Multi-ant colony parallel path planning method and system based on dynamic global elite pool

Abstract

The application discloses a kind of multi-ant colony parallel path planning method and system based on dynamic global elite pool, belong to intelligent optimization algorithm and path planning technical field.For the problems of slow convergence speed, low solution quality, fixed communication strategy and low computational efficiency of existing ant colony path planning method, a high-quality solution set is pre-filled in the main iteration by constructing a global elite pool;Each sub-ant colony triggers asynchronous communication according to the stagnation generation, stores the high-quality solution into the elite pool and feeds back to the new solution generation, realizes the dynamic tilt of communication resources;Each sub-ant colony adjusts the elite utilization rate according to the local optimal cost and dynamically divides the population, realizes the balance of personalized exploration and utilization;By calculating the cost relative improvement rate of the global optimal path and updating the stagnation count accordingly, the iteration is terminated when the preset termination condition is met.The application effectively improves the convergence speed, solution quality and computational efficiency of three-dimensional continuous space path planning, and is suitable for unmanned system autonomous navigation path planning task.

Description

Technical Field

[0001] This invention relates to the field of intelligent optimization algorithms and path planning technology, and in particular to a multi-ant colony parallel path planning method and system based on a dynamic global elite pool. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Path planning is one of the core problems in autonomous navigation for unmanned systems. Its goal is to find the feasible path between the starting point and the destination that minimizes cost while satisfying various constraints in three-dimensional space. Ant colony optimization (ACO) has been widely used for path planning in continuous space due to its excellent global search capabilities. However, existing methods suffer from the following main problems: (1) In existing multi-group ant colony methods, each sub-ant colony usually maintains its own independent local optimum. There is a lack of a unified high-quality solution sharing container among sub-ant colonies. Elite solutions searched by excellent sub-ant colonies cannot be used by other sub-ant colonies in a timely manner, resulting in a waste of global search resources and a decrease in the overall convergence speed. Even if some methods introduce migration operations, only a small number of individuals are transferred between adjacent sub-ant colonies at fixed intervals. The sharing scope is limited and the information timeliness is poor, making it impossible to achieve unified management and on-demand guidance of elite solutions on a global scale.

[0004] (2) The communication strategy among sub-ant colonies is fixed and the allocation of communication resources is unreasonable. Most existing multi-swarm parallel methods adopt a fully synchronous migration strategy or a fixed interval migration strategy, that is, all sub-ant colonies exchange information at the same time or at the same interval, and the communication resources are evenly distributed among the sub-ant colonies. It is impossible to dynamically adjust the communication frequency according to the real-time search performance of each sub-ant colony.

[0005] (3) The algorithm is serialized and has low computational efficiency. The existing ant colony path planning implementation usually processes each ant individually. Operations such as position update, difference statistics, and boundary processing are all executed serially in a loop, which cannot take advantage of the parallel acceleration of matrix vector operations.

[0006] (4) The exploration and utilization ratio of each sub-ant colony is uniform and fixed, which cannot reflect individual differences. The existing method uses the same exploration and utilization parameter settings for all sub-ant colonies, ignoring the objective differences in the different stages of the search process of different sub-ant colonies. It cannot achieve personalized dynamic balance for the search status of each sub-ant colony, and the overall search efficiency needs to be improved.

[0007] (5) Fixed iteration termination conditions lead to invalid iterations or premature termination. Existing ant colony path planning methods typically use a preset fixed number of iterations as the termination condition, failing to adaptively adjust according to the actual convergence state of the algorithm. When the algorithm has fully converged and the solution quality no longer improves, the remaining iterations are still executed, resulting in a significant waste of computational resources. In some complex scenarios, the fixed number of iterations may terminate prematurely before the solution has fully converged, leading to a loss of final solution quality. This static termination strategy cannot guarantee the effective use of computational resources and is difficult to adapt to the differences in convergence speed under different complexity scenarios. Summary of the Invention

[0008] To address the shortcomings of existing ant colony path planning methods, such as slow convergence speed and low solution quality, this invention provides a multi-ant colony parallel path planning method and system based on a dynamic global elite pool. By constructing a global elite pool and pre-filling it before the main iteration, the pool possesses a non-empty, high-quality solution set at the start of iteration, solving the problem of missing elite guidance in the early stages of traditional methods. Asynchronous communication is triggered by each sub-ant colony based on stagnation generation and local optimal cost, dynamically storing high-quality solutions in the elite pool and feeding them back into new solution generation, achieving dynamic allocation of communication resources to high-performing sub-ant colonies. By adjusting the elite utilization rate of each sub-ant colony based on local optimal cost and dynamically dividing the population accordingly, personalized exploration and utilization balance among different sub-ant colonies are achieved. The relative improvement rate of the cost of the globally optimal path is calculated at preset intervals, and the stagnation count is updated accordingly. Iteration terminates when preset conditions are met, avoiding resource waste from a fixed number of iterations. This invention effectively improves the convergence speed, solution quality, and computational efficiency of path planning in three-dimensional continuous space, achieving high-quality and high-efficiency path planning in three-dimensional continuous space.

[0009] On the one hand, a multi-ant colony parallel path planning method based on a dynamic global elite pool is provided, including: Initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on the environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. Construct a global elite pool, generate an initial path code in the spherical coordinate parameter domain, and store it in the global elite pool after local optimization; Each sub-ant colony uses matrix-based batch computation to process the path code of the current population and generate a new path code; Merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony. When the ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, it triggers communication and stores the corresponding path code in the global elite pool. Each sub-ant colony adjusts its elite utilization rate based on the local optimal cost, and classifies its population dynamics into different categories according to the elite utilization rate. Obtain the current optimal path in the global elite pool, calculate its relative cost improvement rate every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

[0010] Furthermore, in the spherical coordinate parameter domain, each path code consists of three continuous variables: step size, pitch angle, and azimuth angle; the maximum value of the step size is determined by the distance between the starting point and the ending point and the number of path segments, and the range of values ​​for the pitch angle and azimuth angle is determined by the direction vector from the starting point to the ending point.

[0011] Furthermore, the local optimization adopts a greedy hill-climbing strategy: Gaussian random perturbation is applied to the initial path encoding on the three components of spherical coordinates. After constraint repair and cost evaluation, only the neighborhood solution with lower path cost is accepted, and the process is iterated for a preset number of times.

[0012] Furthermore, the matrix-based batch calculation includes: organizing the path codes of all individuals in the current population of each sub-ant colony into a step length matrix, a pitch angle matrix, and an azimuth angle matrix; calculating the step length matrix based on the difference between the current population and the parent population; and using the step length matrix to apply a perturbation to the current path code to generate new path codes in batches.

[0013] Furthermore, the optimal retention based on path cost specifically involves merging the path codes in the current population with the new path codes, arranging them in ascending order of path cost, and retaining a predetermined number of individuals with the lowest path cost to update the population.

[0014] Furthermore, the triggered communication includes: immediate communication when the sub-ant colony discovers that the cost of the path code is better than the cost of the current optimal path in the global elite pool, or periodic communication when the sub-ant colony reaches a preset communication interval; after triggering communication, the corresponding path code is stored in the global elite pool, and the communication interval of the sub-ant colony is updated.

[0015] Furthermore, the elite utilization rate is adaptively adjusted based on the recent improvement magnitude of the local optimal cost: when the improvement magnitude is greater than a first preset threshold, the elite utilization rate is increased; when the improvement magnitude is less than a second preset threshold, the elite utilization rate is decreased; otherwise, the current value is maintained, wherein the first preset threshold is greater than the second preset threshold; the proportion of different categories is dynamically allocated as the elite utilization rate changes, and the different categories include three types: elite utilization type, elite guidance type, and autonomous exploration type.

[0016] Furthermore, the preset termination condition is: the stall count reaches the preset maximum stall number and the current iteration number reaches the minimum iteration number.

[0017] On the other hand, a multi-ant colony parallel path planning system based on a dynamic global elite pool is also provided, including: The initialization module is configured to: initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on the environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. The elite pool construction module is configured to: construct a global elite pool, generate an initial path code in the spherical coordinate parameter domain, and store it in the global elite pool after local optimization; The parallel path generation module is configured to: each sub-ant colony uses matrix-based batch computation to process the path encoding of the current population and generate new path encodings; The population update module is configured to: merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony; The communication module is configured to: when the sub-ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, trigger communication and store the corresponding path code in the global elite pool. The adaptive control module is configured to: each sub-ant colony adjusts the elite utilization rate according to the local optimal cost, and dynamically divides its own population into different categories according to the elite utilization rate; The termination and output module is configured to: obtain the current optimal path in the global elite pool, calculate its relative cost improvement rate every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

[0018] In another aspect, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the method described in the first aspect.

[0019] The above technical solution has the following advantages or beneficial effects: (1) This invention achieves unified storage and on-demand guidance of high-quality path encoding across sub-ant colonies by constructing a global elite pool and pre-filling it before the start of the main iteration. This eliminates the problem that traditional multi-swarm ant colony methods cannot guide elites in the early stages of the main iteration due to an empty elite pool, thus accelerating the overall convergence process of the algorithm. In traditional methods, elite solutions searched by excellent sub-ant colonies cannot be utilized by other sub-ant colonies in a timely manner. This invention, through the dynamic management mechanism of the elite pool, including adding high-quality solutions to the pool, removing inferior solutions, and removing redundancy from near-repeating solutions, ensures that the solution set in the pool remains refined and diverse, effectively avoiding the waste of global search resources, improving the optimization efficiency of three-dimensional path planning, and outputting higher-quality feasible paths within a limited number of iterations.

[0020] (2) The asynchronous communication mechanism adopted in this invention enables each sub-ant colony to dynamically adjust its communication interval and elite reading probability based on its own search performance, dynamically tilting limited communication resources towards sub-ant colonies with excellent search performance or those that have stagnated. In traditional methods, all sub-ant colonies distribute communication resources equally at the same time or interval, making it impossible to differentiate scheduling based on the real-time status of the sub-ant colonies. The asynchronous communication mechanism of this invention significantly improves the timeliness of information dissemination and global communication efficiency. Each sub-ant colony can play a unique search value during the iteration process, without any single sub-ant colony monopolizing or making zero contribution. This mechanism enables multi-unmanned systems to quickly respond to environmental changes in collaborative path planning, improving the planning timeliness in complex scenarios.

[0021] (3) The matrix-based batch computation and coordinate transformation optimization adopted in this invention organizes the path codes of all individuals in each sub-ant colony into a matrix form, and completes operations such as parent selection, step size calculation, new solution generation and coordinate transformation in batches, which greatly reduces the cyclic redundancy and computation time under large population scale. Traditional methods process each ant individually, and operations such as position update, difference statistics and boundary processing are all executed in a serial loop, which is computationally inefficient. This invention, through matrix-based parallel processing, has particularly prominent advantages when the population scale is large, significantly improving the overall running speed of the algorithm, shortening the waiting time for unmanned systems to perform path replanning in three-dimensional space, and improving the real-time performance and responsiveness of task execution.

[0022] (4) The adaptive elite utilization mechanism at the sub-ant colony level introduced in this invention maintains an elite utilization parameter independently for each sub-ant colony. The elite utilization is adaptively adjusted based on the improvement of the recent optimal path of each sub-ant colony, and individuals within each sub-ant colony are dynamically divided into three categories: elite utilization type, elite guidance type, and autonomous exploration type. This mechanism enables sub-ant colonies at different search stages to achieve a personalized dynamic balance between exploration and utilization. When the improvement is significant, the elite utilization is increased to strengthen refined development; when the improvement is insufficient, the elite utilization is reduced to broaden the exploration space. This effectively avoids the efficiency bottleneck caused by globally uniform parameter settings, enabling the 3D path planning algorithm to adaptively adjust its search strategy in scenarios with different terrain complexities, maintaining sufficient exploration diversity in areas with dense obstacles, and accelerating convergence to the optimal path in open areas.

[0023] (5) The adaptive termination mechanism proposed in this invention flexibly controls the number of iterations based on the actual convergence state, and saves computing resources to the maximum extent while ensuring the quality of the solution. It overcomes the defects of the traditional fixed iteration method, which continues to execute invalid iterations after the algorithm has fully converged, or terminates too early in complex scenarios, resulting in loss of solution quality. It is applicable to three-dimensional path planning scenarios of different complexities. This mechanism enables unmanned systems to stop calculation in a timely manner after the solution quality has fully converged in three-dimensional path planning tasks of different complexities, avoiding unnecessary computing power consumption. It is applicable to autonomous navigation tasks in various typical low-altitude flight scenarios such as mountains and urban canyons. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 This is a flowchart of the overall method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the optimal path planning results in a typical three-dimensional continuous space scenario according to Embodiment 1 of the present invention; Figure 3 This is a graph showing the evolution of local optimal path cost during the iteration process of each parallel sub-ant colony in Embodiment 1 of the present invention; Figure 4 This is a diagram showing the path planning results of Embodiment 1 of the present invention in a large-scale urban environment with dense obstacles. Figure 5 This is a graph showing the convergence curve of the global optimal path cost in Embodiment 1 of the present invention; Figure 6 This is a curve comparing the convergence performance of the method in Embodiment 1 of this invention with that of the traditional multi-ant colony algorithm; Figure 7This is a bar chart showing the number of times different sub-ant colonies contribute to the optimization of the global optimal solution in Embodiment 1 of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. Those skilled in the art should understand that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] Example 1 This embodiment provides a multi-ant colony parallel path planning method based on a dynamic global elite pool, applicable to autonomous navigation path planning tasks of unmanned systems in three-dimensional continuous space. Existing multi-ant colony methods lack a unified, high-quality solution-sharing container among sub-ant colonies, making it difficult for elite solutions found by superior sub-ant colonies to be utilized by other sub-ant colonies in a timely manner. Furthermore, fixed communication strategies between sub-ant colonies, serialized computation processes, fixed exploration-to-utilization ratios, and rigid iteration termination conditions lead to slow convergence speed, unsatisfactory solution quality, and low computational efficiency. To address these issues, this embodiment designs an improved scheme combining a multi-sub-ant colony parallel ant colony structure with a global elite pool. First, a global elite pool is constructed as a unified container for cross-population information sharing among all sub-ant colonies. The elite pool is pre-filled before the main iteration begins to solve the problem of missing elite guidance in the early iteration stages of traditional methods.

[0029] Building upon this foundation, an asynchronous communication mechanism is introduced, enabling each sub-ant colony to dynamically adjust its communication timing and frequency based on its own search performance, dynamically allocating communication resources to sub-ant colonies with excellent search performance or those that have stagnated. Simultaneously, matrix-based batch computation replaces the traditional serial processing method, improving computational efficiency. Furthermore, an adaptive elite utilization parameter is independently set for each sub-ant colony, achieving a personalized dynamic balance between exploration and utilization. In addition, an adaptive iterative termination mechanism based on a relatively improved threshold is introduced, allowing the algorithm to stop promptly after sufficient convergence, avoiding unnecessary consumption of computational resources. This method, based on a multi-sub-ant colony parallel ant colony architecture and a global elite pool as the core of cross-population information sharing, comprehensively utilizes asynchronous communication, matrix-based computation, sub-ant colony adaptive parameter adjustment, and a relatively improved termination mechanism to achieve efficient and high-quality path planning.

[0030] Figure 1 The following is an overall flowchart of the method according to Embodiment 1 of the present invention, which specifically includes the following steps.

[0031] S101: Initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. S102: Construct a global elite pool, generate initial path codes in the spherical coordinate parameter domain, and store them in the global elite pool after local optimization; S103: Each sub-ant colony uses matrix-based batch computation to process the path code of the current population and generate a new path code; S104: Merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony. S105: When the sub-ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, it triggers communication and stores the corresponding path code in the global elite pool. S106: Each sub-ant colony adjusts its elite utilization rate based on the local optimal cost, and classifies its population dynamics into different categories according to the elite utilization rate. S107: Obtain the current optimal path in the global elite pool, calculate its relative improvement rate of cost every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

[0032] Specifically, in step S101, a three-dimensional continuous domain parallel ant colony model is constructed. First, a three-dimensional continuous path planning space is constructed, and the starting coordinates are set. End point coordinates An environmental model incorporating spatial boundary constraints, obstacle constraints, and flight attitude constraints is established, and the path cost function is determined. : ; in, The total spatial length of the entire path is represented by the sum of the three-dimensional Euclidean distances between all adjacent control points along the path. The expression for the safety penalty value between the path and the obstacle is: ; in, Let be the three-dimensional Euclidean distance from the i-th path control point to the nearest obstacle. A preset safe obstacle avoidance distance (preferably 3 meters in this embodiment) is set; the safety penalty value increases as the distance between the path control point and the obstacle decreases, which is used to constrain the aircraft to stay away from the obstacle; The expression for the flight path attitude smoothing penalty is: ; in, The turning angle between adjacent flight segments. The maximum allowable smooth turning angle for the aircraft; the flight attitude smoothing penalty increases as the turning angle of adjacent segments increases, in order to suppress sharp turns at large angles; , and These are three weighting coefficients, set to 0.4, 0.4, and 0.2 respectively, used to adjust the proportion of the three types of indicators in the total cost. They can be flexibly set during the simulation stage according to different planning scenarios such as mountainous areas and urban low-altitude areas.

[0033] To achieve digital solution of paths in continuous space, spherical coordinates are used to parametrically model the paths in three-dimensional continuous space. The sequence of intermediate control points of the path is represented in spherical coordinates, and each path segment is represented by a step size r, a pitch angle, and a vertical axis. and azimuth It consists of three continuous variables. Let the path consist of n segments, then the complete path encoding of the k-th ant in the m-th sub-ant colony is: ; Where m is the sub-ant colony number, k is the individual ant number, t is the global iteration count, and n is the total number of flight path segments. , and These are the radial step lengths (path lengths) for the 1st, 2nd, and nth segments, respectively. , and These are the pitch angles for the first, second, and nth segments of the path, respectively. , and These are the azimuth angles of the first, second, and nth segments, respectively.

[0034] The feasible domain of each variable is jointly determined by the origin-end direction vectors, spatial boundaries, and attitude constraints. Specifically, the path parameter domain is set as follows: ; in, This represents the maximum step size. The reference azimuth angle is the angle from the starting point to the ending point. It is a direction vector. These are the X and Y axial components of the vector, respectively.

[0035] Define coordinate transformation function To achieve the transformation from spherical coordinates to Cartesian coordinates: ; Triaxial constraint repair operator and Each of the three axes is checked separately. First, it is determined whether the coordinates exceed the planned space boundary or overlap with obstacles. If the coordinates violate the rules, a fine-tuning step size is set along the reference direction vector from the starting point to the ending point, and the coordinate components are iteratively updated. The constraints are continuously checked until the coordinates meet all constraints or the maximum number of iterations is reached. Then, the process is completed. Return to spherical coordinates to complete the path evaluation link.

[0036] The multi-substrate ant colony structure was then divided. The total number of ants... Each individual ant is divided into M independent, parallel search sub-ant colonies, denoted as . The number of individuals in each sub-ant colony is Each sub-colony iterates independently in each round of global iteration, and each sub-colony... Each individual in its population and its local optimal solutions are independently maintained. And through the global elite pool To achieve elite sharing across sub-ant colonies, a three-dimensional continuous domain parallel ant colony model suitable for path planning in three-dimensional continuous space is formed.

[0037] To balance the algorithm's search efficiency, convergence speed, and solution quality, this embodiment selects the following typical values ​​for key parameters: population M = 5, total population size. =80, size of each sub-ant colony =16, Elite Pool Capacity Inferior solution elimination coefficient Elite deduplication threshold Communication interval attenuation coefficient upper bound of communication interval between ant colonies The lower bound of the communication interval between ant colonies Upper bound of read probability Maximum number of attempts for local neighborhood search Relative Improved Threshold Maximum stall count Minimum number of iterations .

[0038] After setting the above parameters, initialize the path matrix, cost vector, and communication interval for each sub-ant colony. Stagnation count and elite utilization rate Initialize the global elite pool to empty.

[0039] In step S102, before entering the main iteration, a pre-filling operation of the global elite pool is first performed to eliminate the problem that traditional ant colony algorithms cannot provide elite guidance in the early stages of iteration due to an empty elite pool. This invention constructs a unified global elite pool. As a container for cross-population information sharing among all sub-ant colonies, it breaks down information barriers between sub-ant colonies. Before the main iteration starts, greedy hill-climbing local optimization is used to preprocess the initial sampled solutions, screen and store several optimized high-quality elite solutions, and complete the pre-filling of the global elite pool, thereby eliminating the problem of traditional methods being unable to guide elites in the early stages of iteration due to an empty elite pool.

[0040] Specifically, multiple initial path codes are randomly generated within the feasible region of spherical coordinates, with a step size r in the interval [missing information]. Internal sampling, where, , The reference step size for a single path segment is used, and the pitch and azimuth angles are uniformly and randomly sampled within the feasible region. After sampling... After completing the transformation from spherical coordinates to Cartesian coordinates, constraint repair is performed to eliminate path out-of-bounds and obstacle crossing problems. Then, path cost evaluation is carried out to screen out effective candidate solutions.

[0041] Next, for each group of valid candidate solutions, an 8-step greedy hill climbing local optimization is performed. Gaussian random perturbation is applied to the three components of the spherical coordinate path encoding, namely the step size, pitch angle, and azimuth angle, with the current component value as the mean, to generate neighborhood candidate codes. After constraint repair and cost evaluation, the neighborhood candidate codes obtained by perturbation are only accepted for neighborhood solutions with lower costs. After multiple iterations, elite individuals with quality better than the randomly initialized solutions are selected and stored in the global elite pool.

[0042] All valid solutions after local optimization are sorted in ascending order of path cost and stored in the global elite pool, completing the pre-filling process. After the above pre-filling operation, the global elite pool has a non-empty set of high-quality solutions before the main iteration starts, providing a reliable starting point for elite guidance in the main iteration.

[0043] In step S103, after the global elite pool is pre-filled, the main iteration loop begins. After the main iteration starts, each sub-ant colony synchronously, independently, and in parallel executes the construction of a three-dimensional continuous path. The entire process adopts matrix-based batch processing to replace the traditional serial loop processing method based on a single ant, which significantly improves computational efficiency.

[0044] Specifically, at the beginning of each iteration, based on the current iteration progress... Synchronously update global attenuation parameters: ; Where t is the current iteration number, The reference iteration baseline number is set to 60 in this embodiment. This is the step size decay coefficient, used to control the magnitude of the search step size decay. This parameter decays linearly from its initial value to its minimum value as the iteration progresses, achieving a smooth transition of the algorithm from wide-area exploration to fine-grained development. This is the global amplitude attenuation coefficient, used to control the overall perturbation amplitude generated by the new solution. For deterministic component weights, The weights are random components, which are adaptively adjusted as the iteration progresses, and satisfy a normalization constraint that their sum equals 1. and For normalized attenuation, this embodiment preferably uses... =0.85, =0.15.

[0045] In each iteration, based on the historical experience of each sub-ant colony, the probability matrix is ​​used... The matrix has M rows, which represents the total number of parallel sub-ant colonies, and the three columns correspond to the three-axis constraint repair operators. and Each element in the matrix represents the probability of the corresponding sub-ant colony selecting one of the three types of operators, and the sum of the probabilities in each row is 1. In the initial iteration, the matrix is ​​initialized with uniform probability. In each iteration, the cost improvement generated by each operator's repair path is calculated. The greater the positive benefit, the higher the probability of selecting the corresponding operator. The probabilities of each row of the matrix are updated dynamically in this way. Each sub-ant colony randomly selects a repair operator based on its own row probability, so as to realize the adaptive adjustment of the repair strategy.

[0046] In the continuous domain, ants do not perform discrete node transfers; instead, they apply directional, continuous perturbations to the current path code to generate a new path code. Each sub-ant colony organizes the path codes of all individuals within the population into three position matrices:

[0047] in, Let m be the number of individuals in the m-th sub-ant colony, and let step size matrix be the number of individuals in the sub-ant colony. It is determined by the column-wise mean of the differences between the current population individuals and their parents, and a global decay factor is introduced. Perform the calculation: ; in, This is a column-wise mean operator that calculates the average value of the input matrix column by column. The path encoding matrix of the distance dimension between the parent and child ants of the m-th child ant colony. The path encoding matrix of the contemporary distance dimension of the m-th sub-ant colony. For numerical stability terms (preferably set to...) Pitch angle matrix and azimuth matrix The same method is used to calculate the path encoding matrix based on the pitch angle and azimuth angle dimensions, respectively.

[0048] The new path encoding is generated by a weighted synthesis of deterministic components pointing to the global optimum and random noise components: ; in, Let be the direction vector used by the k-th ant in the m-th sub-ant colony during the t-th iteration to generate the perturbation. Let be the random perturbation vector of the k-th ant in the m-th sub-ant colony. This is the globally optimal path encoding vector obtained in the t-th iteration.

[0049] Final new path encoding The generating formula is: ; When selecting parents from autonomously exploring individuals in the ant colony, a Gaussian kernel roulette wheel selection method is used, assigning selection probabilities based on the individual's ranking within the population: ; in, The Gaussian kernel width parameter, Let be the normalized probability of the parent selection of the k-th individual. The original weights of the Gaussian kernel corresponding to the k-th individual are: Let be the original Gaussian kernel weight of the j-th individual in the population. Higher-ranking, high-quality individuals receive a higher selection probability, balancing utilization and exploration.

[0050] After performing boundary clipping on all newly generated path codes, they are batch-converted to Cartesian coordinates, the selected repair operator is executed to complete constraint repair, the path cost C(·) is evaluated, and then they are batch-converted back to spherical coordinates. For autonomous exploration individuals, a Levy flight step is additionally introduced. Replace the normal noise component. This is a characteristic parameter of the Lévy distribution, used to adjust the step size distribution characteristics and control the frequency of long-distance jumps. It enhances the ability to perform long-distance jumps in the solution space.

[0051] The complete coordinate transformation and constraint repair process described above is executed only once in batches, which significantly reduces the computational redundancy caused by repeated coordinate transformations of a single ant in traditional methods, and particularly improves the overall operating efficiency in large populations.

[0052] This invention organizes the path encoding of all ants in each sub-ant colony into three position matrices, batch-completing operations such as parent selection, step size calculation, new solution generation, and boundary pruning, fully leveraging the advantages of parallel acceleration. In the coordinate transformation chain, all ants within a batch undergo unified batch conversion from spherical to Cartesian coordinates, constraint repair, and conversion back to spherical coordinates, avoiding the large amount of redundant computation caused by repeated coordinate transformations of individual ants, significantly improving overall operational efficiency, especially at large population scales. In the matrix-based batch solution construction, the generation direction of the new path encoding is a weighted synthesis of deterministic components pointing to the global optimum and random noise components. The weight of the deterministic component gradually increases as the iteration progresses, allowing the algorithm to smoothly transition from early wide-area exploration to later fine-grained development. An additional Lévy flight step size is introduced for exploratory individuals, enhancing the ability to jump long distances across the solution space.

[0053] In step S104, after each sub-ant colony completes the generation of the new path code in step S103, this step further performs elite competition screening and local fine search within the sub-ant colony to improve the quality of the local optimal solution. At the same time, the stagnation state of each sub-ant colony is statistically analyzed to provide a data basis for subsequent asynchronous communication and adaptive adjustment of elite utilization.

[0054] Specifically, the first step is to perform an elite competitive selection process. This involves setting the current ant population in each sub-colony... Path encoding and new generation of individuals The new path codes of each individual are merged (total of 2) (items), sorted in ascending order of path cost, retaining the top [number] items. The individual with the lowest cost is selected as the next generation of the ant colony. Through the above merging and selection operations, the lower-quality individuals in the ant colony are eliminated, while the high-quality individuals are retained and enter the next generation, thus realizing the selection and renewal of elites within the ant colony.

[0055] After the population update is complete, further determination is made as to whether a local fine-grained search has been triggered. When the current best solution of the child ant colony is more than 1% better than that of the parent colony, three additional local searches are performed on the current best solution: ; in, This is the optimal path for the m-th sub-ant colony. The local search perturbation scaling factor (preferably set to 0.08 in this embodiment) controls the Gaussian perturbation amplitude within the optimal neighborhood. It is a Gaussian random perturbation vector that follows a standard multivariate normal distribution.

[0056] After each local search, following constraint repair and cost evaluation, if the path cost of the new path encoded after perturbation... Path cost lower than the current optimal path If the current best solution is replaced, then potential better solutions can be further explored within the current best neighborhood.

[0057] At the same time, record the path cost corresponding to the current best path encoding of each sub-ant colony. This serves as the basis for subsequent asynchronous communication and elite pool updates. Simultaneously, a continuous stagnation algebra is introduced. (t), whose update rule is: if the current generation is an improvement over the previous generation, then (t) Clear to zero, no (t)+1. The above stagnation statistics will provide data support for the communication trigger determination of each sub-ant colony in step S105. The higher the stagnation generation of the sub-ant colony, the greater the probability of it asynchronously reading the elite pool, thereby obtaining more external elite guidance to help it escape the local optimum.

[0058] In step S105, after the elite competition and stagnation statistics within the sub-ant colony are completed, this step further performs asynchronous communication between sub-ant colonies and dynamic maintenance of the global elite pool. This invention employs an asynchronous communication strategy to replace the synchronous migration method with fixed intervals and equal resource allocation in traditional multi-swarm methods. This allows communication resources to be dynamically tilted towards sub-ant colonies with excellent search performance or those that have stagnated, thereby improving the timeliness of global information sharing and communication efficiency.

[0059] Specifically, each sub-ant colony independently maintains a dynamic communication interval parameter. After each iteration, each sub-ant colony independently determines whether to trigger communication, which is triggered if one of the following conditions is met: When the ant colony discovers that the cost of the path encoding is better than the cost of the current best path in the global elite pool, it immediately triggers instant communication, updates the global best path, and shortens the next communication interval according to the following formula, so that high-quality search results can be quickly broadcast to the global elite pool: ; in, This serves as the lower bound for the communication interval iteration of the sub-ant colony. This is the communication interval attenuation coefficient. .

[0060] When the ant colony reaches the current communication interval When this occurs, periodic communication is triggered, and the next communication interval is extended according to the following formula, enabling the ant colony to periodically share search results at a preset minimum cycle: ; in, Let m be the iteration number in which the communication was last triggered by the m-th sub-ant colony. The upper bound of the communication interval for the sub-ant colony is used for iteration.

[0061] After communication is triggered, the global elite pool is dynamically maintained to ensure that the solutions in the pool are refined, diverse, and of high quality. Specifically, the two individuals with the best cost within the sub-ant colony are added to the global elite pool and sorted in ascending order of path cost; [the remaining individuals are removed]. Find similar solutions to remove duplicates; eliminate solutions that satisfy... The low-quality solution. Among them, The path cost function value corresponding to path code P. The path cost of the i-th path in the elite pool. Let $\mathbf{j}$ be the path cost of the $j$-th path in the elite pool. The cost of the current globally optimal solution is... The elite pool is set as a deduplication threshold to ensure that it always maintains a refined and diverse set of high-quality solutions. If the number of solutions in the elite pool exceeds the preset capacity of 20, the top 20 solutions with the lowest path cost are retained.

[0062] At the same time, at the end of each generation, each sub-ant colony performs asynchronous reads according to the probability. (t) Read the global elite pool to obtain the elite solution. When the sub-ant colony stagnates for consecutive generations... If the probability is too high, automatically increase the probability of asynchronously reading from the elite pool. This allows stagnant ant colonies to receive more guidance from external elites, promoting them to break out of local optima and dynamically tilting communication resources toward ant colonies with better search performance, thereby improving global communication efficiency.

[0063] The formula for calculating the probability of reading is: ; in This represents the relative difference between the sub-ant colony and the global optimum. To read the upper bound of the probability, The base read probability is set to 0.3 in this embodiment. Encode the historical best path of the m-th sub-ant colony within iteration t. .

[0064] After reading the elite solution, apply a Gaussian perturbation to the elite path:

[0065] in, To encode the elite path after applying a Gaussian perturbation, The path encoding for elites read from the global elite pool. The standard deviation of the asynchronous elite Gaussian perturbation. This is used to dynamically adjust the intensity of Gaussian perturbation. After spatial constraint repair and path cost evaluation, if the performance of the perturbed individual is better than the current worst individual in the sub-ant colony, then the perturbed individual is replaced and the sub-ant colony is reordered.

[0066] In the above asynchronous communication mechanism, the communication interval of each sub-ant colony is initialized to an interval. The random value within the range is dynamically adjusted during the iteration process based on the frequency with which the sub-ant colony discovers the optimal solution; the asynchronous reading probability of the sub-ant colony to the elite pool takes into account both the relative gap between itself and the global optimum and the number of consecutive stagnant algebras, and both are constrained by upper bounds to prevent communication from being too frequent.

[0067] In step S106, based on the completion of dynamic updates and asynchronous communication of the elite pool, this step further performs adaptive adjustments to the elite utilization rate within each sub-ant colony. This invention maintains an independent elite utilization rate parameter for each sub-ant colony. This parameter is used to dynamically adjust the ratio between search guided by global elite solutions and autonomous exploration within the sub-ant colony. It directly determines the distribution of the three types of individuals within the sub-ant colony and is a crucial link connecting the acquisition of global elite information with the regulation of individual behavior within the sub-ant colony.

[0068] Specifically, every Δt generations (Δt=10 in this embodiment), each sub-ant colony calculates its recent relative improvement based on the optimal cost history recorded in the last Δt generations: ; Wherein, Δt is a preset detection period parameter. In this embodiment, Δt=10, but in practical applications, it can be flexibly set according to the complexity of the problem.

[0069] In this embodiment, two sets of thresholds are set to adaptively adjust the elite utilization rate: the first preset threshold is 5%, and the second preset threshold is 2%. When the improvement... When the utilization rate is between 2% and 5%, maintain the elite utilization rate unchanged; if the improvement is significant... If the improvement is less than 2%, then the utilization rate of elites will be reduced to broaden the space for independent exploration, allowing more individuals to conduct independent searches without elite guidance; if the improvement is significant... If it is greater than 5%, then the utilization rate of elites will be increased to strengthen the utilization and refined development of elites, so that more individuals can quickly converge with the global elites.

[0070] definition The incremental adjustment of elite utilization rate is performed once. The single adjustment step size (0.5) is specifically designed for elite utilization rate, and the increment is determined in three segments based on the improvement magnitude: ; in, This represents the relative improvement in path cost for the m-th sub-ant colony within a continuous iteration window; The unified expression for updating elite utilization is: ; in, Let the elite utilization rate of the m-th sub-ant colony in t iterations be the clamping function. (·), used to limit utilization to a certain range. This is the lower limit of the allowed value for elite utilization (40% in this example). This represents the upper limit of the allowed value for elite utilization (80% in this example).

[0071] This mechanism allows elite utilization to dynamically change based on the actual search performance of the sub-ant colony: when a sub-ant colony shows significant improvement recently, indicating significant search progress, the increment is adjusted accordingly. To improve the utilization rate of elites and enhance their utilization and refined development, thereby accelerating local convergence; when When this occurs, it indicates insufficient improvement or a stagnant search; in this case, adjust the increment. This reduces the utilization rate of elites to broaden the space for independent exploration and encourages individuals to conduct independent searches in a wider solution space. Within the preset threshold range (2% to 5%), adjust the increment. To maintain the current elite utilization rate and ensure strategy stability, the strategy involves increasing utilization when the search is progressing well to accelerate convergence, and decreasing utilization when the search stalls to encourage exploration. This allows each sub-colony to adaptively adjust its exploration and utilization ratio according to its own search status.

[0072] Each sub-ant colony is based on elite utilization rate In each sub-ant colony Individuals are divided into three categories, with their numbers dynamically allocated: ; Among them, elite utilization type Used to quickly converge with global elites, elite-guided type. Used for semi-autonomous, semi-guided search, and autonomous exploration. Used to enhance global search. The classification weight is set to 0.5 in this embodiment.

[0073] The three types of individuals mentioned above each undertake different search functions: the elite-utilizing type accelerates local convergence, the elite-guiding type balances guidance and autonomy, and the autonomous-exploring type maintains population diversity. This enables the ant colonies at different search stages to achieve a dynamic balance between personalized exploration and utilization, effectively avoiding efficiency losses caused by globally uniform parameter settings.

[0074] Figure 7A bar chart showing the contribution of different sub-colonies to the optimization of the global optimal solution is presented, with the vertical axis representing the percentage of contribution. The differences in contribution confirm the effectiveness of the sub-colon-level adaptive elite utilization. The contribution of each sub-colonie ranges from 10% to 35%, with no single sub-colonie monopolizing the market or any sub-colonieie contributing zero contribution. This indicates that the parallel sub-colonie architecture is reasonably designed, the dynamic allocation of asynchronous communication resources is effective, and each sub-colonie can play a unique search role.

[0075] After the elite utilization rate is updated, the updated elite utilization rate will be fed back to the new solution generation step S103 to control the proportion of the three types of individuals in each sub-ant colony in the next round, thereby forming an adaptive control closed loop of "search performance evaluation, elite utilization rate adjustment, population structure update, and new solution generation".

[0076] In step S107, the current optimal path in the global elite pool is continuously monitored during the iteration process. In continuous domain ant colony optimization, traditional discrete ant colony algorithms rely on the concentration of the pheromone matrix to determine convergence. However, this invention uses a Gaussian distribution for sampling in continuous space, where there are no discrete pheromone paths, thus pheromone-related convergence metrics cannot be used. The actual convergence behavior is manifested as a Gaussian distribution variance. This means that the search radius continuously shrinks, the population gradually clusters around the current optimal solution, and ultimately the path cost function no longer improves significantly. Therefore, this invention uses the relative improvement rate of the cost function as a convergence surrogate observation, which has directness and order-of-magnitude adaptability.

[0077] Every K generations, using the global optimal cost from K generations ago as a benchmark, calculate the relative improvement rate of the current global optimal solution compared to K generations ago: ; Update the stall count according to the following rules : ; in, To improve the threshold, To ensure the minimum number of iterations, when simultaneously satisfying... , When the algorithm has fully converged, the iteration automatically terminates. This mechanism uses a maximum stall count. It can flexibly adjust the number of iterations based on the actual convergence state of the algorithm under different complexity scenarios, terminating promptly after the solution quality has sufficiently converged to avoid unnecessary computational resource consumption; simultaneously, it ensures that the algorithm does not stop prematurely before the solution has converged, balancing solution quality and computational efficiency.

[0078] After the algorithm terminates, the path code with the minimum cost is extracted from the global elite pool as the global optimal solution. The coordinate transformation function converts the coordinates into a Cartesian coordinate path control point sequence, completes the optimal path planning in three-dimensional continuous space, and outputs the final feasible path for the unmanned system to perform navigation tasks.

[0079] To further demonstrate the effectiveness of the present invention, simulation experiments were conducted in various scenarios in this embodiment. The specific experimental conditions and results are analyzed below.

[0080] Figure 2 This is a schematic diagram of the optimal path planning result in a three-dimensional continuous space scene. The experimental conditions are set as follows: the simulation space size is 1000m×1000m×600m, the planning area is undulating mountainous terrain, two cylindrical fixed obstacles are set up, and a low-altitude three-dimensional continuous path planning task from a single starting point to a single ending point is performed. Figure 3 The graph shows the evolution curves of local optimal path costs for five parallel sub-ant colonies over the last 10 generations, with the vertical axis representing path cost and the horizontal axis representing the iteration record number. As can be seen from the graph, the cost curves of each sub-ant colony show a continuous downward trend, indicating that each sub-ant colony possesses independent search capabilities and can continuously optimize local solutions within its own search space. The significant differences in the magnitude of the decrease and the inflection point positions among the sub-ant colonies reflect the regulatory role of asynchronous communication and the sub-ant colony-level adaptive elite utilization mechanism on the differentiated convergence rhythm of the sub-ant colonies. There are no long-term horizontal stagnation segments in any of the curves, thanks to the real-time high-quality guidance provided by the dynamic elite pool, the automatic increase in elite reading probability by stagnant sub-ant colonies, and the conditionally triggered fine-grained local search, effectively suppressing local optimum traps. In the later stages, the costs of each sub-ant colony gradually converge, indicating that the global elite pool achieves cross-sub-ant colony sharing of high-quality solutions, and the sub-ant colony search results can be quickly synchronized to the global optimum, driving the overall solution quality towards global optimum convergence.

[0081] Figure 4 This image shows the path planning results in a large-scale urban environment with dense obstacles. The experimental conditions were set at a simulation space of 1000m × 1000m × 350m, simulating a low-altitude urban environment. The area contains dozens of staggered, columnar obstacles, resulting in a dense obstacle distribution, a higher solution space dimension, and a significantly greater number of local optima compared to mountainous scenarios. The results demonstrate that even in a large-scale urban scenario with high complexity and high-density obstacles, the algorithm of this invention can still stably output a complete and smooth obstacle avoidance path without exhibiting failure phenomena such as population divergence or a sharp increase in path cost, proving the algorithm's feasibility.

[0082] Figure 5 Based on Figure 2 The convergence curve of the global optimal path cost in a 3D mountain scene is shown, with the horizontal axis representing the number of iterations and the vertical axis representing the global optimal path cost. The results show that the curve exhibits a rapid decline followed by a stable convergence characteristic, demonstrating the algorithm's advantages of fast convergence speed and good stability.

[0083] Figure 6The graph shows a comparison of the convergence performance of the proposed method and the traditional multi-ant colony algorithm. The horizontal axis represents the number of iterations, and the vertical axis represents the path cost. The control group used the traditional fixed synchronous migration multi-ant colony algorithm. The results show that the proposed method achieves the convergence level of the traditional algorithm (which takes more than 100 iterations) in about 40-50 iterations, improving the convergence speed by more than 50%. After convergence stabilizes, the global optimal cost of the proposed method is significantly lower than that of the traditional algorithm, directly demonstrating the significant improvement in convergence speed and solution quality of the proposed method.

[0084] Figure 7 This paper presents a bar chart illustrating the contribution of each parallel sub-ant colony to the optimization of the global optimal solution. The horizontal axis represents the numbers of the five independent parallel sub-ant colonies, and the vertical axis represents the number of times each sub-ant colony contributes to generating a new global optimal solution during the iteration process. The bar chart shows that the contribution of each sub-ant colony is distributed in the range of 13-55 times. There are no sub-ant colonies with zero contribution and no single sub-ant colony monopolizing the global optimization process. This verifies that the asynchronous communication and adaptive elite utilization mechanism of this invention can evenly distribute search computing power, and each sub-ant colony can independently produce high-quality elite solutions.

[0085] Example 2 This embodiment provides a multi-ant colony parallel path planning system based on a dynamic global elite pool, including: The initialization module is configured to: initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on the environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. The elite pool construction module is configured to: construct a global elite pool, generate initial path codes in the spherical coordinate parameter domain, and store them in the global elite pool after local optimization; The parallel path generation module is configured to: each sub-ant colony uses matrix-based batch computation to process the path encoding of the current population and generate new path encodings; The population update module is configured to: merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony. The communication module is configured to: when the sub-ant colony discovers that the cost of a path code is better than the cost of the current best path in the global elite pool, trigger communication and store the corresponding path code in the global elite pool. The adaptive control module is configured to: each sub-ant colony adjusts the elite utilization rate based on the local optimal cost, and dynamically divides its own population into different categories according to the elite utilization rate; The termination and output module is configured to: obtain the current optimal path in the global elite pool, calculate its relative improvement rate of cost every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

[0086] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation processes are the same, so they will not be repeated here.

[0087] Example 3 This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-ant colony parallel path planning method based on a dynamic global elite pool, characterized in that, include: Initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on the environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. Construct a global elite pool, generate an initial path code in the spherical coordinate parameter domain, and store it in the global elite pool after local optimization; Each sub-ant colony uses matrix-based batch computation to process the path code of the current population and generate a new path code; Merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony. When the ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, it triggers communication and stores the corresponding path code in the global elite pool. Each sub-ant colony adjusts its elite utilization rate based on the local optimal cost, and classifies its population dynamics into different categories according to the elite utilization rate. Obtain the current optimal path in the global elite pool, calculate its relative cost improvement rate every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

2. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, In the spherical coordinate parameter domain, each path code consists of three continuous variables: step size, pitch angle, and azimuth angle. The maximum value of the step size is determined by the distance between the starting point and the ending point and the number of path segments. The range of values ​​for the pitch angle and azimuth angle is determined by the direction vector from the starting point to the ending point.

3. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The local optimization adopts a greedy hill-climbing strategy: Gaussian random perturbation is applied to the initial path encoding on the three components of spherical coordinates. After constraint repair and cost evaluation, only the neighborhood solution with lower path cost is accepted, and the process is iterated for a preset number of times.

4. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The matrix-based batch computation includes: organizing the path codes of all individuals in the current population of each sub-ant colony into a step length matrix, a pitch angle matrix, and an azimuth angle matrix; calculating the step length matrix based on the difference between the current population and the parent population; and using the step length matrix to apply perturbation to the current path codes to generate new path codes in batches.

5. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The specific steps of selecting the best individuals based on path cost are as follows: merging the path codes in the current population with the new path codes, arranging them in ascending order of path cost, and retaining a predetermined number of individuals with the lowest path cost to update the population.

6. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The triggered communication includes: immediate communication when the sub-ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, or periodic communication when the sub-ant colony reaches a preset communication interval; after triggering communication, the corresponding path code is stored in the global elite pool, and the communication interval of the sub-ant colony is updated.

7. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The elite utilization rate is adaptively adjusted based on the recent improvement magnitude of the local optimal cost: when the improvement magnitude is greater than a first preset threshold, the elite utilization rate is increased; when the improvement magnitude is less than a second preset threshold, the elite utilization rate is decreased; otherwise, the current value is maintained, wherein the first preset threshold is greater than the second preset threshold; the proportion of different categories is dynamically allocated as the elite utilization rate changes, and the different categories include three types: elite utilization type, elite guidance type, and autonomous exploration type.

8. The multi-ant colony parallel path planning method based on a dynamic global elite pool according to claim 1, characterized in that, The preset termination condition is: the stall count reaches the preset maximum stall count and the current iteration count reaches the minimum iteration count.

9. A multi-ant colony parallel path planning system based on a dynamic global elite pool, characterized in that, include: The initialization module is configured to: initialize environmental parameters, construct a three-dimensional continuous domain parallel ant colony model based on the environmental parameters, determine the spherical coordinate parameter domain and path cost function, and divide the ant colony into multiple parallel sub-ant colonies. The elite pool construction module is configured to: construct a global elite pool, generate an initial path code in the spherical coordinate parameter domain, and store it in the global elite pool after local optimization; The parallel path generation module is configured to: each sub-ant colony uses matrix-based batch computation to process the path encoding of the current population and generate new path encodings; The population update module is configured to: merge the path codes in the current population with the new path codes, select the best path codes based on path cost to update the population, and record the local optimal cost of each sub-ant colony; The communication module is configured to: when the sub-ant colony discovers that the cost of the path code is better than the cost of the current best path in the global elite pool, trigger communication and store the corresponding path code in the global elite pool. The adaptive control module is configured to: each sub-ant colony adjusts the elite utilization rate according to the local optimal cost, and dynamically divides its own population into different categories according to the elite utilization rate; The termination and output module is configured to: obtain the current optimal path in the global elite pool, calculate its relative cost improvement rate every preset period to update the stagnation count, and when the stagnation count and the number of iterations meet the preset termination condition, extract the path code with the minimum cost from the global elite pool and output it.

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 steps in the multi-ant colony parallel path planning method based on a dynamic global elite pool as described in any one of claims 1-8.

Images