A heterogeneous unmanned cluster multi-target task pre-allocation method and system

Abstract

The application discloses a heterogeneous unmanned cluster multi-target task pre-allocation method and system, and belongs to the technical field of unmanned systems; the method establishes constraint conditions containing time sequence, capability, resources and endurance and an evaluation model with time, energy consumption and risk as targets; after initializing a particle swarm, a reference vector is constructed in a target space, a space included angle is normalized, an angle score is calculated, a comprehensive priority is calculated by combining a non-branch level and a uniformity factor, and the particle is dynamically divided into an elite layer, an exploration layer and a following layer; local search is performed on the elite layer, the following layer is quickly converged, and a shock feedback mechanism simulating the behavior of a myxomycete vein network is introduced into the exploration layer; adaptive myxomycete weights are calculated by taking the normalized angle score as a quality index to disturb and update an improved social learning part, and a Pareto scheme set is iteratively output after feasibility correction and archive maintenance; the application improves the distribution quality of a candidate scheme in a target space, enhances the ability of the algorithm to jump out of a local optimum, and improves search efficiency.

Description

Technical Field

[0001] This invention relates to the field of unmanned systems technology, specifically to a method and system for allocating multi-target tasks in a heterogeneous unmanned swarm. Background Technology

[0002] As unmanned systems are increasingly used in reconnaissance and surveillance, target search, continuous tracking, disaster relief, and emergency support, individual unmanned systems, limited by payload, endurance, mission capabilities, and risk tolerance, struggle to independently complete large-scale, multi-target, and multi-stage missions. Heterogeneous unmanned swarms composed of multiple types of unmanned platforms can accomplish more complex collaborative tasks through complementary capabilities.

[0003] In typical mission scenarios, a target point usually requires more than just one task; it necessitates multiple sub-tasks such as reconnaissance, search, and tracking. The search task is used to locate or pinpoint the target, the reconnaissance task is used to confirm the target's status, and the tracking task is used to continuously monitor changes in the target's situation. The order of these sub-tasks can be configured according to mission requirements; for example, reconnaissance cannot proceed until the search is complete, and tracking cannot proceed until reconnaissance is complete.

[0004] The unmanned systems participating in the mission are not entirely the same. Some unmanned platforms only have reconnaissance capabilities, some mainly have search capabilities, some are suitable for continuous tracking missions, and some have multiple mission capabilities. The speed, payload capacity, remaining endurance, platform value, and risk tolerance of different unmanned platforms also vary. Therefore, before the mission begins, an initial mission allocation scheme needs to be generated for the heterogeneous unmanned swarm by ground control stations, mission planning servers, or edge computing terminals.

[0005] This initial task allocation scheme typically needs to determine the following:

[0006] Which unmanned platform will perform the reconnaissance, search, and tracking tasks at each target point?

[0007] In what order should each unmanned platform execute its tasks?

[0008] When are the tasks expected to begin and end?

[0009] Does each unmanned platform meet the payload, resource, and endurance constraints after executing its mission?

[0010] The current plan includes the task completion time, energy consumption, and overall risks.

[0011] Existing methods for task allocation in unmanned swarms mainly include mixed integer programming, genetic algorithms, particle swarm optimization, multiverse optimization, and snow goose optimization. Among these, swarm intelligence algorithms have been used for task allocation in unmanned swarms due to their relatively simple structure and suitability for handling complex nonlinear problems.

[0012] Existing technologies include task allocation methods for unmanned systems based on improved snow goose optimization algorithms. These methods typically establish a task allocation model based on the starting point and target point of the unmanned platform, with the longest single-platform movement distance and total movement distance as the main indicators, and obtain the final allocation scheme through iterative processing using improved snow goose optimization algorithms.

[0013] Existing technologies include heterogeneous unmanned swarm task allocation methods based on improved multiverse optimization algorithms. These methods establish a task allocation objective function based on the unmanned platform's capabilities, target value, target threat level, and movement distance, and improve the algorithm's search capability through strategies such as chaotic initialization and differential mutation.

[0014] Existing task allocation methods mainly include market mechanisms, mathematical programming, game theory, learning methods, and swarm intelligence optimization methods. Among them, swarm intelligence optimization methods have been widely used for task allocation in unmanned swarms due to their flexibility and strong adaptability to complex nonlinear problems. Research has shown that these methods can establish relatively complete task allocation models by combining factors such as unmanned platform capabilities, target attributes, task rewards, and movement distance, generating executable task solutions. However, in complex heterogeneous unmanned swarm task allocation scenarios, existing swarm intelligence methods still have algorithmic shortcomings: their search process often relies on the guidance of the current best individual or the historical best individual, and the population is prone to prematurely concentrating in locally superior areas, leading to decreased diversity and premature convergence; as the number of unmanned platforms and tasks increases, the search space expands rapidly, and the algorithm's convergence speed and stability tend to decrease; in multi-objective task allocation, candidate solutions may also be concentrated in local regions of the target space, with insufficient coverage of boundary and sparse regions, resulting in uneven distribution of output solutions; simultaneously, some algorithms use similar update rules for individuals of different quality, failing to adequately distinguish the roles of high-quality, ordinary, and low-quality individuals in the search, and there is still room for improvement in local perturbation and escape capabilities. Therefore, a new method for multi-objective task allocation for heterogeneous unmanned swarms is needed, which can improve population diversity, enhance the ability to escape local optima, and improve the distribution quality of candidate task schemes in the target space while maintaining the ease of implementation of swarm intelligence algorithms. Summary of the Invention

[0015] To address the aforementioned issues, this invention discloses a multi-objective task pre-assignment system and method for heterogeneous unmanned swarms.

[0016] The specific technical solution is as follows, mainly including the following technical aspects:

[0017] 1. Task constraint construction and multi-objective evaluation model establishment

[0018] The system acquires information on the unmanned platform, target mission, and environmental threats, and establishes mission constraints including mission timing, platform capability, resource capacity, and maximum endurance. Simultaneously, it constructs a multi-objective evaluation model with mission completion time, energy consumption, and overall risk as optimization objectives. The overall risk index considers the threat level of the target area, mission type risk, and the unmanned platform's own value coefficient to guide the algorithm and avoid high-risk allocations.

[0019] 2. Particle Swarm Initialization and Evaluation

[0020] Each particle is encoded as a complete task pre-assignment scheme, containing task affiliation and execution sequence information to express the sequential relationship between successors and successors. The particle swarm is initialized and a feasibility check is performed based on the constraints. Subsequently, the multi-objective evaluation value of each particle is calculated, and non-dominated solutions are selected based on dominance relationships to establish an initial external file.

[0021] 3. Population dynamic stratification based on reference vector

[0022] Multiple reference vectors are constructed in a three-dimensional target space consisting of task completion time, energy consumption, and comprehensive risks. Each reference direction corresponds to a search region, which is used to guide the particle swarm to search simultaneously in different regions.

[0023] The system calculates the spatial angle between each particle and each reference vector, and normalizes this spatial angle to obtain an angle score. Combining the rank information obtained from non-dominated sorting of the population with a uniformity factor reflecting the current population distribution, a comprehensive priority score is calculated for each particle. Based on this comprehensive priority score, the particle swarm is dynamically divided into three subgroups with different search responsibilities: the top-ranked particles are assigned to the elite layer, the bottom-ranked particles to the follower layer, and the rest to the exploration layer.

[0024] 4. Differentiated updates based on slime mold oscillation feedback

[0025] Implement differentiated update strategies for particles at different levels:

[0026] Elite layer: Retain its advantageous search direction and allow it to conduct localized, refined development in the vicinity of the current high-quality solution;

[0027] Follower layer: Abandon the original flight direction and enhance its ability to converge quickly towards high-quality areas;

[0028] Exploration Layer: An oscillating feedback mechanism simulating the contraction and expansion behavior of the venous network during slime mold foraging is introduced for position updates. Specifically, an oscillation factor that decreases nonlinearly with the number of iterations and a random oscillation parameter simulating venous contraction are constructed. The normalized angle score of individual particles is used as a quality evaluation index, and an adaptive slime mold weight is dynamically calculated by combining the best and worst scores of the population. While particles in the exploration layer retain their inertial flight direction, this adaptive slime mold weight is used to improve the social learning part of the particle swarm optimization algorithm, causing particles to generate oscillating perturbations as they approach the global optimum, thereby maintaining population diversity and enhancing the ability to escape local optima.

[0029] 5. Feasibility revision and iterative output of archives

[0030] The updated particles undergo another feasibility check and correction to ensure they meet constraints related to task capability matching, resource and endurance limits, and task timing. Subsequently, the external archive used to store non-dominated solutions is updated; when the archive capacity exceeds the limit, solutions from sparsely distributed regions in the target space are prioritized. This dynamic hierarchical, differentiated update, and feasibility correction process is iteratively executed until the termination condition is met, outputting the Pareto task pre-allocated solution set.

[0031] Furthermore, the present invention also provides a heterogeneous unmanned swarm multi-objective task pre-assignment system, which achieves task pre-assignment through the coordinated operation of the following functional modules:

[0032] 1. Task Constraints and Evaluation Module: Used to acquire basic data, establish time, capability, resource and endurance constraints, and calculate the task completion time, energy consumption and overall risk of the particle scheme.

[0033] 2. Reference Vector Dynamic Hierarchy Module: This module is used to construct reference vectors in the target space, calculate the spatial angle between the particle and the reference vector and normalize it to obtain an angle score, combine the non-dominated ranking level and the population evenness factor to calculate a comprehensive priority score, and dynamically divide the particle swarm into three subgroups—elite layer, exploration layer and follower layer—to undertake different search responsibilities.

[0034] 3. Slime Mold Oscillation Feedback Update Module: This module performs differentiated position updates on the stratified particles. For particles in the exploration layer, a non-linearly decreasing oscillation factor and random oscillation parameters are constructed. Adaptive slime mold weights are generated using the normalized angle score of the particles as a quality evaluation index. These weights are then incorporated into the velocity update formula of the particle swarm algorithm to improve the social learning part. This simulates the contraction and expansion behavior of the slime mold vein network during the algorithm's search process, achieving perturbation search. Simultaneously, the elite layer is controlled to perform local fine-grained search, while the follower layer performs rapid convergence updates.

[0035] 4. Feasibility and Archive Maintenance Module: This module is used to convert task schemes and particles to each other, ensure that the updated particles meet all actual constraints, and maintain and update non-dominated solutions in the external archive based on the distribution sparsity.

[0036] 5. Task Scheme Output Module: This module converts the non-dominated solution set in the final external archive into a readable task pre-assignment scheme output.

[0037] The advantages of this invention are: 1. This invention improves the distribution quality of candidate solutions in the target space through the dynamic hierarchical mechanism of reference vectors, making the Pareto solution set output by the algorithm have a wider coverage and more uniform distribution, which is conducive to obtaining multiple types of solutions such as time priority, energy priority, risk priority and comprehensive balance.

[0038] 2. This invention enhances the ability of the particle swarm optimization algorithm to escape local optima through the slime mold oscillation feedback mechanism, which can alleviate the problem of premature convergence of the traditional multi-objective particle swarm optimization algorithm in complex combinatorial optimization problems.

[0039] 3. This invention adopts a differentiated update method for particles of different masses, so that high-quality particles, medium-quality particles and low-quality particles play different roles in the search process, thereby improving the search efficiency and convergence of the algorithm. Attached Figure Description

[0040] Figure 1 Flowchart for assigning core search strategies to tasks;

[0041] Figure 2 The overall structure of the task allocation system. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] This invention provides a multi-target task pre-assignment system and method for heterogeneous unmanned swarms. The system can be deployed in a ground control station, a mission planning server, or an edge computing node of the unmanned swarm, and is used to generate an initial task assignment scheme for heterogeneous unmanned swarm collaboration before task execution. System inputs include unmanned platform information, target task information, task constraint information, and environmental threat information; system outputs include the task execution sequence of each unmanned platform, the task assignment for each target point (reconnaissance, search, tracking, etc.), the estimated task execution time, the task completion time corresponding to the scheme, energy consumption, comprehensive risk, and a multi-target Pareto candidate scheme set.

[0044] The system described in this invention mainly includes: a task information input module, a task constraint modeling module, a multi-objective evaluation module, a particle encoding and decoding module, a reference vector dynamic hierarchical module, a slime mold oscillation feedback update module, a feasibility check and correction module, an external file maintenance module, and a task plan output module.

[0045] Task information input module

[0046] The task information input module is used to receive and organize the basic data required for task pre-assignment, including unmanned platform information, target information, task type information, and environmental threat information.

[0047] The unmanned platform information includes the platform number, platform type, current location or starting point, speed, maximum endurance, payload capacity, remaining resources, platform value coefficient, and type of mission it can perform. Based on mission capabilities, unmanned platforms can be categorized into reconnaissance platforms, search platforms, tracking platforms, and integrated platforms. Reconnaissance platforms primarily perform reconnaissance missions; search platforms primarily perform search missions; tracking platforms primarily perform continuous tracking missions; and integrated platforms can perform multiple tasks including reconnaissance, search, and tracking.

[0048] Define a collection of heterogeneous unmanned platforms Set of task objective points Each subtask consists of reconnaissance (r), search (s), and tracking (e). At the mission point The attributes on are defined as follows: , Let j be the coordinates of the task point. The amount of resources required to execute this subtask; The reward value obtained for completing this task point; The environmental threat level of this mission location; The execution time required for this type of subtask.

[0049] The target information includes the target number, target spatial location, target value, target threat level, and the resources required for each type of sub-task. Each target point can contain sub-tasks such as reconnaissance, search, and tracking, and the related sub-tasks can be set to a fixed execution order according to task requirements. For example, for the same target point, the search task can be executed first, followed by the reconnaissance task, and finally the tracking task.

[0050] The environmental threat information includes the threat level around the target point, the regional risk coefficient, or other parameters that can reflect the risks of the unmanned platform performing its mission. This information is used to calculate the overall risk cost when the unmanned platform enters different target areas to perform its mission.

[0051] Task constraint modeling module

[0052] The task constraint modeling module is used to construct pre-assigned constraints for tasks based on input information. This module mainly includes task timing constraints, unmanned platform capability constraints, resource capacity constraints, and maximum endurance constraints.

[0053] The task timing constraints are used to ensure that reconnaissance, search, and tracking tasks with successive relationships at the same target point are executed in a prescribed order. For example, if a search task is not completed, subsequent reconnaissance tasks at that target point cannot begin; if a reconnaissance task is not completed, tracking tasks at that target point cannot begin.

[0054] The unmanned platform capability constraints are used to ensure that tasks can only be assigned to unmanned platforms with the corresponding capabilities. For example, reconnaissance tasks can only be performed by unmanned platforms with reconnaissance capabilities, search tasks can only be performed by unmanned platforms with search capabilities, and tracking tasks can only be performed by unmanned platforms with tracking capabilities.

[0055]

[0056] The resource capacity constraint is used to ensure that the resources consumed by the unmanned platform in performing its tasks do not exceed its own carrying capacity or remaining resource capacity. Different types of sub-tasks may correspond to different resource consumption; for example, reconnaissance tasks consume reconnaissance payload resources, search tasks consume search payload resources, and tracking tasks consume continuous tracking resources.

[0057]

[0058] The maximum range constraint ensures that the total distance traveled by the unmanned platform after completing its mission sequence does not exceed its maximum available range. This constraint can be further included as a return distance constraint, depending on the specific application requirements, to ensure that the unmanned platform can return to its base or designated recovery point after completing its mission.

[0059]

[0060] Multi-objective evaluation module

[0061] The multi-objective evaluation module is used to comprehensively evaluate each candidate task allocation scheme. This invention selects task completion time, energy consumption, and overall risk as the main optimization objectives.

[0062] The task completion time measures the total time required for the entire heterogeneous unmanned swarm to complete all tasks. This metric is determined by the movement time required for each unmanned platform to execute its task, the task operation time, and the waiting time caused by the successive relationships between tasks. The shorter the task completion time, the higher the task execution efficiency.

[0063]

[0064] The energy consumption figure measures the motion cost incurred during the execution of tasks by a heterogeneous unmanned swarm. This indicator can be calculated based on the unmanned platform's movement distance and energy consumption coefficient per unit distance, or it can be calculated separately based on the energy consumption models of different unmanned platforms. Lower energy consumption indicates a more rational use of energy resources in the task allocation scheme.

[0065]

[0066] The comprehensive risk is used to measure the level of risk faced by unmanned platforms during mission execution. This indicator considers the threat level of the target area, the risk of the mission type, and the value coefficient of the unmanned platform itself. High-value unmanned platforms incur higher risk costs when entering high-threat areas; therefore, this indicator guides the algorithm to avoid unnecessary high-risk mission allocation.

[0067]

[0068] Through these three objectives, the present invention can generate task allocation schemes with different preferences. For example, the time-priority scheme tends to increase the degree of parallel execution, the energy-priority scheme tends to reduce ineffective movement and long-distance transfers, and the risk-priority scheme tends to reduce the probability of high-value unmanned platforms entering high-threat areas.

[0069] Particle encoding and decoding module

[0070] The particle encoding and decoding module is used to convert the task allocation scheme into an individual form that can be processed by the particle swarm algorithm, and to restore the particles after the algorithm update to a specific task execution scheme.

[0071] During the encoding process, a particle represents a complete task pre-assignment scheme. The particle contains the unmanned platform number for each target point's sub-tasks such as reconnaissance, search, and tracking, as well as the task execution order within each unmanned platform. To ensure the encoding can express the sequential relationship between tasks, the particle stores not only task affiliation information but also task execution sequence information.

[0072] During the decoding process, the system recovers the mission paths of each unmanned platform based on the particle encoding, and calculates the estimated start and end times of each sub-task in sequence.

[0073] Reference Vector Dynamic Hierarchy Module

[0074] The reference vector dynamic hierarchical module is used to improve the problems of uneven solution set distribution and insufficient population diversity in traditional multi-objective particle swarm optimization algorithms.

[0075] This module first constructs multiple reference directions within a three-dimensional target space comprised of task completion time, energy consumption, and overall risk. Each reference direction corresponds to a search region within the target space, guiding the particle swarm to search simultaneously within different regions.

[0076] Let the dimension of the target space be... The number of divisions in each dimension is Then the total number of reference vectors The calculation formula is:

[0077]

[0078] For any reference vector in three-dimensional space The components are:

[0079]

[0080] in Corresponding to the objective function directional components; Ensure that it is a unit vector; Polar angle; This is the azimuth angle.

[0081] Subsequently, the system performs a comprehensive score on the particles based on their matching degree with the reference direction, their non-dominated level, and the current distribution of the population in the target space. The comprehensive score considers both the optimization quality of the particle itself and its contribution to the coverage of the target space.

[0082] Let the th in the current population be The objective function vector for each particle is: Calculating particles and the reference vectors The spatial angle between :

[0083] The smaller the size, the more likely the particle is to be smaller. Evolutionary direction and reference vector The more consistent the values, the greater their contribution to the distribution of the region. All particles are sorted in ascending order of their minimum included angles, and the angle scores are obtained after normalization. .

[0084] Non-dominated ranking of the population shows that particles at lower ranks have better convergence. Fitness scores are obtained after normalizing the rankings. Normalization is performed using the range method:

[0085]

[0086] To adjust the weights of convergence and distribution, a population evenness factor is introduced. In three-dimensional space, this factor is calculated based on the Euclidean distance between particles:

[0087]

[0088] Among them, the interparticle distance :

[0089]

[0090] Space diagonal Used to eliminate the influence of dimensions:

[0091]

[0092] Based on the above calculations, the first The overall priority score of each particle Defined as:

[0093]

[0094] Based on the comprehensive scoring results, the system divides the particle population into three subgroups: the elite layer, the exploration layer, and the follower layer.

[0095] The elite layer consists of particles with higher current quality and is mainly responsible for performing fine-grained searches near high-quality regions to improve the convergence accuracy of the solution set.

[0096] The exploration layer consists of medium-mass particles, which are mainly responsible for expanding the search range, maintaining population diversity, and jumping out of local superior areas through an oscillating feedback mechanism.

[0097] The following layer consists of low-mass particles and is mainly responsible for quickly converging towards high-quality regions to improve overall convergence efficiency.

[0098] Through the above dynamic layering, different particles undertake different search responsibilities during the algorithm iteration process, thereby avoiding the premature convergence problem caused by all particles using the same update rule.

[0099] slime mold oscillation feedback update module

[0100] The slime mold oscillation feedback update module is used to enhance the local optimum escape capability of the particle swarm algorithm. This module is inspired by the contraction and expansion behavior of the venous network during slime mold foraging.

[0101] During natural foraging, slime molds adjust their network channels according to food quality. When food quality is high, the relevant channels are enhanced; when food quality is low, the channels are weakened, prompting the slime molds to explore other areas. This invention introduces this idea into the particle update process, so that when particles approach a high-quality task allocation scheme, they do not simply converge linearly, but retain a certain range of oscillating perturbations.

[0102] Constructing a nonlinearly decreasing oscillation factor , .in This represents the current iteration number. This represents the maximum number of iterations. To simulate the randomness of expansion and contraction of myxomyophyte veins in different directions, random oscillation parameters simulating vein contraction are constructed: In the calculation formula of the slime mold algorithm, the weights... The thickness of veins was simulated. If the food source quality is high, the veins become thicker, and vice versa. Based on the previous ranking results, a comprehensive score was used as the "quality evaluation index" in the SMA weighting formula. Improved SMA weighting. :

[0103]

[0104] in Adaptive slime mold weights; The overall score for particle i; The optimal overall score for the population; This is the worst overall score for the population. The random oscillation factor has a value range of [0,1]. Rank the particles; Population size.

[0105] Will When integrated into the task allocation algorithm of MOPSO, the particle update formula after fusion is:

[0106] 1) Subpopulation A ( )

[0107] The particle must belong to the top 50%, therefore A larger weighting can maintain momentum in the advantageous direction and allow for more refined exploration in specific areas.

[0108]

[0109] 2) Subpopulation B ( )

[0110] subpopulation The upper and lower halves of the particle array each cover 50% of the total, balancing global exploration and local exploitation. Furthermore, a myxobacterial vein oscillation and contraction mode is introduced to improve the "social learning part" of the traditional particle swarm optimization algorithm. The improved formula is as follows:

[0111]

[0112] 3) Subpopulation C ( )

[0113] Subpopulation C is generally in the disadvantaged region, and its primary goal is to converge quickly, since it inevitably belongs to the bottom 50%. And it is relatively small. Smaller inertia means that the particle almost completely abandons its previous flight direction and instead leans more towards... Learning. The formula is:

[0114]

[0115] Specifically, for elite-level particles, the system retains their advantageous search direction, allowing them to conduct local exploration near the current high-quality solution; for exploratory-level particles, the system introduces an oscillation factor that changes with the iteration process, causing them to generate a certain range of perturbations while approaching the global high-quality solution; for follower-level particles, the system enhances their ability to move towards high-quality regions, enabling low-mass particles to quickly participate in effective searches.

[0116] Through this mechanism, the algorithm maintains strong global exploration capabilities in the early stages and gradually shifts to refined local searches in the later stages. Simultaneously, oscillatory feedback reduces the probability of particle swarms prematurely clustering in local regions, increasing the likelihood of obtaining high-quality Pareto solutions.

[0117] Feasibility check and correction module

[0118] The feasibility check and correction module is used to ensure that the task allocation scheme after each particle update still meets the actual constraints.

[0119] After particle updates, issues may arise such as tasks being assigned to unmanned platforms that lack the necessary capabilities, unmanned platforms exceeding resource limits or endurance limits, or invalid task timing. This module performs checks on the updated particles sequentially.

[0120] If a mismatch between mission capabilities and requirements is found, the mission will be reassigned to an unmanned platform with the corresponding capabilities and that meets the resource conditions. If the resource consumption of an unmanned platform is found to be exceeding the limit, the mission sequence of that unmanned platform will be adjusted, and some missions will be transferred to other feasible unmanned platforms. If the endurance limit is found to be exceeded, the mission insertion position will be recalculated or the executing unmanned platform will be changed. If a mission timing conflict is found, the mission start time will be rearranged to ensure that reconnaissance, search, tracking and other missions are performed in the correct order.

[0121] Through feasibility checks and corrections, the system can reduce the impact of invalid candidate solutions on algorithm iteration, improve search efficiency, and ensure that the final output solution is engineering-executable.

[0122] External file maintenance module

[0123] The external archive maintenance module is used to save the non-dominant task allocation scheme obtained during the search process.

[0124] In multi-objective optimization, there may not be an absolute superiority or inferiority relationship between different solutions. For example, one solution may have a shorter task completion time but higher risk, while another solution may have lower risk but longer completion time. In this case, both can be retained as candidate solutions.

[0125] This invention stores currently obtained non-dominated solutions in an external archive. When a new solution is superior to existing solutions in the archive, the system adds it to the archive and deletes the dominated solution. When the archive capacity exceeds the limit, the system prioritizes retaining solutions from sparsely distributed regions in the target space and deletes redundant solutions from overly dense regions. In this way, the external archive can simultaneously maintain the convergence and distribution of the solution set.

[0126] Task plan output module

[0127] The task scheme output module is used to convert the non-dominated solutions in the final external archive into readable task pre-assignment results.

[0128] The output includes:

[0129] Which unmanned platform will perform the reconnaissance, search, and tracking tasks at each target point?

[0130] The task execution order of each unmanned platform;

[0131] The estimated start and end times for each subtask;

[0132] The task completion time, energy consumption, and overall risk corresponding to each candidate solution;

[0133] Recommended solutions include those prioritizing time, energy consumption, risk, and overall balance.

[0134] Commanders can select one of the options as the initial execution plan based on the actual task requirements, or input the plan into the subsequent task reassignment module or path planning module.

Claims

1. A method for pre-assigning multi-objective tasks in a heterogeneous unmanned swarm, characterized in that, Includes the following steps: Step 1: Obtain information on the unmanned platform, target mission, and environmental threats, and establish a basic dataset for mission pre-assignment; Step 2: Based on the unmanned platform's mission capabilities, resource capacity, maximum endurance, and the sequential relationship between missions, construct mission constraints; Step 3: Construct a multi-objective evaluation model that includes task completion time, energy consumption, and overall risk; Step 4: Initialize the particle swarm, with each particle corresponding to a complete heterogeneous unmanned swarm task allocation scheme, and perform a feasibility check on the initial particles; Step 5: Calculate the multi-objective evaluation value of each particle, and screen non-dominated solutions according to the dominance relationship to establish the initial external file; Step 6: Construct a reference vector in the target space, calculate the comprehensive priority based on the matching degree between the particle and the reference vector, the non-dominance level, and the population distribution state, and dynamically divide the particle population into an elite layer, an exploration layer, and a follower layer. Step 7: Implement a differentiated update strategy based on the particle level: perform a local fine search for the elite layer, introduce slime mold oscillation feedback for perturbation search for the exploration layer, and enhance the ability of the follower layer to move closer to high-quality areas. Step 8: Perform a feasibility check on the updated particle's ability to perform tasks, resources, endurance, and timing, and correct any parts that do not meet the constraints; Step 9: Recalculate the particle target value, update the external archive, and retain a representative scheme based on the distribution sparsity when the archive capacity exceeds the limit; Step 10: Determine if the termination condition is met. If it is, output the Pareto task pre-allocation scheme set in the external file.

2. The method for pre-assigning multi-objective tasks in heterogeneous unmanned swarms according to claim 1, characterized in that, The task constraints in step 2 include: Task timing constraints: Ensure that reconnaissance, search, and tracking sub-tasks with successive relationships at the same target point are executed in the prescribed order; Unmanned platform capability constraints: Ensure that tasks can only be assigned to unmanned platforms with the corresponding capabilities; Resource capacity constraint: Ensure that the resources consumed by the unmanned platform in performing tasks do not exceed its remaining resource capacity; Maximum range constraint: Ensure that the total distance traveled by the unmanned platform after completing its mission sequence does not exceed its maximum available range.

3. The method for pre-assigning multi-objective tasks in heterogeneous unmanned swarms according to claim 1, characterized in that, In step 3: The task completion time is determined by the movement time required for each unmanned platform to perform the task, the task operation time, and the waiting time caused by the succession of tasks. The energy consumption is calculated based on the unmanned platform's travel distance and energy consumption coefficient per unit distance. The overall risk is calculated based on the threat level of the target area, the risk of the mission type, and the value coefficient of the unmanned platform itself.

4. The method for pre-assigning multi-objective tasks in heterogeneous unmanned swarms according to claim 1, characterized in that, The specific process of calculating the overall priority and dividing the population in step 6 is as follows: Multiple reference directions are constructed in the three-dimensional target space, and the spatial angle between the particle and the reference directions is calculated. The population is non-dominated and ranked, and the ranking is normalized to obtain the fitness score; A population evenness factor based on Euclidean distance between particles is introduced, and the comprehensive priority score of particles is calculated by combining the fitness score. Based on the overall priority score, high-quality particles are classified into the elite layer, medium-quality particles into the exploration layer, and low-quality particles into the follower layer.

5. The method for pre-assigning multi-objective tasks in heterogeneous unmanned swarms according to claim 1 or 4, characterized in that, The differential update strategy in step 7 is as follows: For elite particles, their advantageous search direction is retained, and a larger inertial weight is used for localized refined exploration; For particles in the exploration layer, a nonlinearly decreasing oscillation factor and random oscillation parameter are constructed, and an adaptive slime mold weight is calculated by combining the particle comprehensive score. This improves the social learning part of the particle swarm algorithm, so that the particles generate oscillation perturbations while approaching the high-quality solution. For particles in the following layer, a smaller inertial weight is used to make them abandon their previous flight direction and quickly move towards the global high-quality region.

6. The method for pre-assigning multi-objective tasks in a heterogeneous unmanned swarm according to claim 1, characterized in that, The feasibility check and correction in step 8 include: If the task capabilities do not match, the task will be reassigned to an unmanned platform that has the corresponding capabilities and meets the resource requirements. If the unmanned platform's resources exceed the limit, the task sequence of the unmanned platform will be adjusted and some tasks will be transferred to other feasible unmanned platforms; If the battery life exceeds the limit, the mission insertion position will be recalculated or the unmanned platform will be replaced. If there is a time conflict between tasks, the start time of the tasks will be rearranged so that the subtasks with a successor-successor relationship satisfy the order.

7. The method for pre-assigning multi-objective tasks in heterogeneous unmanned swarms according to claim 1, characterized in that, The method for updating external files in step 9 is as follows: when a new solution is better than an existing solution in the file, add it to the file and delete the dominated solution; when the file capacity exceeds the limit, prioritize retaining solutions in sparsely distributed areas of the target space and delete redundant solutions in dense areas.

8. A heterogeneous unmanned swarm multi-objective task pre-assignment system, characterized in that, include: The task information input module is used to receive information about the unmanned platform, target information, task type information, and environmental threat information. The task constraint modeling module is used to construct task timing constraints, unmanned platform capability constraints, resource capacity constraints, and maximum endurance constraints. The multi-objective evaluation module is used to evaluate the task completion time, energy consumption, and overall risk of candidate solutions; The particle encoding and decoding module is used to convert between task allocation schemes and particle forms; The reference vector dynamic hierarchical module is used to calculate the comprehensive priority based on the degree of matching between the particle and the reference vector, the non-dominance level, and the population distribution state, and to divide the particle into an elite layer, an exploration layer, and a follower layer. The slime mold oscillation feedback update module is used to perform differentiated updates and oscillation disturbance searches based on particle levels. The feasibility check and correction module is used to perform constraint checks and corrections on the updated particles. The external archive maintenance module is used to store non-dominated solutions and maintain the archive according to the degree of distribution sparsity; The task scheme output module is used to output Pareto task pre-assignment scheme sets.

9. The heterogeneous unmanned swarm multi-objective task pre-assignment system according to claim 8, characterized in that, In the reference vector dynamic hierarchical module, the comprehensive priority score is calculated by combining the fitness score after non-dominated sorting normalization with the population evenness factor based on the Euclidean distance between particles.

10. The heterogeneous unmanned swarm multi-objective task pre-assignment system according to claim 8, characterized in that, The Pareto task pre-allocation scheme set output by the task scheme output module includes recommended schemes based on time priority, energy consumption priority, risk priority, and comprehensive balance, as well as the task execution order and estimated time for each unmanned platform.

Images