not feasible);

[0091] The grid value is 0.5, entering the buffer zone, and the penalty is calculated based on the distance;

[0092] The grid value is 0, which is a safe area and no penalty.

[0093] Establish a three-dimensional mapping table of time-area-threat coefficient, read the system time or task setting time, event-driven rules are greater than periodic rules, and are greater than basic rules. If multiple rules overlap in the same area, the rule with the highest threat coefficient is selected.

[0094] like Figure 4 As shown, the specific steps of S3 using the K-means clustering algorithm to divide each group of paths are as follows:

[0095] S3.1.1. K-means clustering subpopulation division: Using the fitness vector as the clustering input feature, the K-means clustering algorithm is used to divide the search individuals into multiple subpopulations. Individuals in each subpopulation have similar performance, which facilitates efficient collaborative optimization in the local search space. This clustering process is performed dynamically to adapt to changes in individual status.

[0096] S3.1.2. Selecting guide individuals based on fitness values: Select α, β, and δ as guide individuals. Within each subpopulation, α is the individual with the best fitness in the current subpopulation, β is the second best, and δ is the third best.

[0097] S3.1.3. Position Update: For each search individual, update its path combination vector within its subpopulation, calculate the distance between the individual and individuals α, β, and δ, and use the three guide individuals to calculate candidate positions based on the principle of the gray wolf optimization algorithm to guide the position update of the remaining search individuals.

[0098] S3.1.4, Boundary and feasibility correction: perform boundary detection and track feasibility correction on the updated individual position;

[0099] S3.1.5. Subpopulation merging and iterative updating: Merge all updated subpopulations into a new global population, perform K-means clustering again, and enter the next round of iteration.

[0100] In order to better break the limitation of linear decrease in standard GWO, in S3.1.3, when updating the position, a nonlinear convergence factor is introduced. The parameter a is used as a control parameter to affect the value of the algorithm convergence factor A, which indirectly affects the iterative update of the position of a single gray wolf.

[0101] The Gray Wolf Optimization Algorithm (GWA) is a metaheuristic optimization algorithm that simulates the hunting behavior of gray wolf packs. Its core concept is based on the social hierarchy (α, β, δ, ω) of gray wolf packs and their behaviors of surrounding, chasing, and attacking prey. In the standard GWO, individual position updates depend on the convergence factor a and the coefficient vector r. The convergence factor a decreases linearly from 2 to 0, and the control algorithm transitions from global search to local exploitation.

[0102] When a adopts a nonlinear function, the value range of A no longer changes linearly from 2 to 0, but exhibits nonlinear fluctuations. After the introduction of the nonlinear convergence factor, the calculation logic of A and C in the position update formula remains unchanged, but the nonlinear characteristics of a change the dynamic range of A, thereby indirectly affecting the position iteration, thereby avoiding premature convergence caused by linear decrease and improving the algorithm's optimization ability in multi-modal function optimization.

[0103] In order to better perform position updates, S3.1.3 position updates are performed by combining dynamic weighted averaging and static averaging. The fitness difference between α and δ is compared with the dynamic threshold. When the difference exceeds the threshold, the dynamic weighted averaging strategy is used. When the difference is lower than the threshold, the original static averaging method is used.

[0104] Fitness is used in optimization algorithms to evaluate the quality of individuals (such as a set of path solutions in drone path planning). The fitness values ​​of different individuals or different iterations of the same algorithm are calculated, and the fitness differences are obtained through differences or ratios.

[0105] The dynamic threshold is a value that changes with the algorithm process or environmental factors. It can be dynamically adjusted based on historical fitness difference data. For example, the mean of the fitness differences of several past iterations plus a certain multiple of the standard deviation is taken as the threshold T. Alternatively, based on the characteristics of the task, for example, at the beginning of the task in a complex environment, a larger threshold can be set and then gradually reduced as the algorithm progresses.

[0106] When the fitness difference is greater than the threshold T, it means that the current algorithm has found a better solution or the solution space has changed significantly. In this case, a dynamic weighted average strategy is used. This strategy dynamically adjusts the weights based on factors such as individual fitness, allowing the algorithm to move faster towards a high-quality solution. When the fitness difference is less than the threshold T, it means that the solution space tends to be stable or the algorithm has converged. The original static average method is used to average information such as historical positions, smoothly update the trajectory, and prevent the algorithm from oscillating near the local optimum.

[0107] In order to better output the optimal planned path, when S4 outputs the optimal planned path, its judgment conditions include the following:

[0108] Maximum number of iterations: Set a fixed upper limit and terminate the calculation when the number of iterations is reached to ensure that the calculation time is controllable;

[0109] Convergence threshold judgment: If the improvement of the global optimal fitness for several consecutive generations is less than the threshold (such as 1e-6), the algorithm is considered to have converged and terminated early;

[0110] Member elite stagnation judgment: integrated with the KO algorithm, if the optimal solution has not been updated for more than N generations, it is considered to have fallen into a local optimum and can trigger termination or reinitialization;

[0111] Setting a clear maximum number of iterations (e.g., 200 generations) directly controls the upper bound of the algorithm's runtime. In urban multi-UAV collaborative missions, UAVs typically need to complete path planning within a limited time (e.g., time-sensitive logistics delivery requirements, or prime time constraints for emergency rescue). This condition ensures that the algorithm does not fall into infinite iterations, leading to planning delays and meeting real-time requirements.

[0112] The algorithm is judged to have converged by checking whether the improvement of the global optimal fitness is less than a threshold (e.g. 1e-6) for several consecutive generations (e.g. 10 generations). This condition can identify whether the algorithm is close to the optimal solution and avoid premature termination when the solution space fluctuates.

[0113] If the optimal solution has not been updated for more than N generations (e.g., 15 generations), the algorithm is considered to have fallen into a local optimum and triggers termination or reinitialization. This mechanism effectively overcomes the vulnerability of traditional GWO algorithms to local optimums, guiding the algorithm to explore new solution spaces through restart strategies (e.g., multi-start restart). For example, if a drone's path planning is stuck in a local optimum near a cluster of tall buildings, the elite stagnation detection triggers the algorithm to reinitialize the population and find a more optimal path around the obstacle.

[0114] The triple judgment conditions use a combination strategy of "time constraint + accuracy verification + local optimal detection" to ensure the real-time and reliability of the algorithm in engineering practice, and improve the path planning quality through theoretical convergence and anti-local optimal design, achieving a three-dimensional balance of "efficiency-accuracy-robustness".

[0115] The present invention provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the multi-machine collaborative path planning method are implemented.

[0116] The second object of the present invention is to provide a multi-UAV collaborative path planning system, including any one of the above-mentioned multi-UAV collaborative path planning methods in a city, such as Figure 5 As shown, it includes an information collection module 100, a path generation module 200, a clustering update module 300 and an optimal determination module 400;

[0117] The information acquisition module 100 obtains the initial position, target position and obstacle information of each UAV, and uses the grid method to model and define the urban low-altitude airspace mission scenario;

[0118] The path generation module 200 uses the MP-GWO algorithm to generate the initial path for multi-aircraft cooperation, constructs a multi-objective fitness vector that includes energy, altitude, no-fly zone threat, spatial collision, and time cooperation costs, and sets differentiated penalty coefficients for different types of no-fly zones;

[0119] The cluster update module 300 embeds K-means clustering into the optimization main loop, divides the subpopulations according to the fitness vector, and each subpopulation independently performs the nonlinear position update guided by α, β, and δ, and merges and iterates after correcting the path feasibility;

[0120] The optimal decision module 400 adopts the triple decision conditions of fixed number of iterations, convergence threshold and elite stagnation, and outputs the current optimal planning path when any one of the conditions is met.

[0121] like Figure 6 As shown in the figure, at the beginning, the initial position, target position and obstacle information of each UAV are obtained, and the initial path is generated by multi-machine collaboration. The path of each UAV is randomly generated using the GWO algorithm (represented by a position vector). Each group of paths corresponds to a "wolf". The fitness vector of the target of each "wolf" (that is, each UAV path) is calculated and used as input to perform K-means clustering on all search individuals, thereby obtaining k sub-populations with similar internal structures and consistent goals. Within each sub-population, a local guided search of α, β and δ wolves is performed. Specifically, the current individual and the guided positions of α, β and δ wolves are calculated, and an update strategy is executed to merge the updated sub-populations into the total population of the new generation in preparation for entering the next generation of iteration. The triple judgment conditions of fixed number of iterations, convergence threshold and elite stagnation are used to output the current optimal planning path when any condition is met.

[0122] In addition, it also includes a multi-UAV collaborative path planning device in the city, including a memory and a processor. When the memory and processor are executed, they are applied to any of the above-mentioned multi-UAV collaborative path planning systems.

[0123] The memory is used to store various data, algorithm programs, and operating status information required for drone path planning, and has high read and write speeds (to meet the real-time data update needs of drones), large capacity (to support massive data storage for multi-machine collaboration), and reliability (anti-interference, power-off data protection).

[0124] As the computing core of the device, the processor is responsible for executing path planning algorithms, processing real-time data, and coordinating multi-UAV collaboration. It has multi-core high-performance computing capabilities (supporting real-time operation of complex algorithms), low-power design (adapting to the long-term operation requirements of the equipment) and anti-interference capabilities (to cope with urban electromagnetic environments).

[0125] In addition, the invention also includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements any of the above-mentioned multi-UAV collaborative path planning systems;

[0126] Optionally, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0127] Optionally, the present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the multi-UAV collaborative path planning system of the above aspects.

[0128] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-machine collaborative path planning method in a city, characterized by: The following steps are involved: S1. Obtain the initial position, target position, and obstacle information of each UAV, and use the grid method to model and define the urban low-altitude airspace mission scenario; S2. Generate the initial multi-aircraft coordination path using the MP-GWO algorithm, construct a multi-objective fitness vector that includes energy, altitude, no-fly zone threat, spatial collision, and time coordination costs, and set differentiated penalty coefficients for different types of no-fly zones. S3, embed K-means clustering into the optimization main loop, divide the subpopulation according to the fitness vector, and each subpopulation independently performs the nonlinear position update guided by α, β, and δ. After correcting the path feasibility, merge and iterate; S4, adopt the triple judgment conditions of fixed number of iterations, convergence threshold and elite stagnation, and output the current optimal planning path when any condition is met.

2. The method for multi-machine collaborative path planning in an urban area according to claim 1, characterized in that: The specific steps of S2 generating the initial path for multi-machine collaboration are as follows: S2.1.1 Population Initialization: Given a flight space and mission objectives, generate several search individuals. Each search individual represents a complete set of UAV path planning solutions. Each individual is encoded as a combined vector of multiple UAV tracks, representing the sequence of track points of all UAVs from the starting point to the end point. S2.1.2, Fitness evaluation: Perform path quality evaluation on each individual and construct a multi-objective fitness vector. Each individual corresponds to a set of multi-dimensional fitness features.

3. The method for multi-machine collaborative path planning in an urban area according to claim 4, characterized in that: The multi-objective fitness vector constructed in S2.1.2 includes the following cost indicators: Energy consumption cost: calculated based on the total length of the path, and the length of each path segment is obtained by accumulating the Euclidean distance between waypoints; Flight altitude cost: When the drone's flight altitude exceeds the preset allowable range, a height penalty cost is incurred; No-fly zone threat cost: When a path crosses or approaches a no-fly zone, a penalty is generated based on the type of no-fly zone and the degree of proximity. Different types of no-fly zones, including but not limited to vegetation areas, built-up areas, large-scale activity areas, and government-controlled areas, are assigned different threat coefficients. Spatial collision cost: Compare the actual flight distance between any two drones at the trajectory point with the safe distance. If the safe distance is exceeded, the cumulative number of collisions will result in a penalty. Time coordination cost: Evaluate the time window range for each drone to reach the target point. If the actual arrival time of the drone exceeds the reasonable time range, the delay penalty cost will be increased.

4. The method for multi-machine collaborative path planning in an urban area according to claim 3, characterized in that: S2.1.2 constructs a multi-objective fitness vector, introduces regional determination logic into the fitness function, superimposes corresponding penalty values ​​based on whether the path passes through a specified area, and supports dynamic adjustment of no-fly zone restriction rules based on time.

5. The method for multi-machine collaborative path planning in an urban area according to claim 1, characterized in that: The specific steps of S3 using the K-means clustering algorithm to divide each group of paths are as follows: S3.1.

1. K-means clustering subpopulation division: Using the fitness vector as the clustering input feature, the K-means clustering algorithm is used to divide the search individuals into multiple subpopulations. Individuals in each subpopulation have similar performance, which facilitates efficient collaborative optimization in the local search space. This clustering process is performed dynamically to adapt to changes in individual status. S3.1.

2. Selecting guide individuals based on fitness values: Select α, β, and δ as guide individuals. Within each subpopulation, α is the individual with the best fitness in the current subpopulation, β is the second best, and δ is the third best. S3.1.

3. Position Update: For each search individual, update its path combination vector within its subpopulation, calculate the distance between the individual and individuals α, β, and δ, and use the three guide individuals to calculate candidate positions based on the principle of the gray wolf optimization algorithm to guide the position update of the remaining search individuals. S3.1.4, Boundary and feasibility correction: perform boundary detection and track feasibility correction on the updated individual position; S3.1.

5. Subpopulation merging and iterative updating: Merge all updated subpopulations into a new global population, perform K-means clustering again, and enter the next round of iteration.

6. The method for multi-machine collaborative path planning in an urban area according to claim 6, characterized in that: When updating the position in S3.1.3, a nonlinear convergence factor is introduced, and parameter a is used as a control parameter to affect the value of the algorithm convergence factor A, thereby indirectly affecting the iterative update of the position of a single gray wolf.

7. The method for multi-machine collaborative path planning in an urban area according to claim 1, characterized in that: When S4 outputs the optimal planning path, its judgment conditions include the following: Maximum number of iterations: Set a fixed upper limit and terminate the calculation when the number of iterations is reached to ensure that the calculation time is controllable; Convergence threshold judgment: If the improvement of the global optimal fitness for several consecutive generations is less than the threshold (such as 1e-6), the algorithm is considered to have converged and terminated early; Member elite stagnation judgment: integrated with the KO algorithm, if the optimal solution has not been updated for more than N generations, it is considered to have fallen into a local optimum, which can trigger termination or reinitialization.

8. A system for implementing multi-machine collaborative path planning within a city, comprising the multi-machine collaborative path planning method within a city according to any one of claims 1 to 8, characterized in that: It includes an information collection module (100), a path generation module (200), a clustering update module (300) and an optimal determination module (400); The information acquisition module (100) obtains the initial position, target position and obstacle information of each UAV, and uses a grid method to model and define the urban low-altitude airspace mission scenario; The path generation module (200) generates a multi-aircraft collaborative initial path using the MP-GWO algorithm, constructs a multi-objective fitness vector including energy, altitude, no-fly zone threat, spatial collision and time collaboration cost, and sets differentiated penalty coefficients for different types of no-fly zones; The cluster update module (300) embeds K-means clustering into the optimization main loop, divides subpopulations according to fitness vectors, and each subpopulation independently performs nonlinear position update guided by α, β, and δ, and merges and iterates after correcting the path feasibility; The optimal decision module (400) adopts a triple decision condition of fixed number of iterations, convergence threshold and elite stagnation, and outputs the current optimal planning path when any one of the conditions is met.

9. A device for urban multi-machine collaborative path planning, comprising a memory and a processor, characterized in that: When executed, the memory and processor are applied to the system for implementing multi-machine collaborative path planning in a city as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, it implements a multi-machine collaborative path planning system within a city as described in any one of claims 1-7.

Images