Methods, systems, equipment, and media for planning heterogeneous unmanned swarm air-ground collaborative missions.

Abstract

This invention relates to the field of unmanned swarm technology, and particularly to a method, system, device, and medium for planning air-ground cooperative tasks in heterogeneous unmanned swarms. The method includes: a control device constructing a multi-objective optimization-oriented air-ground cooperative task planning model based on the parameters of the unmanned aerial vehicles (UAVs) and unmanned vehicles (UAVs) in the heterogeneous unmanned swarm; employing a pre-corrected discrete multi-objective sheep migration optimization algorithm to solve the multi-objective function and problem constraints, obtaining multiple non-dominated solutions; the discrete multi-objective sheep migration optimization algorithm is an algorithm based on the multi-objective optimization air-ground cooperative task planning model and the attribute acquisition of the unmanned swarm; performing a multi-objective comprehensive scoring on each non-dominated solution, and determining the heterogeneous unmanned swarm air-ground cooperative task planning scheme based on the scoring results and the air-ground cooperative task scenario. This method solves the problems of low coding efficiency, weak search capability, and difficulty in constraint handling in the heterogeneous intelligent unmanned swarm cooperative task planning problem.

Description

Technical Field

[0001] This invention relates to the field of unmanned swarm technology, and in particular to a method, system, device and medium for planning air-ground collaborative tasks of heterogeneous unmanned swarms. Background Technology

[0002] With the rapid development of intelligent technologies, heterogeneous intelligent unmanned swarm systems have shown broad application prospects in reconnaissance and surveillance, precision strikes, area control, and battlefield support. For air-ground collaborative mission planning scenarios involving heterogeneous intelligent unmanned swarms, the system consists of two heterogeneous platforms: airborne unmanned equipment and ground-based unmanned equipment. These two types of platforms differ significantly in mobility, payload capacity, endurance, and survivability. Airborne unmanned equipment possesses advantages in rapid penetration and wide-area perception, but has limited payload and weak sustained combat capability. Ground-based unmanned equipment has strong payload capacity and long battlefield loiter time, but relatively slow mobility and is significantly constrained by terrain. Due to the significant differences in performance characteristics and mission adaptability between airborne and ground-based unmanned equipment, only through reasonable collaborative planning and mission coordination can the complementary advantages of the two platforms be fully utilized.

[0003] The heterogeneous intelligent unmanned swarm collaborative task planning problem is a complex combinatorial optimization problem that requires simultaneous optimization of two aspects: first, task assignment, i.e., determining which unmanned equipment will perform each task; and second, execution order, i.e., determining the order in which each unmanned equipment performs tasks. This problem has the following characteristics: a discrete decision space, multi-objective optimization, multiple constraints such as load constraints, time window constraints, task type constraints, and a large-scale search space.

[0004] Currently, research on this type of problem mainly employs two coding schemes: (1) Two-layer coding scheme: The task attribution and execution order are encoded into two independent vectors, such as the attribution vector. Indicates which unmanned equipment is assigned the task, the order vector. This indicates the order in which tasks are executed. The drawback of this approach is that the two encoding layers are independent, making it difficult for the optimization operator to simultaneously optimize both attribution and order, resulting in low search efficiency.

[0005] (2) Single-layer permutation coding scheme: All tasks are encoded into a permutation sequence, and the sequence is divided into multiple sub-sequences of unmanned equipment by delimiters or rules. The disadvantage of this scheme is that the coding redundancy is high, and it is difficult to effectively constrain task type matching when dealing with heterogeneous intelligent unmanned clusters.

[0006] To address the aforementioned problems, the existing technology has the following shortcomings: A1: The coding scheme is unreasonable: the two-layer coding scheme cannot simultaneously optimize task attribution and execution order, and the single-layer coding scheme has difficulty in constraint handling in heterogeneous systems. A2: Insufficient design of discrete operators: Operators of continuous optimization algorithms fail in discrete space, requiring the design of specialized discrete operators; A3: Weak multi-objective optimization capability: Most algorithms only focus on a single objective and ignore the need for multi-objective optimization; A4: Insufficient constraint handling capability: There is a lack of effective handling mechanisms for complex constraints such as load constraints and time window constraints, resulting in the generation of a large number of infeasible solutions. Summary of the Invention

[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method, system, device and medium for heterogeneous unmanned swarm air-ground collaborative task planning, which solves the problems of low coding efficiency, weak search capability and difficulty in constraint processing in heterogeneous intelligent unmanned swarm collaborative task planning.

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include: In a first aspect, the present invention provides a method for planning air-ground collaborative tasks in heterogeneous unmanned swarms, comprising: S100: The control device constructs a multi-objective optimization-oriented air-ground cooperative mission planning model based on the parameters of the UAVs and unmanned vehicles in the heterogeneous unmanned swarm; the objective function in the air-ground cooperative mission planning model includes: a total energy consumption cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints. S200. The control device uses a pre-corrected discrete multi-objective sheep migration optimization algorithm to solve the constructed multi-objective function and problem constraints, and obtains multiple non-dominated solutions in the specified storage area. The discrete multi-objective sheep migration optimization algorithm is an algorithm based on a multi-objective optimized air-ground cooperative task planning model and attribute correction of unmanned swarms. S300. The control device performs multi-objective comprehensive scoring on each non-dominated solution in the designated storage area, and determines the heterogeneous unmanned swarm air-ground collaborative task planning scheme based on the scoring results and the air-ground collaborative task scenario.

[0009] Optionally, in S100, the objective function for total energy consumption cost is... for: ; Where E represents a candidate non-dominated solution (or simply candidate solution); h represents the type of unmanned equipment, and Where u represents the type of aerial unmanned equipment, g represents the type of ground unmanned equipment, and a represents the number of unmanned equipment. This represents the number of unmanned equipment of type h. , M represents the total number of unmanned aerial vehicles (UAVs), and N represents the total number of unmanned ground vehicles (UAVs). This represents the energy consumption coefficient per unit distance traveled by the h-th type of unmanned equipment; This indicates the total path length under the task planning scheme of candidate solution E, where the h-th class and a-th unmanned equipment starts from the starting point, visits the task points assigned to the equipment in sequence, and returns to the starting point. Total Time Penalty Function The calculation method is as follows: ; Where K represents the total number of tasks, and i represents the index variable, taking values ​​from 1 to K, for each task in the candidate solution E. The arrival time is calculated based on the speed and path length of the unmanned equipment. Each task contains a task time window. , The earliest start time for the task. The latest start time for the task; like Then calculate the delay time. ;like Then the unmanned equipment will need to wait until Only then can we begin providing services. This is the time penalty coefficient, used to adjust the penalty intensity per unit of overdue time in the objective function. This coefficient is pre-calibrated based on the task's time sensitivity, allowable overdue level, and the discriminative power of optimization results under different parameter values. Its preferred value range is... .

[0010] Optionally, S200 includes: 3.1 Treat each sheep as a candidate solution, initialize the sheep flocking algorithm parameters, and randomly generate a task planning encoding sequence for each sheep as an initial candidate solution; the initialized sheep flocking algorithm parameters include: population size. Maximum number of discrete searches L during the grazing phase; maximum number of algorithm iterations. External archive B and archive capacity ; 3.2. Decode each candidate solution in the initial candidate solution into a task sequence for each unmanned equipment. Perform load capacity constraint checks, time window constraint checks, and access type constraint checks on each task sequence to obtain... and This forms a multi-objective fitness vector for candidate solutions. ; 3.3 The grazing phase is used to perform a local search in the vicinity of the current solution to improve the task allocation and execution order of each candidate solution in the flock; an improved solution is obtained after the grazing phase ends. 3.4. Merge the candidate solutions in the current flock with the original external archive B, and update the external archive B according to the Pareto dominance relationship; 3.5. The process of executing the migration phase is used to guide the flock out of the local area and expand the search range when the grazing phase can no longer effectively generate non-dominated solutions. Additionally, update the external archive, and when the maximum number of iterations is reached, output the non-dominated solutions in the updated external archive B.

[0011] Optionally, 3.3 includes: 3.31. Divide the target space into D parts along two target dimensions. For the r-th target dimension of the target space; r=1 represents the total energy cost dimension, r=2 represents the time window penalty dimension, determine the grid position of solution E in the two-dimensional target space, and the grid crowding of the grid containing solution E. ; 3.32. Based on the determined grid congestion, select the first solution from the external archive B according to the low-congestion area priority rule of the target space. Each candidate solution E is mapped to a point (F1(E), F2(E)) in the target space. 3.33. Based on the grid crowding, the probability of the first sheep choosing the solution is: After selecting the lead sheep solution, the lead sheep performs no more than L discrete searches; each follower sheep also performs no more than L discrete searches. The discrete searches use exchange, reversal, insertion, redistribution, and crossover operators to perturb the task planning encoding sequence and generate new candidate solutions. ; The fitness of the new candidate solutions is then recalculated. ; For the z-th sheep, let its historical optimal solution be... If a new candidate solution Pareto dominates the current individual's historical optimal solution. Or the two do not control each other, but The grid congestion is lower than the historical best solution. Then update the current solution and the historical best solution for the sheep. An improved solution is obtained after the grazing phase ends.

[0012] Optionally, 3.4 includes: If the updated external archive B exceeds the capacity C, the grid congestion strategy is used to clean up the external archive until the external archive size does not exceed C. Specifically, the change in the sequence of non-dominated solution target values ​​in the external archive is used as the migration trigger condition. Let the first... The external archive before the start of the next iteration of the grazing phase is The external archive after the grazing phase ends and non-dominated screening and grid congestion cleanup are completed. For any candidate solution E in the external archive, its objective value is determined by the total energy consumption cost objective. and time window penalty target constitute; Among them, the external archived target value sequence is constructed before the grazing phase begins. : ; Construct the external archive target value sequence after the grazing phase ends : ; in, Indicates the first The sequence of target values ​​of all non-dominated solutions in the external archive before the start of the next iteration grazing phase; Indicates the first After the grazing phase ends and the external archive is updated, the sequence of objective values ​​for all non-dominated solutions in the external archive; E represents any candidate solution in the external archive; This represents the target total energy cost corresponding to candidate solution E; This represents the time window penalty target value corresponding to candidate solution E; sort{·} indicates sorting the target value pairs in the target value sequence according to the comparison order. Compare again Sort the values ​​in ascending order to compare each item in the target value sequences before and after grazing. If the condition is met... = If this condition is met, the migration phase is triggered, and step 3.5 is executed; if the conditions are met... ≠ If so, return to step 3.3 to perform the next round of grazing search.

[0013] Optionally, the 3.5 includes: At the start of the migration, the target space grid congestion calculation method is used to select the solution from the low-congestion region in the external archive B as the lead sheep solution. Each sheep migrates in sequence according to the task planning code sequence. Specifically, for the z-th following sheep, according to the probability coefficient , , Randomly select from three migration strategies: based on probability. Choose local coding rearrangement, based on probability. Learn the task planning encoding sequence from the previous sheep, using probability... Learn from your own historical best coding; Migration updates can be represented as: , ; ; ; Among them, the head sheep will be dissected Assign to untie, express It follows the interval A random variable that is uniformly distributed on the upper bound. Indicates the z-th ( The current task planning encoding sequence for the sheep; This represents the new task planning coding sequence generated after the migration; This represents the task planning encoding sequence of the sheep preceding the z-th sheep. This represents the historical optimal task planning encoding sequence for the z-th sheep; This indicates an insertion operator used to adjust the execution order of tasks within the current solution; This represents the crossover operator, used to reorganize task order and task affiliation information between two task planning encoding sequences; after generating a new solution, its fitness is recalculated, and the current sheep and its historical best solution are updated according to Pareto dominance; all sheep in the population... Once all the sheep have completed their migration and renewal, the migration phase ends.

[0014] Optionally, S300 includes: Perform multi-objective comprehensive scoring on the non-dominated solutions in external archive B to obtain the score set of the solutions. ; For any nondominated solution E, first... and Perform min-max normalization separately: ; ; Where S represents any solution in the external archive B, It represents a very small positive number and is used to avoid the denominator being zero; Then construct the comprehensive scoring function: ; in, It is the normalized total energy cost. It is the normalized time window penalty, parameter Used to adjust the weights of different objectives. The target preference weights are used; finally, the non-dominated solution E with the smallest score is selected. As a planning scheme for air-ground collaborative missions, namely: ,in Let represent the independent variable that minimizes the objective function.

[0015] Secondly, the present invention also provides a heterogeneous unmanned swarm air-ground cooperative mission planning system, comprising: The model building unit is used to construct a multi-objective optimization-oriented air-ground cooperative mission planning model based on the parameters of UAVs and unmanned vehicles in a heterogeneous unmanned swarm; the objective function in the air-ground cooperative mission planning model includes: a total energy consumption cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints. The sheep herd solution unit is used to solve the constructed multi-objective function and problem constraints using a pre-modified discrete multi-objective sheep migration optimization algorithm, and obtain multiple non-dominated solutions in the specified storage area. The discrete multi-objective sheep migration optimization algorithm is an algorithm based on a multi-objective optimization air-ground cooperative task planning model and attribute correction of unmanned swarms. The planning scheme selection unit is used to perform multi-objective comprehensive scoring on each non-dominated solution in the specified storage area, and determine the planning scheme for the heterogeneous unmanned cluster air-ground collaborative task based on the scoring results and the air-ground collaborative task scenario.

[0016] Thirdly, the present invention also provides an unmanned aerial vehicle (UAV) device, comprising: a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory, specifically performing the steps of the heterogeneous unmanned swarm air-ground cooperative mission planning method described in the first aspect above.

[0017] Fourthly, the present invention also provides a computer storage medium, comprising: the computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the heterogeneous unmanned swarm air-ground cooperative task planning method as described in any of the first aspects.

[0018] The beneficial effects of this invention are: The method in this application constructs a multi-objective optimized air-ground cooperative mission planning model based on the attributes and mission parameters of the unmanned swarm. The objective functions of this model are the total energy consumption cost objective function and the total time penalty function. Then, based on a pre-corrected discrete multi-objective herding migration optimization algorithm, the constructed multi-objective functions and problem constraints are solved to obtain non-dominated solutions in a specified storage area. All non-dominated solutions are then comprehensively scored in multiple objectives. Based on the scoring results and the air-ground cooperative mission scenario, a heterogeneous unmanned swarm air-ground cooperative mission planning scheme is determined to maximize mission benefits and improve combat effectiveness.

[0019] The above method solves the problems of low coding efficiency, weak search capability, weak multi-objective optimization capability, and difficulty in constraint handling in the collaborative task planning of heterogeneous intelligent unmanned clusters. Attached Figure Description

[0020] Figure 1 A flowchart of a heterogeneous unmanned swarm air-ground collaborative task planning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a heterogeneous unmanned swarm air-ground collaborative task planning system provided in an embodiment of the present invention; Figure 3 To Figure 1 The diagram illustrates the interface parameter settings during the feasibility verification process of the method shown. Figure 4 To Figure 1 The diagram illustrates the initialization of the sheep herding algorithm during the feasibility verification process of the method shown. Detailed Implementation

[0021] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] In terms of existing algorithms for solving air-ground collaborative task planning, swarm intelligence algorithms such as genetic algorithms, particle swarm algorithms, and ant colony algorithms perform well in continuous optimization problems, but often require complex encoding and decoding schemes and specialized operators in discrete optimization problems.

[0023] Currently, the sheep migration optimization algorithm is a novel swarm intelligence algorithm and a new biomimetic swarm intelligence optimization method. Compared with existing swarm unmanned equipment optimization algorithms, the sheep migration optimization algorithm is less prone to getting trapped in local optima. By simulating the grazing and migration behavior of sheep, it exhibits good global search capabilities and convergence performance. However, the standard sheep migration optimization algorithm is mainly designed for solving continuous optimization problems and cannot be directly applied to solving discrete optimization problems. Therefore, in this application, the sheep migration optimization algorithm is modified to enable its application to discretized air-ground collaborative tasks.

[0024] Example 1: like Figure 1 As shown, this embodiment provides a flowchart of a heterogeneous unmanned swarm air-to-ground collaborative task planning method. This embodiment includes a control device that communicatively connects multiple unmanned aerial vehicles (UAVs) and multiple unmanned ground vehicles (UGVs). Each UAV and each UGV is equipped with its own intelligent device, such as an intelligent agent, which interacts with the control device. The method of this embodiment includes the following steps: S100: The control device constructs a multi-objective optimization-oriented air-ground cooperative mission planning model based on the parameters of the UAVs and unmanned vehicles in the heterogeneous unmanned swarm; the objective function in the air-ground cooperative mission planning model includes: a total energy consumption cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints.

[0025] For example, the total energy cost objective function for: ; Where E represents a candidate solution; h represents the type of unmanned equipment, and , where u represents the type of aerial unmanned equipment and g represents the type of ground unmanned equipment; This represents the number of unmanned equipment of type h, where, , ; This represents the energy consumption coefficient per unit distance traveled by the h-th type of unmanned equipment; This indicates the total path length under the task planning scheme of candidate solution E, where the h-th class of unmanned equipment, the a-th unit, starts from the starting point, visits the task points assigned to the equipment in sequence, and returns to the starting point.

[0026] Total Time Penalty Function The calculation method (for calculating the penalty for violating the corresponding time window): ; Where K represents the total number of tasks, and i represents the index variable, taking values ​​from 1 to K, for each task in the candidate solution E. The arrival time is calculated based on the speed and path length of the unmanned equipment. Each task contains a task time window. , The earliest start time for the task. This is the latest start time for the task. If (The latest start time of the task) is used to calculate the delay time. ;like (Earliest start time of the mission), then the unmanned equipment needs to wait until... Service can only begin after this period. There is no penalty during the waiting period, but it will affect the arrival time of subsequent tasks. This is the time penalty coefficient, used to adjust the penalty intensity per unit of overdue time in the objective function. This coefficient is pre-calibrated based on the task's time sensitivity, allowable overdue level, and the discriminative power of optimization results under different parameter values. Its preferred value range is... .

[0027] In this embodiment, the unmanned equipment set is as follows: ; ; Where M is the total number of unmanned aerial vehicles and N is the total number of unmanned ground vehicles; This indicates the m-th unmanned aerial vehicle. This represents the nth unmanned ground vehicle; This represents the maximum payload capacity of the m-th unmanned aerial vehicle. This represents the maximum payload capacity and travel speed of the nth unmanned ground vehicle. and energy consumption coefficient per unit distance traveled ; The task set is Each task Has access type attribute ,in Mission points can only be accessed by unmanned aerial vehicles. The mission points can only be accessed by unmanned ground equipment. Mission points can be accessed by both air and ground-based unmanned equipment, when the mission... Assigned to aerial unmanned equipment mission sequence At that time, it is necessary to meet the following requirements. When the task Assigned to unmanned ground equipment sequence At that time, it is necessary to meet the following requirements. .

[0028] For any aerial unmanned equipment Apply heavy capacity constraints: ;in Representing unmanned aerial vehicles in performing missions The load-bearing requirements.

[0029] For any ground unmanned equipment Apply heavy capacity constraints: ;in Representing ground-based unmanned equipment in performing missions The load-bearing requirements.

[0030] Time window constraint: when When this is done, a corresponding penalty term is introduced into the total time penalty objective function. For candidate solution E, unmanned equipment arrives at the mission. The actual time.

[0031] S200: The control device uses a pre-modified discrete multi-objective sheep migration optimization algorithm to solve the constructed multi-objective function and problem constraints, and obtain the non-dominated solution in the specified storage area.

[0032] The discrete multi-objective sheep migration optimization algorithm in this embodiment is an algorithm that modifies the air-ground cooperative task planning model based on multi-objective optimization and the attributes of unmanned swarms.

[0033] S300: The control device performs multi-objective comprehensive scoring on each non-dominated solution in the designated storage area, and determines the heterogeneous unmanned swarm air-ground collaborative task planning scheme based on the scoring results and the air-ground collaborative task scenario.

[0034] The method described in this embodiment can optimize air-ground collaborative missions to maximize mission benefits and improve combat effectiveness in order to enhance the overall system effectiveness.

[0035] Step S200 above may include the following sub-steps: Sub-step 3.1: Treat each sheep as a candidate solution individual, initialize the sheep flocking algorithm parameters, and randomly generate a task planning encoding sequence for each sheep as an initial candidate solution; the initialization of sheep flocking algorithm parameters includes: population size. Maximum number of discrete searches L during the grazing phase; maximum number of algorithm iterations. External archive B and archive capacity .

[0036] Sub-step 3.2: Decode each candidate solution in the initial candidate solution into a task sequence for each unmanned equipment. Perform load capacity constraint checks, time window constraint checks, and access type constraint checks on each task sequence to obtain... and This forms a multi-objective fitness vector for candidate solutions. .

[0037] Sub-step 3.3: Perform the grazing phase processing. The grazing phase is used to perform a local search in the vicinity of the current solution to improve the task allocation and execution order of each candidate solution in the flock; an improved solution is obtained after the grazing phase ends.

[0038] Sub-step 3.4: Merge the candidate solutions in the current flock with the original external archive B, and update the external archive B according to the Pareto dominance relationship.

[0039] Sub-step 3.5: Perform the processing of the migration phase. The migration phase is used to guide the flock out of the local area and expand the search range when the grazing phase can no longer effectively generate non-dominated solutions. Additionally, update the external archive, and when the maximum number of iterations is reached, output the non-dominated solutions in the updated external archive B.

[0040] This embodiment can solve the constructed multi-objective function and problem constraints based on a pre-corrected discrete multi-objective sheep migration optimization algorithm, and obtain the non-dominated solution in the specified storage area. It effectively solves the problems of low coding efficiency, weak search ability, weak multi-objective optimization ability and difficulty in constraint processing in the collaborative task planning of heterogeneous intelligent unmanned clusters.

[0041] Example 2: To better understand the solution in Embodiment 1 above, the following embodiments will be described in detail.

[0042] The method in this embodiment is applied to a delivery task involving multiple UAVs (unmanned aerial vehicles) and multiple UGVs (unmanned ground vehicles, or simply unmanned vehicles), each with its own task requirements and time window. In this embodiment, multiple UAVs and multiple UGVs are collectively referred to as an unmanned swarm, a heterogeneous intelligent unmanned swarm, or a heterogeneous unmanned swarm.

[0043] Once the main problem objects and environmental variables in the overall task scenario are determined, step 202 below describes the entire optimization problem. There are two overall objectives: one is to minimize the total energy consumption cost (corresponding to the aforementioned...). The other is to minimize the overall task time penalty (corresponding to the aforementioned). ).

[0044] 201. The control device initializes the heterogeneous intelligent unmanned cluster set A and the task set T based on the input information of the heterogeneous intelligent unmanned cluster.

[0045] In this embodiment, the unmanned swarm set A includes two categories: ground unmanned equipment (unmanned vehicles) and aerial unmanned equipment (drones). , Where M is the total number of unmanned aerial vehicles and N is the total number of unmanned ground vehicles; This indicates the m-th unmanned aerial vehicle. Represents the nth unmanned ground equipment; unmanned swarm set. .

[0046] The parameters of autonomous vehicles include: maximum payload capacity. speed of travel and energy consumption coefficient per unit distance traveled ; The parameters of the drone include: the drone's maximum payload capacity. Flight speed and energy consumption coefficient per unit distance traveled ; The parameters in the task set T include: task coordinates Task requirements Task Time Window Task service time and task access type ;in, It is the earliest start time of the mission. This is the latest start time for the task.

[0047] 202. The control device constructs a multi-objective optimization-oriented air-ground cooperative mission planning model. Specifically, the multi-objective optimization functions in the constructed air-ground cooperative mission planning model include: the total energy consumption cost objective function. Total Time Penalty Function .

[0048] Among them, the objective function of total energy consumption cost for: ; Total Time Penalty Function The calculation method is as follows: ; At this point, the constraints for drone and unmanned vehicle delivery may include: the total quantity of delivered materials cannot exceed the maximum load capacity, some task points are located in dense forests and can only be accessed by unmanned vehicles; others are on the water and can only be accessed by drones. These are the normal constraints of the problem. Therefore, problem constraints should be established, including: task access type constraints, load capacity constraints, and time window constraints.

[0049] In this embodiment, the heterogeneous intelligent unmanned cluster attributes and problem model constraints include: After constructing the task set and optimizing the objectives, the attributes of the heterogeneous intelligent unmanned swarm participating in the collaborative task execution are initialized, assuming the unmanned equipment set is denoted as . , ; in This indicates the m-th unmanned aerial vehicle. Let n be the nth unmanned ground vehicle; initialize the maximum payload capacity of both the aerial and ground unmanned vehicles. speed of travel and energy consumption coefficient per unit distance traveled .

[0050] To address the characteristics of heterogeneous intelligent unmanned swarms collaboratively executing tasks, we introduce task access type constraints, payload capacity constraints, and time window constraints to limit the feasible solution space of the problem. Let the task set be... Each task Has access type attribute When the task Assigned to aerial unmanned equipment mission sequence At that time, it is necessary to meet the following requirements. When the task Assigned to unmanned ground equipment sequence At that time, it is necessary to meet the following requirements. This ensures that missions are performed only by unmanned equipment of the same type.

[0051] At the same time, for any aerial unmanned equipment Apply heavy capacity constraints: ; For any ground unmanned equipment Apply heavy capacity constraints: ; And introduce time window constraints, set For unmanned equipment to reach the mission The actual time, when When necessary, a corresponding penalty term for breach of contract is introduced into the objective function.

[0052] 203. A pre-corrected discrete multi-objective sheep migration optimization algorithm is used to solve the constructed multi-objective function and problem constraints to obtain the output non-dominated solution.

[0053] In this embodiment, the solution is obtained iteratively through individual encoding, fitness calculation, policy execution, and updating. The solution for the discrete multi-objective sheep migration optimization algorithm model in this embodiment may include the following seven steps, namely steps 203.1 to 203.7: Step 203.1: Define each sheep as a candidate solution individual, initialize the sheep flocking algorithm parameters, and randomly generate the task planning encoding sequence for each sheep as an initial candidate solution.

[0054] In this embodiment, the encoding and decoding functions for the task planning encoding sequence include: Mission planning and coding sequence coding: Given a set of mission sequences for each unmanned equipment Initialize empty encoded sequence Iterate through each sequence in R Task sequence Each task number in Unmanned equipment number , }, the binary pair Add to the encoding sequence .

[0055] Task planning encoding sequence decoding: Given a task planning encoding sequence ,initialization Empty list traverse the encoded sequence Each pair in , will the task Add to task sequence Returns a set of task sequences. .

[0056] In addition, the parameters of the sheep herding algorithm include: population size. Maximum number of discrete searches L during the grazing phase; maximum number of algorithm iterations. External archive B and archive capacity The external archive here refers to the naming of the specified storage content.

[0057] Step 203.2: Decode each candidate into a task sequence for each unmanned equipment, perform load capacity constraint checks, time window constraint checks, and access type constraint checks on each task sequence, and calculate the total energy consumption cost target. and time window penalty items This forms a multi-objective fitness vector for candidate solutions. ; Step 203.3: Perform the grazing phase calculations. The grazing phase is used to perform a local search in the vicinity of the current solution to improve the task allocation and execution order of candidate solutions in the flock, as described in substeps A1 to A6.

[0058] Sub-step A1: First, based on the grid congestion determined in sub-step A3, select the leading solution from the external archive B according to the low-congestion region priority rule of the target space. Each candidate solution E can be mapped to a point (F1(E), F2(E)) in the target space. The target space mentioned above refers to the two-dimensional space composed of the multi-objective fitness vectors (F1, F2).

[0059] It should be noted that in the actual implementation process, sub-step A1 and sub-step A3 can be performed simultaneously, or it can be understood that sub-step A3 is executed first, and then sub-step A1 is executed.

[0060] Sub-step A2: Divide the target space into D parts along the two target dimensions, where D represents the number of grid divisions in each target dimension; for the r-th target dimension of the target space; r=1 represents the total energy cost dimension, r=2 represents the time window penalty dimension, and the grid number of solution E in this dimension is: (1); in, F represents the grid number of solution E in the r-th objective dimension; E represents the candidate solution for the current grid position to be calculated; (E) represents the function value of solution E on the r-th objective; This represents the minimum value of all solutions in external archive B on the r-th objective; D represents the maximum value of all solutions in the external archive B at the r-th objective; B represents the external archive; D represents the number of grid divisions in the objective space for each objective dimension; It represents a very small positive number and is used to avoid the denominator being zero; · This indicates the floor function.

[0061] Sub-step A3: Let Let E represent the grid position of solution E in the two-dimensional target space. Then, the grid crowding of the grid containing solution E is defined as: (2); in, G(S) represents the grid congestion of the target space grid where solution E is located; S represents any candidate solution in the external archive B; G(S) represents the two-dimensional target space grid where solution S is located; G(E) represents the two-dimensional target space grid where solution E is located; G(S)=G(E) means that solution S and solution E are in the same grid. Let |·| represent the set of all solutions in external archive B that are located on the same grid as solution E; |·| represents the number of elements in the set. Therefore, The smaller the value, the sparser the region where the solution is located, that is, the higher the low crowding degree, which is more conducive to maintaining the diversity of the distribution of Pareto solution set in the target space.

[0062] Sub-step A4: Based on the above grid crowding, the selection probability of the first sheep solution can be expressed as: (3); Where P(E) represents the probability that solution E is selected as the first solution; ρ(S) represents the grid crowding of any solution S in the archive. This represents the summation of all solutions S in the external archive B. The formula indicates that solutions with lower grid crowding have a higher probability of being selected as the lead solution. After the lead solution is selected, it performs no more than L discrete searches; each follower solution also performs no more than L discrete searches.

[0063] Sub-step A5: Discrete search selects exchange, inversion, insertion, redistribution, and crossover operators to perturb the task planning coding sequence and generate new candidate solutions. .

[0064] The execution steps of the exchange operator are (B1) to (B4): (B1) Copy the current encoded sequence ; (B2) Randomly select two different location indices and ,in and ; (B3) Exchange positions and Elements on; ; (B4) Return the new encoded sequence .

[0065] Furthermore, the execution steps of the inversion operator are (C1) to (C4): (C1) Copy the current encoded sequence ; (C2) Randomly select two position indices and ,make sure ; (C3) Reverse subsequence ; (C4) Returns the new encoded sequence .

[0066] Furthermore, the insertion operator The execution steps are (D1) to (D4): (D1) Copy the current encoded sequence ; (D2) Randomly select a location Remove elements from the sequence ; (D3) Randomly select another location The element to be removed Insert at position ; (D4) Returns the new encoded sequence .

[0067] The insertion operator adjusts the execution order by moving the position of individual tasks. It is a local search operator suitable for fine-grained optimization within the neighborhood of the solution.

[0068] Furthermore, the execution steps of the redistribution operator are (E1) to (E6): (E1) Copy the current encoded sequence ; (E2) Randomly select a location Get the task number for this location. and the currently allocated unmanned equipment ; (E3) Determine the feasible set of unmanned equipment based on the mission type; (E4) Randomly select a new unmanned equipment number from the feasible set. ; (E5) Update encoding ; (E6) Returns the new encoded sequence .

[0069] The redistribution operator in the algorithm is used to change the affiliation of a task, reassigning the task to another unmanned vehicle.

[0070] Furthermore, the cross operator The execution steps are (F1) to (F5): (F1) Select parent coding sequence and ; (F2) Randomly generate intersection points ; (F3) From Take the first half ; (F4) From Extracting from the middle in sequence The tasks that appear in the game: Each pair in ,like Not in If it appears in the middle, then it will Add to (Keep constant).

[0071] (F5) Returns the child encoding sequence .

[0072] The crossover operator is used to achieve knowledge transfer during the migration phase by generating new solutions by combining the features of two excellent solutions. The crossover operator guarantees that the offspring solution contains all tasks and that each task appears only once, while inheriting some task allocation and execution order information from the parents.

[0073] Sub-step A6: Then recalculate the fitness of the new candidate solutions. : .

[0074] For the z-th sheep, let its historical optimal solution be... If a new candidate solution Pareto dominates the current individual's historical optimal solution. Or the two do not control each other, but The grid congestion is lower than the historical best solution. Then update the current solution and the historical best solution for the sheep. An improved solution is obtained after the grazing phase ends.

[0075] Step 203.4: After the grazing phase ends, merge the candidate solutions in the current flock with the original external archive B, and update the external archive according to the Pareto dominance relationship.

[0076] If the updated external archive size exceeds capacity C, a grid congestion strategy is used to clean up the external archive, prioritizing the deletion of solutions in congested grids until the external archive size does not exceed C.

[0077] For example, the execution steps of the grid congestion strategy are as follows: (1) Determine the range of the target space: calculate the minimum and maximum values ​​of the objective function for all solutions; (2) Divide the target space into D × D grids; (3) Calculate the grid coordinates of each solution; (4) Calculate the congestion of the grid where each solution E is located. (5) If the archive size exceeds C, sort by grid density from high to low and delete it first. (6) Return to the cleaned-up archive.

[0078] The grid congestion strategy described above is used to maintain the diversity of the external archive. By removing solutions from dense regions, it preserves the more evenly distributed Pareto front, enabling decision-makers to select satisfactory solutions from different regions of the solution space.

[0079] This embodiment uses whether the sequence of non-dominated solution target values ​​in the external archive changes as the migration trigger condition. Let the first... The external archive before the start of the next iteration of the grazing phase is The external archive after the grazing phase ends and non-dominated screening and grid congestion cleanup are completed. For any candidate solution E in the external archive, which is a feasible candidate solution, its objective value is determined by the total energy consumption cost objective. and time window penalty target constitute.

[0080] Specifically, construct the external archived target value sequence before the grazing phase begins. : (4) Construct the external archive target value sequence after the grazing phase ends : (5) in, Indicates the first The sequence of objective values ​​of all feasible nondominated solutions in the external archive before the start of the next iteration grazing phase; Indicates the first After the grazing phase ends and the external archive is updated, the objective value sequence of all feasible non-dominated solutions in the external archive; E represents any feasible candidate solution (i.e., a feasible solution) in the external archive. This represents the target total energy cost corresponding to candidate solution E; This represents the time window penalty target value corresponding to candidate solution E; {·} indicates that the target value pair is compared first. Compare again The values ​​are sorted in ascending order to ensure that the two target value sequences before and after grazing can be compared item by item.

[0081] If the following conditions are met: = If the result is positive, it means that after the grazing phase ends, the number of non-dominated solutions in the external archive has not changed, and the target value pairs corresponding to each non-dominated solution have not changed. This indicates that the grazing search has not generated any new Pareto non-dominated solutions, nor has it improved the target values ​​of existing non-dominated solutions in the external archive. Therefore, the current search is determined to have entered a stagnant state, and the migration phase is triggered, and step 203.5 is executed.

[0082] If the following conditions are met: ≠ If the result is positive, it means that after the grazing phase ends, the number of non-dominated solutions in the external archive changes, or at least the objective value pair corresponding to a non-dominated solution changes. This indicates that the grazing search has effectively updated the Pareto front and does not trigger the migration phase. Therefore, we continue to return to step 203.3 to execute the next round of grazing search.

[0083] Step 203.5: Perform the migration phase calculation. The migration phase is used to guide the flock out of the local area and expand the search range when the grazing phase can no longer effectively generate non-dominated solutions. At the start of the migration, according to the target space grid congestion calculation method defined in Step 203.3, a solution in a low-congestion area from the external archive B is selected as the lead solution. That is, the archive solution with smaller ρ(E) is preferentially selected as the migration guide solution, causing the flock to move to the area with a sparser distribution of solutions in the target space, thereby enhancing the global search capability and solution set diversity.

[0084] Select the lead sheep Afterwards, the sheep that followed migrated in sequence according to their coded sequences.

[0085] Specifically, the first following sheep uses the solution of the first sheep as its predecessor, and the remaining following sheep use the code of the previous sheep as their predecessor. For the z-th following sheep, according to the probability coefficient... , , Randomly select from three migration strategies: based on probability. Choose local coding rearrangement, based on probability. Learn the code of the previous sheep with probability Learn from your own historical best coding.

[0086] For example, in this embodiment, it can be set to... =0.5, =0.25, =0.25.

[0087] Migration updates can be represented as: , ; ; ; Among them, the head sheep will be dissected Assign to untie, express It follows the interval A random variable that is uniformly distributed on the upper bound. This represents the current task planning encoding sequence for the z-th sheep; This represents the task planning coding sequence generated after migration; This represents the task planning encoding sequence of the sheep preceding the z-th sheep. This represents the historical best code for the z-th sheep; This indicates an insertion operator used to adjust the execution order of tasks within the current decoding process. This represents the crossover operator, used to reorganize task order and task affiliation information between two encodings. After generating a new solution, its fitness is recalculated, and the current sheep and its historical best solution are updated according to Pareto dominance. All sheep in the population... Once all sheep have completed their migration update, the migration phase ends, and we proceed to step 203.6.

[0088] Step 203.6: Update external save file B. Compare all candidate solutions from the current sheep population with the original external save file. By merging the solutions in the solution list, we obtain the candidate set. ;right Any two feasible solutions and If the following conditions are met: , If there exists at least one objective that is strictly less than , then we say that... Pareto dominance ,Right now Pareto dominion is indicated by the symbol This indicates that the non-dominated solution set is obtained through this selection process. ; non-dominated solution set Write to external archive B. If the size of the external archive exceeds the capacity threshold C, continue to clean up using the grid congestion strategy: divide the target space into D×D grids and calculate the grid congestion of the grid containing each solution E. Delete first Solutions in larger crowded grids are used until the number of solutions in the external archive B is less than or equal to the capacity threshold C, thus preserving Pareto nondominance while maintaining the diversity of solution set distribution.

[0089] Step 203.7: Determine if the maximum number of iterations has been reached. If the solution is not reached, return to step 203.3 to continue solving; otherwise, output the non-dominated solution in the external archive. Step 204: Perform multi-objective comprehensive scoring on the non-dominated solutions in external archive B to obtain the score set of the solutions. Based on the task scenario, the solution with the minimum score is selected as the air-ground collaborative task planning scheme.

[0090] The control device in this embodiment obtains overall environmental information and problem modeling, and solves and calculates the entire problem in a centralized manner. The solution obtained is the task set and task execution order of each UAV and each unmanned vehicle; each UAV and unmanned vehicle is responsible for executing these tasks.

[0091] For example, the method for selecting the final task planning scheme in step 204 is as follows: Perform multi-objective comprehensive scoring on the non-dominated solutions in external archive B to obtain the score set of the solutions. The scoring uses a normalized weighted sum function. For any non-dominated solution E, its total energy cost objective is first evaluated. and time window penalty target Perform min-max normalization separately: ; ; Then construct the comprehensive scoring function: ; in, It is the normalized total energy cost. It is a normalized time window penalty. It is a tiny positive number to prevent the denominator from being zero; parameter Used to adjust the weights of different objectives. Weights for target preferences; The larger the value, the more emphasis is placed on total energy cost. The smaller the value, the more emphasis is placed on the time window penalty. Ultimately, the non-dominated solution E with the smallest score is chosen. As a planning scheme for air-ground collaborative missions, namely: .

[0092] This embodiment is based on a pre-modified discrete multi-objective sheep migration optimization algorithm. It solves the constructed multi-objective function and problem constraints to obtain the non-dominated solution in the specified storage area. This solves the problems of low coding efficiency, weak search ability, weak multi-objective optimization ability and difficulty in constraint processing in the collaborative task planning of heterogeneous intelligent unmanned clusters.

[0093] Combination Figures 3 to 4 As shown, the methods described in Embodiments 1 and 2 above have been verified to be feasible. They can realize and effectively solve the problems in the prior art, give full play to the capabilities of air and ground unmanned equipment, combine complex mission requirements for unified collaborative planning, realize the rational allocation and efficient use of cluster resources, and improve the overall mission efficiency of unmanned equipment clusters.

[0094] Example 3: The problem of collaborative material delivery by a heterogeneous intelligent unmanned swarm is described as follows: In a 100 m × 100 m map, there are 30 task points and 15 obstacles. Two aerial unmanned vehicles (UAVs) and two ground-based UAVs are deployed for collaborative material delivery. The aerial UAVs have a speed of 15 m / s, and the ground-based UAVs have a speed of 8 m / s. The maximum payload of the aerial UAVs is 50 kg, and the maximum payload of the ground-based UAVs is 100 kg. Of the 30 task points, 7 can only be performed by aerial UAVs, 9 can only be performed by ground-based UAVs, and 14 can be performed by any type of UAV. Each task point has material requirements, time windows, and service time requirements. The optimization objective is to minimize the total energy cost and the time window penalty.

[0095] Algorithm parameter settings: We set the population size based on the complexity and scale of the task. = 50, maximum number of iterations G max = 100, maximum discrete search count L = 1, update probabilities for migration stages are: p1 = 0.5, p2 = 0.25, p3 = 0.25, archive capacity C = 100, grid partitioning D = 10.

[0096] Algorithm execution process: Step 301: Construct a heterogeneous intelligent unmanned swarm task set within the two-dimensional task region. In this embodiment, the task region is 100 m × 100 m in size and contains 15 obstacles. Construct the task set. Each task Includes task space location Task access type Task requirements Task Time Window and task service time Attributes; of the 30 missions, 7 can only be performed by aerial unmanned equipment (T2, T8, T...). 13 ,T 14 , T 20 , T 29 , T 30 Nine missions can only be performed by ground-based unmanned equipment (T1, T3, T5, T9, T...). 10 , T 19 , T 22 ,T 24 , T 25 ), 14 missions can be performed by any type of unmanned equipment (T4, T6, T7, T... 11 , T 12 , T 15 , T 16 , T 17 ,T 18 , T 21 , T 23 , T 26 , T 27 , T 28 The mission payload requirement is a random number within the range of 1–10 kg, and the mission time window is... The task service time is generated from the task scenario and used as input for task constraints. The standard is 120 seconds. The heterogeneous intelligent unmanned swarm A in this embodiment consists of 2 aerial unmanned vehicles and 2 ground unmanned vehicles. The aerial unmanned vehicle is numbered 1 (…). ) and 2 ( The ground unmanned equipment is numbered 3 ( ) and 4 ( Flight speed of unmanned aerial vehicles Set to 15 m / s, maximum load capacity All are set to 50 kg; the travel speed of unmanned ground equipment Set to 8 m / s, maximum load capacity Both are set to 100 kg. The energy consumption coefficient per unit distance traveled for aerial unmanned equipment and ground unmanned equipment is denoted as follows: and And used as a parameter for calculating cost targets.

[0097] Step 302: Initialize the attributes of the heterogeneous intelligent unmanned cluster and construct constraints. Construct based on total system energy cost With time window penalty A multi-objective optimization model for optimizing objectives; This embodiment sets two optimization objectives. The first objective... The first objective is the total energy consumption cost of the system, used to characterize the total energy consumption cost incurred by each unmanned equipment from its starting point, visiting its mission sequence, and returning to the starting point; the second objective is... This is a time window penalty target, used to characterize the lateness penalty incurred when the actual arrival time of a task exceeds the latest start time. Lateness penalty coefficient. Set to 10; if no one is equipped before the earliest start time of the mission. Upon arrival, wait until Once the service begins, the waiting time will not incur any penalty for breach of contract, but it will be included in the arrival time of subsequent tasks.

[0098] The model includes mission access type constraints, payload capacity constraints, and time window constraints. The mission access type constraint ensures that only air-accessible missions are assigned to air-based unmanned equipment, and only ground-accessible missions are assigned to ground-based unmanned equipment. The payload capacity constraint ensures that the total payload required by any air-based unmanned equipment does not exceed 50 kg, and the total payload required by any ground-based unmanned equipment does not exceed 100 kg. Candidate solutions that violate the mission access type constraint or payload capacity constraint are treated as infeasible solutions in the fitness evaluation.

[0099] Step 303: Solve the above multi-objective optimization model based on the modified discrete multi-objective sheep migration optimization method, which includes the processes of task planning coding sequence design, population initialization, fitness calculation, grazing search, migration update, and external archive maintenance.

[0100] Sub-step 303.1: Construct a task planning encoding sequence to represent the collaborative task planning scheme and initialize the relevant parameters of the sheep herding algorithm. Each sheep is defined as a candidate solution, and each candidate solution is represented by a task planning encoding sequence, with the encoding unit being a binary tuple. ,in Indicates the task number. Indicates the assigned unmanned equipment number. The order of the encoded units in the sequence represents the relative order of task access; decoding is performed according to the unmanned equipment number. The tasks are assigned to the corresponding unmanned equipment task sequences. For example, the encoding sequence for a sheep (a candidate solution) is: [(3, 1), (15, 2), (8, 1), (22, 3), (18, 4),...], which means that task 3 is assigned to air unmanned equipment number 1, task 15 is assigned to air unmanned equipment number 2, task 8 is also assigned to air unmanned equipment number 1 and executed after task 3, task 22 is assigned to ground unmanned equipment number 3, and task 18 is assigned to ground unmanned equipment number 4.

[0101] Initialize the parameters related to the sheep herding algorithm and the algorithm population size. Setting it to 50 sets the maximum number of algorithm iterations. The maximum number of discrete searches L during the grazing phase is set to 100. During initialization, a task planning encoding sequence for 50 sheep is randomly generated, and the fitness of each initial candidate solution is calculated; an initial external archive B is constructed based on Pareto dominance relationships. The external archive capacity C is set to 100, and the number of grid divisions in the target space D is set to 10. A joint task planning encoding sequence is randomly generated for each sheep, i.e., each initial solution.

[0102] Sub-step 303.2: For any candidate solution E, decode the task planning encoding sequence of all 4 unmanned equipment in turn to obtain 4 unmanned equipment task sequences, and then calculate the path length of each unmanned equipment from the starting point, visiting the task points in sequence and returning to the starting point.

[0103] Subsequently, the arrival time of each mission was calculated based on the speed of the aerial unmanned equipment (15 m / s) and the speed of the ground unmanned equipment (8 m / s). According to the task requirements The total payload of each unmanned piece of equipment is calculated based on the lateness penalty coefficient. The time window penalty is calculated, and the fitness vector of the candidate solution E is finally obtained. .

[0104] Sub-step 303.3: Perform the grazing phase and enter the optimization loop. During the grazing phase, apply the discrete search operator to each sheep, calculate the fitness of the new solution, and update the external archive. Taking the 10th iteration as an example, apply the swap operator to the 9th sheep in the population, randomly select position 5 and position 12 to swap, evaluate the fitness of the new solution, and if it is better than the historical best solution, then update the historical best solution.

[0105] Sub-step 303.4: After the grazing phase ends, merge the current flock with the original external archive B, and update the external archive according to the Pareto dominance relationship. If the size of the external archive exceeds C=100, a grid congestion strategy of D=10 is used to clean up the archive, prioritizing the deletion of solutions in congested grids, until the archive size does not exceed 100. This embodiment uses whether the target value sequence in the external archive changes as the migration trigger condition. Specifically, it considers whether the target values ​​of each solution in the external archive B change. Form a target value sequence H; if the target value sequence before and after grazing is the same, it is determined that the current grazing search has not produced a new effective Pareto improvement, triggering the migration condition and executing sub-step 303.5; if the archived features change, return to sub-step 303.3 to enter the next round of grazing search.

[0106] Sub-step 303.5: The migration phase is used to expand the search area when the grazing phase is stalled. In this embodiment, at the start of the migration, the first sheep solution is still selected from the external archive B according to the low-crowding area priority rule. After selecting the lead sheep, the following sheep migrate sequentially according to the encoded sequence; for the z-th following sheep, three migration learning probabilities are set. , , Among them, probability Choose a local coding rearrangement strategy, based on probability. Learn the code of the previous sheep using the crossover operator. With probability Learn its own historical best code through the crossover operator. The first following sheep uses the aforementioned lead sheep's solution as its precedence reference, and the remaining following sheep use the code of the previous sheep as their precedence reference.

[0107] Migration update generates new task planning coding sequence Then, recalculate its fitness and update the current sheep and its historical best based on Pareto dominance. All in the population Once all sheep have completed their migration and updates, the migration phase ends, and the process proceeds to sub-step 303.6.

[0108] Sub-step 303.6: Update the external archive, merge the 50 sheep in the current population with the solutions in the external archive, use Pareto dominance to filter non-dominated solutions, use grid crowding strategy to clean up the archive, and ensure that the archive size does not exceed 100.

[0109] Sub-step 303.7: Check the termination condition, repeat sub-steps 303.3 to 303.6 until the termination condition is met. In the next iteration, the non-dominated solution in the external archive is output.

[0110] Step 304: Perform multi-objective comprehensive scoring on the non-dominated solutions in the external archive, and select the solution with the minimum score as the air-ground collaborative task planning scheme according to the task scenario: specifically, for each non-dominated solution, the total energy consumption target is... Time window penalty target Normalization is performed separately to obtain the normalized target value. and And construct a comprehensive evaluation function for any nondominated solution E: .

[0111] This embodiment adopts a balanced preference, with the target preference weight λ set to 0.5, meaning the total energy consumption cost target weight is 0.5, and the time window penalty target weight is 0.5. A comprehensive score is selected. Minimum nondominated solution E As the final air-ground collaborative mission planning scheme, it outputs the mission allocation results and mission execution order for the four unmanned equipment.

[0112] Under the parameters of the above embodiments, the algorithm obtains the Pareto front solution set after 100 iterations. Compared with the initial random scheme, the final non-dominated solution set shows significant improvement in both the total energy cost objective and the total time window penalty objective; specifically, the total energy cost objective can be reduced from the initial 1981.28 to 1336.56, and the time window penalty objective can be reduced from the initial 6652.13 to 0. The results show that the present invention, by describing both task affiliation and execution order simultaneously through task planning encoding sequences, and combining local search during the grazing phase, stagnation-triggered migration search, and external archive grid congestion maintenance, can effectively improve the solution quality of heterogeneous intelligent unmanned swarm air-ground collaborative task planning.

[0113] Finally, for the 30 tasks, the execution order of each UAV and UGV is as follows: (Number 1): T8,T 20 , T 21 ,T 17 , T 26 , T 27 , T 13 ; (Number 2): T 29 , T2, T 14 , T4, T 30 ; (Number 3): T 10 , T 28 , T3, T 12 , T 24 , T9, T 23 , T 25 , T 22 T7, T5, T 19 ; (Number 4): T1, T6, T16 , T 11 , T 15 , T 18 .

[0114] The above method can maximize mission benefits and improve combat effectiveness, while solving problems such as low coding efficiency, weak search capability, weak multi-objective optimization capability, and difficulty in constraint handling in the collaborative mission planning of heterogeneous intelligent unmanned swarms.

[0115] On the other hand, embodiments of the present invention also provide a heterogeneous unmanned swarm air-ground collaborative task planning system, such as... Figure 2 As shown, the system includes: The model building unit is used to construct a multi-objective optimization-oriented air-ground cooperative mission planning model based on the parameters of UAVs and unmanned vehicles in a heterogeneous unmanned swarm; the objective function in the air-ground cooperative mission planning model includes: a total energy consumption cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints. The sheep herd solution unit is used to solve the constructed multi-objective function and problem constraints using a pre-modified discrete multi-objective sheep migration optimization algorithm, and obtain the non-dominated solution in the specified storage area. The discrete multi-objective sheep migration optimization algorithm is an algorithm designed and modified based on the multi-objective optimization air-ground cooperative task planning model and the attributes of unmanned swarms. The planning scheme selection unit is used to perform multi-objective comprehensive scoring on each non-dominated solution in the specified storage area, and determine the planning scheme for the heterogeneous unmanned cluster air-ground collaborative task based on the scoring results and the air-ground collaborative task scenario.

[0116] The aforementioned system can maximize mission benefits and enhance combat effectiveness.

[0117] Thirdly, embodiments of the present invention also provide an unmanned aerial vehicle (UAV) device, including: a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory, specifically performing the steps of the heterogeneous unmanned swarm air-ground cooperative task planning method described in any of the above embodiments.

[0118] Fourthly, embodiments of the present invention also provide a computer storage medium, comprising: the computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the heterogeneous unmanned swarm air-ground collaborative task planning method as described in the above embodiments.

[0119] The methods described in the above embodiments will not be detailed here.

[0120] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0121] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0122] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0123] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0124] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for planning air-ground collaborative tasks in heterogeneous unmanned swarms, characterized in that, include: S100: The control device constructs an air-ground cooperative mission planning model oriented towards multi-objective optimization based on the parameters of UAVs and unmanned vehicles in a heterogeneous unmanned swarm. The objective function in the air-ground collaborative task planning model includes: a total energy cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints. S200. The control device uses a pre-corrected discrete multi-objective sheep migration optimization algorithm to solve the constructed multi-objective function and problem constraints, and obtains multiple non-dominated solutions in the specified storage area. The discrete multi-objective sheep migration optimization algorithm is an algorithm modified based on the multi-objective optimization air-ground cooperative task planning model and the attributes of the unmanned swarm. S200 includes: 3.1 Treating each sheep as a candidate solution individual, initializing the sheep flocking algorithm parameters, and randomly generating a task planning encoding sequence for each sheep as an initial candidate solution; initializing the sheep flocking algorithm parameters includes: population size Maximum number of discrete searches L during the grazing phase; maximum number of algorithm iterations. External archive B and archive capacity ; 3.

2. Decode each candidate solution in the initial candidate solution into a task sequence for each unmanned equipment. Perform load capacity constraint checks, time window constraint checks, and access type constraint checks on each task sequence to obtain... and This forms a multi-objective fitness vector for candidate solutions. ; 3.3 The grazing phase involves performing a local search near the current solution to improve the task allocation and execution order of candidate solutions in the flock. An improved solution is obtained after the grazing phase. Specifically, based on grid congestion, a lead sheep solution is selected from the external archive B according to the low-congestion region priority rule of the target space. The lead sheep performs no more than L discrete searches. Each following sheep also performs no more than L discrete searches. Discrete searches use exchange, reversal, insertion, redistribution, and crossover operators to perturb the task planning encoding sequence and generate new candidate solutions. 3.

4. Merge the candidate solutions in the current flock with the original B, and update B according to the Pareto dominance relationship; 3.

5. Execute the processing of the migration phase, which is used to guide the flock out of the local area and expand the search range when the grazing phase cannot continue to generate non-dominated solutions effectively; and update B, and when the maximum number of iterations is reached, output the non-dominated solutions in the updated B. S300. The control device performs multi-objective comprehensive scoring on each non-dominated solution in the designated storage area, and determines the heterogeneous unmanned swarm air-ground collaborative task planning scheme based on the scoring results and the air-ground collaborative task scenario.

2. The heterogeneous unmanned swarm air-ground collaborative task planning method according to claim 1, characterized in that, In S100, Total energy cost objective function for: ; Where E represents a candidate non-dominated solution, i.e., a candidate solution; h represents the type of unmanned equipment, and Where u represents the type of aerial unmanned equipment, g represents the type of ground unmanned equipment, and a represents the number of unmanned equipment. This represents the number of unmanned equipment of type h. This represents the energy consumption coefficient per unit distance traveled by the h-th type of unmanned equipment; This indicates the total path length under the task planning scheme of candidate solution E, where the h-th class and a-th unmanned equipment starts from the starting point, visits the task points assigned to the equipment in sequence, and returns to the starting point. Total Time Penalty Function The calculation method is as follows: ； Where K represents the total number of tasks, and i represents the index variable, taking values ​​from 1 to K, for each task in the candidate solution E. The arrival time is calculated based on the speed and path length of the unmanned equipment. Each task contains a task time window. , The earliest start time for the task. The latest start time for the task; like Then calculate the delay time. ;like Then the unmanned equipment will need to wait until Only then can we begin providing services. This is the time penalty coefficient, used to adjust the penalty intensity per unit of overdue time in the objective function; its value range is... .

3. The heterogeneous unmanned swarm air-ground cooperative task planning method according to claim 1, characterized in that, 3.3 includes: 3.

31. Divide the target space into D parts along two target dimensions. For the r-th target dimension of the target space; r=1 represents the total energy cost dimension, r=2 represents the time window penalty dimension, determine the grid position of solution E in the two-dimensional target space, and the grid crowding of the grid containing solution E. ; 3.

32. Based on the grid congestion, select the first solution from B according to the low-congestion area priority rule of the target space. Each candidate solution E is mapped to a point (F1(E), F2(E)) in the target space. 3.

33. Based on the grid crowding, the probability of the first sheep choosing the solution is: After selecting the lead sheep solution, the lead sheep performs no more than L discrete searches; each follower sheep also performs no more than L discrete searches. The discrete searches use exchange, reversal, insertion, redistribution, and crossover operators to perturb the task planning encoding sequence and generate new candidate solutions. ; The fitness of the new candidate solutions is then recalculated. ; For the z-th sheep, let its historical optimal solution be... If a new candidate solution Pareto dominates the current individual's historical optimal solution. Or the two do not control each other, but The grid congestion is lower than the historical best solution. Then update the current solution and the historical best solution for the sheep. An improved solution is obtained after the grazing phase ends.

4. The heterogeneous unmanned swarm air-ground cooperative task planning method according to claim 3, characterized in that, 3.4 includes: If the size of B after the update exceeds the archive capacity C, the grid congestion strategy will be used to clean up the external archive until the size of the external archive does not exceed C. Specifically, the change in the sequence of non-dominated solution target values ​​in the external archive is used as the migration trigger condition. Let the first... The external archive before the start of the next iteration of the grazing phase is The external archive after the grazing phase ends and non-dominated screening and grid congestion cleanup are completed. For any candidate solution E in B, its objective value is determined by... and constitute; Among them, the external archived target value sequence is constructed before the grazing phase begins. : ； Construct the external archive target value sequence after the grazing phase ends : ； in, Indicates the first The sequence of target values ​​of all non-dominated solutions in the external archive before the start of the next iteration grazing phase; Indicates the first After the grazing phase ends and the external archive is updated, the sequence of objective values ​​for all non-dominated solutions in the external archive; E represents any candidate non-dominated solution in B. This represents the total energy cost target value corresponding to E; This represents the time window penalty target value corresponding to E; sort{·} indicates sorting the target value pairs in the target value sequence according to the comparison order. Compare again Sort the values ​​in ascending order to compare each item in the target value sequences before and after grazing. If the condition is met... = If this condition is met, the migration phase is triggered, and step 3.5 is executed; if the conditions are met... ≠ If so, return to step 3.3 and continue with the next round of grazing search.

5. The heterogeneous unmanned swarm air-ground cooperative task planning method according to claim 4, characterized in that, The 3.5 includes: At the start of the migration, the target space grid congestion calculation method is used to select the solution from the low-congestion region of B as the lead sheep solution. Each sheep migrates in sequence according to the task planning code sequence. Specifically, for the z-th following sheep, according to the probability coefficient , , Randomly select from three migration strategies: based on probability. Choose local coding rearrangement, based on probability. Learn the task planning encoding sequence from the previous sheep, using probability... Learn from your own historical best coding; Migration updates are represented as: ， ; ; ； Among them, the head sheep will be dissected Assign to untie, express It follows the interval A random variable that is uniformly distributed on the upper bound. This represents the current task planning encoding sequence for the z-th sheep. ; This represents the new task planning coding sequence generated after the migration; This represents the task planning encoding sequence of the sheep preceding the z-th sheep. This represents the historical optimal task planning encoding sequence for the z-th sheep; This indicates an insertion operator used to adjust the execution order of tasks within the current solution; This represents the crossover operator, used to reorganize task order and task affiliation information between two task planning encoding sequences; after generating a new solution, its fitness is recalculated, and the current sheep and its historical best solution are updated according to Pareto dominance; all sheep in the population... Once all the sheep have completed their migration and renewal, the migration phase ends.

6. The heterogeneous unmanned swarm air-ground cooperative task planning method according to claim 5, characterized in that, The S300 includes: Perform multi-objective comprehensive scoring on the non-dominated solutions in external archive B to obtain the score set of the solutions. ; For any nondominated solution E, first... and Perform min-max normalization separately ; ； Where S represents any solution in the external archive B. It represents a very small positive number and is used to avoid the denominator being zero; Then construct the comprehensive scoring function: ； in, It is the normalized total energy cost. It is the normalized time window penalty, parameter Used to adjust the weights of different objectives. The target preference weights are used; finally, the non-dominated solution E with the smallest score is selected. As a planning scheme for air-ground collaborative missions, namely: ,in Let represent the independent variable that minimizes the objective function.

7. A heterogeneous unmanned swarm air-ground collaborative mission planning system, characterized in that, include: The model building unit is used to build an air-ground cooperative mission planning model for multi-objective optimization based on the parameters of UAVs and unmanned vehicles in a heterogeneous unmanned swarm. The objective function in the air-ground collaborative task planning model includes: a total energy cost objective function that needs to be minimized simultaneously. Total Time Penalty Function The constraints in the air-ground collaborative task planning model include: task access type constraints, load capacity constraints, and time window constraints. The sheep flocking optimization unit is used to solve the constructed multi-objective function and problem constraints using a pre-modified discrete multi-objective sheep migration optimization algorithm, and obtain multiple non-dominated solutions in a specified storage area. Specifically, it includes: 3.

1. Treating each sheep as a candidate solution individual, initializing the sheep flocking algorithm parameters, and randomly generating a task planning encoding sequence for each sheep as an initial candidate solution; the initialization of sheep flocking algorithm parameters includes: population size. Maximum number of discrete searches L during the grazing phase; maximum number of algorithm iterations. External archive B and archive capacity ; 3.

2. Decode each candidate solution in the initial candidate solution into a task sequence for each unmanned equipment. Perform load capacity constraint checks, time window constraint checks, and access type constraint checks on each task sequence to obtain... and This forms a multi-objective fitness vector for candidate solutions. ; 3.3 The grazing phase involves performing a local search near the current solution to improve the task allocation and execution order of candidate solutions in the flock. An improved solution is obtained after the grazing phase. Specifically, based on grid congestion, a lead sheep solution is selected from the external archive B according to the low-congestion region priority rule of the target space. The lead sheep performs no more than L discrete searches. Each following sheep also performs no more than L discrete searches. Discrete searches use exchange, reversal, insertion, redistribution, and crossover operators to perturb the task planning encoding sequence and generate new candidate solutions. 3.

4. Merge the candidate solutions in the current flock with the original B, and update B according to the Pareto dominance relationship; 3.

5. Execute the processing of the migration phase, which is used to guide the flock out of the local area and expand the search range when the grazing phase cannot continue to generate non-dominated solutions effectively; and update B, and when the maximum number of iterations is reached, output the non-dominated solutions in the updated B. The discrete multi-objective sheep migration optimization algorithm is an algorithm based on a multi-objective optimization air-ground cooperative task planning model and attribute correction of unmanned swarms. The planning scheme selection unit is used to perform multi-objective comprehensive scoring on each non-dominated solution in the specified storage area, and determine the planning scheme for the heterogeneous unmanned cluster air-ground collaborative task based on the scoring results and the air-ground collaborative task scenario.

8. A drone device, characterized in that, include: The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory, specifically performing the steps of the heterogeneous unmanned swarm air-ground collaborative task planning method according to any one of claims 1 to 6.

9. A computer storage medium, characterized in that, include: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the heterogeneous unmanned swarm air-ground collaborative task planning method as described in any one of claims 1 to 6.

Images