Unmanned aerial vehicle cluster load resource planning method based on multi-attribute coding and parallel variable neighborhood search

Abstract

The invention provides an unmanned aerial vehicle cluster load resource planning method based on multi-attribute coding and parallel variable neighborhood search, which belongs to the technical field of unmanned aerial vehicle resource planning, and comprises the following steps: firstly, determining target characteristics and task requirements, and determining unmanned platforms and available resource attributes; calling a load configuration algorithm to generate a configuration scheme, adopting multi-attribute coding to represent the configuration scheme, preferentially constructing an initial scheme based on task requirements, then adopting a self-adaptive variable neighborhood multi-population parallel evolution method to solve, generating a task configuration scheme meeting an optimization target and constraint conditions, and uploading the task configuration scheme; determining whether to approve to use the scheme or not by using expert experience or a decision support system according to the feasibility and efficiency of the scheme and the integrating degree of the scheme and a target; starting from the rationality and economy of the load resource allocation method for the unmanned aerial vehicle cluster, a higher cost-effectiveness ratio is realized on the premise of ensuring the task efficiency.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) resource planning technology, specifically, it relates to a UAV swarm payload resource planning method based on multi-attribute encoding and parallel variable neighborhood search. Background Technology

[0002] As unmanned aerial vehicle (UAV) platforms develop towards miniaturization, intelligence, and low cost, they leverage their advantages of flexible resource allocation, low manufacturing costs, and high mobility. Through differentiated configuration of payload resources, they achieve dynamic combination and collaboration of functions across different platforms, to some extent replacing the mission execution capabilities of large UAV platforms and gradually becoming an important means of warfare. However, accurately characterizing the matching and interaction relationship between payload resources and mission requirements, and rationally and efficiently obtaining payload resource configuration schemes for UAV swarms to maximize mission effectiveness under constraints such as the increasing number of mission platforms, resource category coupling, limited payload capacity, and diversified mission resource requirements, is becoming a challenging problem that needs to be solved.

[0003] According to publicly available information, countermeasures against drone swarms can be broadly categorized into three types: detection and identification, platform destruction, and link interference. For detection and identification countermeasures, priority should be given to drone platforms with low-scattering materials and long endurance, utilizing terrain-detection blind spots to reduce the probability of radar detection. For destruction countermeasures, robust drone platforms should be selected, and the number of platforms should be appropriately increased based on estimated destruction probabilities, while employing a robust control architecture to ensure overall mission effectiveness. For link interference countermeasures, drone platforms equipped with multiple communication and positioning payloads should be prioritized to enhance resistance to link interference. Based on the above analysis, before mission execution, it is necessary to select from various types and quantities of available payload resources and rationally allocate them to drone platforms with different mission capabilities to ensure multi-type mission capabilities against high-value target sets.

[0004] Payload resource allocation methods for UAV swarms must meet the requirements of rationality and economy. Rationality means that resource allocation should be based on task requirements set by expert experience or other means, clarifying the matching relationship between the platform and the carryable payload resources, and improving task execution efficiency in a task-demand-oriented manner while meeting the constraints of the platform's task capabilities. Economy means that the payload resource allocation solution method needs to optimize different types and numbers of payload resources carried by the UAV swarm, achieving a higher cost-effectiveness ratio while ensuring task efficiency. Summary of the Invention

[0005] Based on the above, this invention proposes a payload resource planning method for UAV swarms based on multi-attribute encoding and parallel variable neighborhood search. Starting from the rationality and economy of payload resource allocation methods for UAV swarms, a payload resource allocation method that meets mission requirements is designed.

[0006] This invention is achieved through the following technical solution: a method for planning payload resources in UAV swarms based on multi-attribute encoding and parallel variable neighborhood search. The method specifically includes the following steps: Step 1. Determine target characteristics and mission requirements: The command center, combining reconnaissance information and expert experience, defines a set of target mission attributes, including the resource requirements, expected benefits, and mission priority for each mission; Step 2. Determine the attributes of unmanned platforms and available resources: Clarify the task capabilities of different types of unmanned platforms and the relevant attributes of available resources, and archive the interaction relationships between different elements; Step 3. Call the load configuration algorithm to generate a configuration scheme: The load configuration algorithm uses multi-attribute encoding to represent the configuration scheme, constructs an initial scheme based on task requirements, and then uses an adaptive variable neighborhood multi-population parallel evolution method to solve it, generating a task configuration scheme that meets the optimization objective and constraints and uploading it; Step 4. Expert review and approval: Using expert experience or decision support systems, determine whether to approve the use of the plan based on its feasibility, efficiency, and alignment with the objectives.

[0007] Further, in step 1, Define target set For each target to be executed T i Its relevant attributes during the resource allocation phase can be represented by the set in Formula 1: (1) in This indicates the task priority of the objective. This represents a vector of rewards that can be obtained by performing different types of tasks on the current objective; This indicates the different task types that need to be performed to achieve the current goal. This indicates the predetermined damage threshold for the current target. Indicates whether the target has been destroyed; Indicates completion of the objective The corresponding set of expected resource vectors required for each task. Indicates the types of resources present in the resource pool; This indicates the degree of threat encountered from the target when performing the corresponding task.

[0008] Furthermore, in step 2, Collection of drone platforms For any unmanned platform Its relevant attributes during the resource allocation phase are represented by the set in Formula 3: (3) Indicates platform Whether or not to participate in the execution of a specific objective or task; Indicates platform type Indicates platform The task target number to be executed. This determines the priority of the task. For task type, For the cost of using the platform, The maximum number of resources that can be carried is, The maximum load weight is; This indicates the platform's resistance to damage. Indicates platform The types and quantities of resources currently carried; The available resource set S is assumed to form a resource pool, consisting of selectable reconnaissance, strike, assessment, and communication resources. The corresponding resource categories and resource quantities are represented as follows: and .

[0009] Furthermore, in step 3, the load configuration algorithm is implemented based on a load resource configuration optimization model, the construction of which includes: Define the decision matrix of the drone swarm as follows , x ij For binary decision variables, 1 represents the platform. Execution Objectives Task, 0 indicates not to be executed; Define a set of constraints, including task completion constraints, performance constraints, resource constraints, and scheme redundancy constraints. The objective function is defined based on a set of constraints, aiming to minimize the cost of drone deployment and resource costs while maximizing mission benefits.

[0010] Furthermore, in step 3, the multi-attribute encoding representation configuration scheme specifically involves: constructing... The matrix is ​​a two-dimensional matrix, with each row representing the payload configuration status of a drone platform. The matrix fields include platform participation status, drone index, platform type, target number, target priority, task type, types and quantities of various resources, damage probability, available payload space, and total cost. The target and resource information for platforms that do not participate in missions is set to zero, indicating that the drone has not been assigned any mission.

[0011] Furthermore, in step 3, the specific steps of constructing the initial solution based on task requirements include: Initialize the all-zero encoding matrix, configured platform flag vector, target requirement completion flag vector, and target construction sequence list; Traverse the target construction sequence list. For targets that are not completed and have not reached the maximum number of constructions, randomly select a platform from the available platforms, update the platform status, target and task type information of the encoding matrix, construct a resource vector by combining the resource pool, the maximum number of platform mounts and load constraints, and update the resource pool and target requirements. Construction stops when all objectives are achieved, no platform is available, or no resources are available. This process is repeated multiple times to generate subpopulations and the complete population set.

[0012] Furthermore, in step 3, the specific process of multi-population parallel evolution includes: The total population is divided into multiple subpopulations, and each subpopulation independently performs selection, crossover, and mutation evolution operations; A gene pool is constructed to store the fitness, subpopulation index, and individual index of all subpopulations. Based on information sharing mechanism, neighborhood structure design and adaptive neighborhood selection, solution reception and judgment mechanism, and chromosome repair strategy, after normalizing the fitness, high-quality individuals are selected to form a replacement set through roulette wheel selection. Each subpopulation randomly selects a corresponding number of individuals to form the replaced set. The replacement set is used to update the subpopulation. Finally, a task configuration scheme that meets the optimization objective and constraints is generated and uploaded.

[0013] A payload resource planning system for UAV swarms based on multi-attribute encoding and parallel variable neighborhood search; The system includes a target task module, a platform and resource module, a payload configuration module, and an approval module. The target task module is used to determine target characteristics and task requirements: the command center combines reconnaissance information and expert experience to define a set of target task attributes, including the resource requirements, expected benefits and task priority of each task. The platform and resource module is used to determine the attributes of unmanned platforms and available resources: clarify the task capabilities of different types of unmanned platforms and the relevant attributes of available resources, and archive the interaction relationships between different elements; The load configuration module is used to call the load configuration algorithm to generate a configuration scheme: the load configuration algorithm uses multi-attribute encoding to represent the configuration scheme, constructs an initial scheme based on task requirements, and then uses an adaptive variable neighborhood multi-population parallel evolution method to solve the problem, generating a task configuration scheme that meets the optimization objective and constraints and uploading it. The approval module is used for expert review and approval: based on the feasibility, efficiency and alignment with the objectives of the proposed solution, experts use their experience or decision support systems to decide whether to approve its use.

[0014] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0015] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] Beneficial effects of the invention This invention addresses the problems of unreasonable and inefficient payload resource allocation under conditions of heterogeneous mission capabilities and diverse target requirements in UAV swarm payload resource planning. First, by analyzing the coupling relationship between the UAV platform and mission attributes, the resource allocation problem is modeled as a multi-factor coupled bivariate nonlinear integer programming problem. Considering multiple constraints such as platform payload capacity, mission resource requirements, and total resource volume, a multi-objective optimization model under multiple constraints is constructed. Based on this model, a hybrid genetic adaptive large neighborhood search method is designed. Through multi-attribute encoding, demand-oriented heuristic initialization, and an adaptive neighborhood selection mechanism, accumulated experience is fully utilized during the iteration process to effectively improve the exploration efficiency of the knowledge space, thereby solving for a resource allocation scheme that meets mission requirements. Experimental results show that this method achieves a good balance between mission benefits and configuration costs, improves the overall combat effectiveness of small UAV swarms in multi-mission execution, and demonstrates good scalability and efficiency under missions of different scales. Attached Figure Description

[0017] Figure 1 This is a flowchart of the drone swarm payload configuration process of the present invention; Figure 2 Here is the algorithm flowchart for HPGALNS; Figure 3 This is a chromosome coding strategy based on multi-attribute features; Figure 4 This is a schematic diagram of local crossover and mutation operations; Figure 5 For the overall gene neighborhood structure; Figure 6 It is a local gene neighborhood structure; Figure 7 This is an initialization method based on a multi-attribute encoding strategy; Figure 8 Resource load configuration method; Figure 9 Strategies for chromosome resource repair; Figure 10 For adaptive large neighborhood search. Detailed Implementation

[0018] 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.

[0019] Unless otherwise specified, the experimental methods used in the following examples are conventional methods. Unless otherwise specified, the materials, reagents, methods, and instruments used are all conventional materials, reagents, methods, and instruments in the art, and can be obtained commercially by those skilled in the art.

[0020] This invention proposes a method for planning UAV swarm payload resources based on multi-attribute encoding and parallel variable neighborhood search, such as... Figure 1 The following are included: Step 1: Determining Target Characteristics and Mission Requirements. The command center, combining reconnaissance information and expert experience, defines a set of target mission attributes. These attributes include key details such as the resource requirements, expected benefits, and mission priority for each mission.

[0021] Define target set For each target to be executed T i Its relevant attributes during the resource allocation phase can be represented by the set in Formula 1: (1) in The task priority of the target can be preset by combining target reconnaissance information, expert experience and knowledge, and task requirements. This represents the vector of rewards that can be obtained by performing different types of tasks on the current objective, and can be further represented as... Each task corresponds to a different type of mission, and the specific expected reward score for each mission can be obtained through prior reconnaissance and expert evaluation. This indicates the different types of tasks that the current target needs to perform, corresponding to reconnaissance, strike, assessment, and communication tasks in that order. This indicates the predetermined damage threshold for the current target, which can be preset based on mission requirements and expert experience. Indicates whether the target has been destroyed.

[0022] Indicates completion of the objective The corresponding set of expected resource vectors required for each task, where Indicates the execution target The number of the s-th type of resource required for the task. This indicates the types of resources present in the resource pool.

[0023] This indicates the degree of threat encountered from the target when performing a corresponding task, which can be further expressed as... Pre-sets can be made by searching for adversarial information databases or through experience-based predictions. Multiple countermeasures are typically present around mission targets, increasing the difficulty of mission execution. For reconnaissance, assessment, and communication missions, platforms need to maintain a relative distance from the target, making them vulnerable to various threats; while in strike missions, platforms are more susceptible to combined interference from multiple countermeasures. Since the damage probability of actual countermeasures is influenced by a combination of factors, establishing an accurate damage probability model is very difficult. To simplify the analysis, the damage probabilities of different countermeasures are simplified to highlight the differences between countermeasures while avoiding the analytical complexity caused by too many coupling factors. Different Gaussian probability distributions of varying shapes are used to differentiate the damage probabilities of different platform types, as shown in Equation 2. Furthermore, considering the platform's resilience, the interaction relationships can be represented using a two-dimensional matrix.

[0024] (2) Where K g Representing types A and K e Representing type B, K a Represents the use of type C, V type This indicates the type of platform participating in the mission. For soft-kill tactics such as link interference, it is assumed that the platform can automatically use the corresponding communication signaling system and navigation method to counter them.

[0025] Step 2: Determine the attributes of unmanned platforms and available resources. Clearly define the mission capabilities of different types of unmanned platforms and the relevant attributes of available resources, ensuring that the interaction relationships between different elements are fully considered and accurately documented during the mission planning phase.

[0026] Collection of drone platforms For any unmanned platform Its relevant attributes during the resource allocation phase can be represented by the set in Formula 3: (3) Indicates platform Whether or not to participate in the execution of a specific objective or task; Indicates platform The type is determined by the currently available types of UAVs. In practical applications, platforms with different maneuverability undertake different functional responsibilities. For example, quadcopter platforms are used as central nodes for information transmission to ensure the stability of data transmission during missions; loitering munitions, small fixed-wing aircraft, and compound-wing platforms carry corresponding sensor equipment or weapon payloads to perform tasks such as reconnaissance, strike, damage assessment, and communication jamming, reflecting the heterogeneous performance characteristics of the platforms. Indicates platform The task target number (i.e., the target) to be executed. This refers to the priority of the task; the corresponding task type. These correspond to reconnaissance, strike, assessment, and communication tasks, respectively. The platform's usage cost is... The maximum number of resources that can be carried is The maximum load weight is ; This represents the platform's resilience to damage, i.e., its ability to continue executing tasks after being attacked. It can be estimated based on the platform's fuselage materials and structural design, and different normal distributions are used for approximation here.

[0027] The platform currently carries the following types of resources: in Indicates platform The first one carried l The first task corresponding to this class s Number of load types.

[0028] The available resource set S is assumed to form a resource pool, consisting of selectable reconnaissance, strike, assessment, and communication resources. The corresponding resource categories and resource quantities are represented as follows: and .

[0029] To strike payload set For example, among which This represents the set of attributes for each type of offensive payload.

[0030] The subscript 2 indicates that the payload belongs to the attack type. These represent the weight, quantity, and unit cost of this type of load, respectively. This indicates the probability of damage to each target by this type of payload. The properties of reconnaissance, assessment, and communication payloads can be referenced from the definition of strike payloads, but their probability of damage to the target... The probability of damage is zero. The set of weights for all types of resources is represented by M.s .

[0031] Depending on the type of resources carried out, the unmanned platform ensemble V exhibits significant resource heterogeneity, and can be categorized as follows: reconnaissance drone ensemble These drones can carry portable reconnaissance equipment such as visible light detectors and infrared sensors to perform tasks such as target search and status detection within a designated area.

[0032] attack drone collection These drones carry different types and numbers of weapon payloads to carry out strike missions against predetermined targets, and may launch suicide attacks on targets if necessary. Assessment of drone ensemble These drones carry assessment equipment such as visible light detectors to measure the status of damaged mission targets and use communication relay drones to return the status information to the ground command center.

[0033] Communication drone collection These drones carry a variety of portable communication devices, including radio modules, cellular communication modules, and satellite communication modules, to ensure the exchange of local information between clusters and the transmission and reception of control commands between the cluster and the rear ground station.

[0034] It is worth noting that the unmanned platform was not carrying any resource payload. Previously, it did not have task category characteristics. Only when the specific task is determined to be performed Only then will they be categorized into the corresponding drone type set.

[0035] Simultaneously, all platforms will be equipped with basic communication payloads to ensure the exchange of local information, and will selectively carry anti-interference payloads on top of this. Furthermore, various types of platforms performing different tasks collectively form the task platform set V. T Represented as Formula 4: (4) Step 3: Invoke the payload configuration algorithm. Use the appropriate payload configuration algorithm to generate and upload the corresponding task configuration scheme.

[0036] From the perspective of resource allocation objectives, the primary task of UAV swarms is to carry resource payloads that meet the requirements of mission execution, and on this basis, improve the cost-effectiveness ratio. Through the analysis of the interaction between available payload resources S, target mission characteristics T, and unmanned platform characteristics V in steps 1 and 2, a resource allocation model is further established with the goal of maximizing the expected mission efficiency.

[0037] Therefore, the decision matrix of the drone swarm is first defined as follows: Based on this, a set of relevant task constraints C is defined from aspects such as task completion, performance limitations, resource limitations, and solution redundancy. m Specifically: Mission Completion: This constraint can be further considered from two aspects: first, meeting the predetermined damage threshold; and second, ensuring resource requirements are matched. For unmanned platforms carrying attack resources, the total number of payloads should be estimated based on the kill probability of different payloads against targets and the target damage threshold. Assume the damage matrix of the weapon payloads used for attack against different targets is as follows: ,in Indicates the first i Type of weapon against the first j The probability of damage to a target. This probability distribution can be obtained through statistical analysis of a large amount of experimental data. For ease of analysis, it is assumed that it follows an interval... The data is uniformly distributed on the target surface, and the interval data can be set based on expert prior knowledge or historical experimental data. Furthermore, the payloads of each weapon are relative to the target. The probability of damage to each drone is independent. Based on the above analysis, the attack drone ensemble The damage constraints on the target imposed by all weapon payloads can be expressed as Equation 5. (5) in Indicates the target of the attack The number of drones to meet , This represents the damage threshold set for the target; an integer decision variable. Indicates unmanned platform V i Carrying for striking targets The t The number of loads.

[0038] To ensure the matching degree of non-strike resources and the redundancy of the scheme, the constraints in Formula 6 can be further defined to ensure that the payload carried by the UAV swarm can cover or meet all the mission resource requirements of the target to be executed. Indicates unmanned platform Whether to participate in the execution of objectives The task is a binary decision variable, meaning that each unmanned platform can only choose to participate or not participate when executing the task. ≥1 indicates the resource redundancy of each target task. This indicates a rounding up operation, ensuring that resource allocation meets or exceeds the needs of each task.

[0039] (6) Furthermore, considering the limited destructive power of the attack payload, it is assumed that the unmanned platform performing the attack mission can only engage with a single target, i.e., it cannot perform strike missions against multiple targets. Therefore, the constraints in Equation 7 are established. For other types of unmanned platforms, under sufficient resource conditions, they can participate in the execution of multiple targets and multiple missions.

[0040] (7) Performance limitations: To ensure the performance of unmanned platforms The number of resources carried shall not exceed its maximum mount count. The following constraints are given, as shown in Formula 8. This indicates the total number of all available resource categories.

[0041] (8) In addition, to ensure the unmanned platform The total weight of the resources carried shall not exceed its maximum load capacity. The performance constraints in Formula 9 are given. Indicates the first k The weight of the resource.

[0042] (9) Resource finiteness constraint: Considering the limited resource scale for load configuration, a total resource constraint is defined in Equation 10. Indicates the first l The first task type corresponding to the s The total amount of available resources.

[0043] (10) In addition, to prevent logical errors during resource allocation from causing the total number of participating platforms to exceed the maximum available number, cluster size constraints are defined in Formula 11.

[0044] (11) Redundancy constraints: To ensure the cluster can continue executing tasks even under destructive attacks, its overall damage level must first meet a safety threshold. S a Therefore, the constraints in Formula 12 were first defined at the platform physical level to ensure that the platform's security margins for executing different task types meet the requirements. (12) Furthermore, to ensure the cluster can maintain basic communication even under attack, the number of platforms performing communication tasks within it needs to satisfy the constraint in Formula 13, where... Nc This indicates the number of neighbors a single communication node can support. S c For redundant extended scalars.

[0045] (13) By analyzing the characteristics of key elements in the mission environment, the optimization objectives of resource allocation for heterogeneous UAV clusters are summarized as: (1) maximizing mission benefits; (2) minimizing UAV deployment costs and (3) minimizing resource costs.

[0046] Different optimization objectives are related by weights Establishing connections allows for adjusting weights to meet different objectives and needs in actual combat. To distinguish between constraint violations in configuration schemes, a penalty function-based method is used to differentiate infeasible solutions. The above analysis is summarized as the optimization objectives in Formulas 14 and 15.

[0047] (14) The optimization objective in Equation 14 above aims to minimize mission cost. Part 1 represents the total cost of the UAV swarm deployed in the combat operation; Part 2 represents the total cost of payload resources used in the combat operation. (15) In Equation 15 above, the optimization objective is to maximize the task benefits. The first part represents the available payload space of the drone swarm; the second part represents the cumulative benefits that can be obtained by performing different tasks of the objective.

[0048] In summary, the heterogeneous UAV resource allocation problem can be reduced to a mixed-integer nonlinear programming problem under multiple constraints, and its mathematical model is expressed as Equation 16, where... Indicates the penalty coefficient. DOV This indicates the degree of violation of each constraint.

[0049] (16) The solution space of the optimization problem P1 increases exponentially with the increase in the scale of UAVs, the number of tasks, and the types of resources, making it a typical NP-hard problem. Traditional exhaustive and exact algorithms often fail to provide sufficiently efficient solutions in practical applications. Therefore, in order to obtain reasonable resource allocation results within a limited time, a heuristic optimization algorithm is proposed, taking into account the heterogeneous characteristics and task requirements of UAV swarms, aiming to improve the efficiency of the resource allocation process.

[0050] Considering the advantages of genetic algorithms, such as requiring few parameters, ease of implementation, and strong scalability, they have been widely used to solve various combinatorial optimization problems with NP-hard properties. To solve problem P1, a hybrid parallel genetic algorithm with adaptive large neighborhood search based on multi-attribute encoding (HPGALNS) is proposed. This method includes a chromosome encoding / decoding strategy, a population initialization method considering task requirements, parallel evolutionary local neighborhood structure design and selection among subpopulations, and an adaptive neighborhood selection mechanism. Its flowchart is shown below. Figure 2 As shown.

[0051] Specifically, firstly, considering multiple factors such as the heterogeneity of UAV platforms, the diversity of mission requirements, and the diversity of available payloads, a multi-feature fusion coding strategy is designed. Taking into account mission characteristics and UAV platform features, with the goal of maximizing the fulfillment of mission requirements, multiple feasible solution populations are initialized. As the foundation for parallel iteration, a feasible solution update framework from genetic algorithms is adopted to explore more profitable feasible solutions. To ensure that individual solutions do not violate task constraints, a corresponding chromosome repair mechanism is designed to guarantee that the generated offspring individuals are valid and satisfy the constraints. This mechanism can automatically correct solutions that do not meet the constraints during the genetic algorithm process, avoiding the search for invalid solutions. In addition, an information sharing mechanism is introduced, which effectively utilizes the different state information of the subpopulation in the solution space by establishing a high-quality resource pool and using a resampling and replacement method. To further improve the quality of solutions, various neighborhood structures N are designed. k This method is used to help explore local optima. It combines the neighborhood structure adaptive selection method based on historical belief sets with the probabilistic acceptance of inferior solutions mechanism to conduct local searches and continuously adjust and optimize the structure of solutions. While ensuring the diversity of global searches, it improves the accuracy and efficiency of local searches.

[0052] Step 3.1 Configuration scheme representation method based on multi-attribute encoding: To ensure that the chromosomes used in the optimization process can reasonably reflect the actual resource allocation scheme, a chromosome encoding strategy based on multi-attribute features was designed, such as... Figure 3 As shown.

[0053] Each row represents the payload configuration status of a drone platform, thus providing a comprehensive picture of the drone's participation and resource allocation in the mission. Specifically, it represents a set of alternative resource configuration schemes. It is a size of A two-dimensional matrix, whereN V S represents the number of allocable unmanned platforms, and S represents the number of types of available payload resources. N g This indicates status information about the current task platform, excluding resource payload information, and can be dynamically adjusted according to different task types. Selected in the embodiment N g =9, the meaning of the encoding format is as follows: The first column indicates whether the drone is currently participating in the overall mission execution. This is achieved through binary decision variables. (right answer (using a flag bit in the array) to indicate, if This indicates that the platform participates in the task, otherwise... This indicates that the platform does not participate in the task.

[0054] The second and third columns represent the index and category number of the current drone, respectively. This helps identify the identity and type of drone.

[0055] Columns four through six indicate the priority of the drone pair. goal Execution type is The mission, including reconnaissance, strike, and assessment, indicates the mission allocation of the drone.

[0056] The middle columns sequentially represent the types and quantities of resources carried, according to the order of reconnaissance, strike, assessment, and communication.

[0057] The last three columns represent the probability of the drone damaging the target, the available payload space, and the total cost of carrying the corresponding resources. This information helps to assess the effectiveness of the mission and the consumption of resources.

[0058] When a drone is not participating in a mission, its target and resource information are both set to zero. For example, the drone with index 3 in the third row has zero target and resource information, indicating that the drone has not been assigned to any mission.

[0059] Figure 3 The encoding method shown converts the UAV state information set Target task information set The resource information set S, along with UAV performance constraints and overall resource constraints, are integrated under a unified information structure. This encoding strategy not only intuitively displays the resource allocation process but also clearly expresses the coupling relationship between resources and the platform. In this way, the interactions between various factors can be comprehensively reflected, providing necessary support for the subsequent design of evolutionary mechanisms.

[0060] Step 3.2 Construction method of configuration scheme based on task requirements After defining the feasible solution encoding strategy and gene structure, a feasible solution construction method based on task requirements and applicable to resource configuration of heterogeneous UAV swarms is further presented. The implementation process of this method is as follows: Figure 7 Algorithm 2-1 is shown. Its core idea is to comprehensively consider the target resource requirements. Drone attributes Based on the available resource pool size S, and drawing on the sequential construction method, starting from the task resource requirements, we gradually select and load the corresponding task resources onto the corresponding UAV platform to maximize the satisfaction of the target's resource requirements.

[0061] Specifically, input variables This is used to limit the number of drones participating in a single objective mission to a feasible range, ensuring resource allocation for most objectives when there are sufficient platforms, and avoiding the situation where most resources are concentrated on a few objectives. The initialization phase first sets the used drone flag vectors ( The first column element in the table (defaults to) And update when a specific drone is selected to participate in the corresponding mission. (Line 9). Target requirement completion vector This means that all targets require corresponding resources to be matched and updated when resource configuration is complete. (Line 15). Target construction list This indicates the order in which the resource construction process is performed on the task objective. Two construction orders are provided here: Build in descending priority order This method maximizes mission efficiency by prioritizing resource allocation to mission objectives with greater potential gains when available payload resources are limited.

[0062] Random construction This method can maintain the balance of the configuration scheme across different objectives.

[0063] In constructing individuals During the process, the two are selected using a random number threshold, thereby increasing the diversity of configuration options. It should be noted that for each target... The process of constructing its own resource vector (lines 9-17) will only be executed when available resources are not empty, available drones exist, and there is a mission requirement (line 8), and the available resource pool S and target mission requirements will be updated. (Line 12), where the pseudocode for the specific constructor (line 11) is as follows: Figure 8Algorithm 2-2 shows the process. The construction process ends when all target task requirements are met, all UAVs are occupied, or there are no available resources (lines 20-22). Otherwise, the process proceeds sequentially through the target set. The same construction process is adopted.

[0064] It should be noted that, through Figure 7 Algorithm 2-1 generates the initial individuals This ensures that it does not violate the constraints of Formulas 5 and 10. However, for the overall task resource constraint in Formula 6, it is also necessary to consider whether the resources are sufficient. If resources are sufficient, this constraint can be satisfied. If resources are scarce, different configuration schemes need to be optimized to obtain the most efficient resource allocation result. This is achieved by repeatedly executing Algorithm 2-1. N pop Next, a subpopulation can be obtained. Furthermore, by repeating the execution... N Sub This allows for the construction of a complete population set. P all .

[0065] exist Figure 8 In Algorithm 2-2, it is necessary to first calculate the number of resources already installed on the current platform. And the remaining number of available hanging points Based on this, the required number of resources for each type... Both greedy and incremental strategies are used for adjustment (lines 6-11) to adjust the resource matching rate of a single platform. Further, it is determined whether the resource matching rate exceeds the size of each resource S in the available resource pool (line 16), ultimately yielding the resource allocation vector. By employing the resource vector construction process described above, local details of the configuration scheme can be adjusted, thereby improving the diversity of configuration schemes to a certain extent.

[0066] Step 3.3, Multi-population Parallel Evolutionary Method Based on Adaptive Variable Neighborhood Analysis of the encoding strategy matrix reveals that differences between feasible solutions in the solution space are typically manifested through local features. To more effectively explore different resource combinations, subpopulations with different state distributions are initialized in the solution space. Furthermore, by leveraging the superior parallel computing characteristics of genetic algorithms, the search area can be expanded by interacting with local information during the search process, thus avoiding getting trapped in local optima.

[0067] To avoid getting trapped in local optima with a single selection strategy, an adaptive selection strategy adjustment mechanism based on distribution entropy is adopted. Population entropy is used to reflect the distribution information of solutions, and the selection strategy is dynamically adjusted to better balance development (developing local areas) and exploration (expanding the search space). In step 3.1, the chromosome encoding strategy based on multi-attribute features, whether the UAV is activated, the target mission number, and the resources it carries play a crucial role in the quality of the final resource allocation scheme.

[0068] Since different task types exist in the encoding structure, in order to reduce individual violations of task constraints caused by cross-operations, it is first necessary to determine the different individuals separately. and Execute the same task type set of gene locations and Based on this, two operators are defined: global gene exchange and local gene exchange. The first operator generates a new individual by exchanging all gene coding structures of the same task type on different individuals within the same population; the second operation generates a new individual by exchanging resource coding fragments of the same task type on different individuals within the same population.

[0069] In addition, to improve the diversity of configuration schemes, three types of mutation operations are proposed. The first mutation operation modifies the current drone's participation in mission execution, attempting to reduce the number of drones to achieve higher mission benefits. The second mutation operation changes the target number of the current drone's mission to explore whether it is suitable for different target requirements. It is worth noting that this mutation operation does not involve changing the mission type to ensure the compatibility between resource structures. The third mutation operation changes the number of mission resources carried by the current drone to verify whether mission benefits are better under different resource configurations. The above crossover and mutation diagrams are shown below. Figure 4 As shown.

[0070] Neighborhood Structure Design: To further improve the efficiency of resource allocation schemes for heterogeneous UAV swarms, seven neighborhood structures were designed for the coding scheme to enhance the algorithm's local search capability. The core idea is a neighborhood structure... The local optimal solution is not necessarily another neighborhood structure. The optimal solution can be found by oscillating the current optimal solution across different neighborhood structures to escape local optima. Since individuals with higher fitness have greater improvement potential, the variable neighborhood search in this invention only targets the top-ranked individuals, thereby reducing computation time.

[0071] Based on the coding structure defined above and Formula 2, it can be seen that the key to changing the configuration efficiency lies in adjusting... The matching relationship between the entire UAV and the mission, as well as the resources carried by the UAV, were considered. Therefore, a set of neighborhood operators was designed from two perspectives. N k This ensures that it can perform local searches of the solution space at different scales, as illustrated in the diagram below. Figure 5 and Figure 6 As shown. It is important to note that the neighborhood structure is for self-updating of existing configuration results within a single chromosome, rather than for the interaction of configuration results between multiple chromosomes.

[0072] Overall gene replacement operator Through subpopulations The set of individuals with the highest fitness is obtained by sorting them in descending order. And randomly select parental individuals from them. Furthermore, this individual corresponds to the set of drones participating in the mission. Randomly select the gene index that needs to be replaced. They also exchange information about target tasks and resources in the corresponding genes.

[0073] Whole gene insertion operator If there are payload resources in the resource pool (S≠ And there are instances where some drones are idle. 0) ≠ There are still unmet target task requirements. In the case of (≠0), then in the idle drone collection The system randomly selects a UAV index as the corresponding gene locus and initializes its corresponding task type and resource payload in resource pool S. .

[0074] Whole gene splitting operator This neighborhood structure attempts to include drones participating in the mission but with larger available payloads. V i Onboard mission resources Reassign to drones of the same type and objective and exclude yourself from the mission. 0, thereby reducing task costs. For example... Figure 5 The communication payload of drone number 1 is transferred to the communication payload of drone number 5 to maximize the payload utilization of a single drone platform.

[0075] Local gene substitution operator To minimize mission costs, this neighborhood structure is implemented across the available drone ensemble. 1. Select the drone category as described above. By replacing the current drone with different drones, we can explore whether different deployment options can bring higher mission benefits.

[0076] Local gene perturbation operator To make the best use of the drone payload space, this neighborhood structure indexes the currently selected genes. The number of resource loads is randomly changed without violating relevant constraints to explore whether there is a more efficient configuration.

[0077] Local gene exchange operator This neighborhood structure selects to perform the same task type. The candidate set of drones And select individual indexes from them to participate in resource exchange. and and according to the commutation operator It exchanges some of its resource loads.

[0078] Local gene mutation operator To encourage the algorithm to prioritize higher-priority objectives and thus achieve better task outcomes, this neighborhood structure is based on the currently executing task type. Based on the target set T, a set of candidate targets with the same task type is selected, and a new target with a priority no lower than the current priority is selected to replace the existing target.

[0079] Chromosome Repair Strategy: Considering that configuration schemes may violate task constraints during exchange and mutation operations, constraint violations are categorized as follows to avoid impacting search efficiency: (1) The number of resources carried exceeds the number of available attachment points; (2) The weight of resources carried exceeds the maximum payload; (3) The type of task performed by the UAV does not match the type of resources carried, such as platform V i The mission is being carried out for reconnaissance, but it's equipped with different resources. For this third scenario, the approach is to keep the mission type unchanged and perform a secondary initialization of the equipped resources within the public resource pool S. For the first two scenarios, the approach is... Figure 9 Repair strategies in Algorithm 2-3.

[0080] To prevent an infinite loop from occurring due to the lack of feasible repair strategies for the current individual, a maximum number of repair attempts is defined during the input phase. The current load weight was first calculated during the initialization phase. and total amount of resources carried If the difference between the two exceeds their respective performance limits (line 4), then resources will be used from existing resources. Perform the deduction (line 8) and update the current available resource pool S (line 9) and load status (line 10).

[0081] Adaptive neighborhood selection: To utilize the state information during the search process and enable the algorithm to prioritize searching neighborhood structures with a high probability of potential updates, a set of historical beliefs was established for each subpopulation. . The best individual in the subpopulation is stored in the middle. The number of times the best individual was successfully updated and the corresponding score were recorded when different neighborhood structures were implemented. Then, Formula 17 was used to update the neighborhood in the next iteration. The probability of being selected.

[0082] (17) in, Representing neighborhood structure In the j The probability of being selected in the +1st iteration, where r ∈ [0, 1] is a constant that is dynamically adjusted with each iteration. Representing neighborhood structure Up to the j The local optimum was successfully updated during the next iteration. Number of times, neighborhood structure The cumulative score up to the j-th iteration.

[0083] This represents the highest score in the neighborhood structure. The following scoring rules were designed: = 0: Neighborhood structure , i = 1, 2, ..., |N k The score was 0 during algorithm initialization. = 5: Neighborhood Structure , i = 1, 2, ..., |N k | at the j If the solution generated in the next iteration is the optimal solution for the subpopulation, then the neighborhood structure is worth 5 points. = 3: Neighborhood Structure , i = 1, 2, ..., |N k | at the j If the solution generated in the next iteration is better than the population average solution, then the neighborhood structure is 3 points. = 1: Neighborhood structure , i = 1, 2, ..., |N k | at the jIf the solution generated in the next iteration is better than the top 80% of the population solutions, then the neighborhood structure scores 1 point. The selected probability of each neighborhood is obtained from Formula 17 and then normalized. Each time the neighborhood is switched, a roulette wheel strategy is used to select the next neighborhood structure. The search process is as follows: Figure 10 Algorithm 2-4 is shown.

[0084] Information sharing mechanism: In order to make full use of search information among different subpopulations, information sharing mechanism is established in each subpopulation. After all individuals undergo three evolutionary operations, individual information needs to be shared. This process begins by constructing a gene pool containing all subpopulations. As shown in Formula 18, where N sub Indicates the number of subpopulations. N pop This indicates the number of individuals within each subpopulation.

[0085] (18) Formula 18 indicates that each row of the gene pool stores the fitness and index value of individuals from different subpopulations. To ensure that individuals with higher returns have a higher probability of information sharing, the first column of the gene pool's fitness vector is first processed. Normalization and selection using roulette strategy N rep Individual composition replacement set Secondly, in each subpopulation Random selection N rep Individual composition replaced individual set Finally, the corresponding index individuals in the subpopulation are replaced to complete the sharing of evolutionary information, and the replacement process is shown in Formula 19.

[0086] (19) Solution acceptance criteria: A probability-based acceptance criterion is adopted. If a temporary solution is generated during iteration... The benefits are greater than the current solution. Then use replace If the current solution Better than the globally optimal solution If the benefit of the temporary solution is lower than the current objective, then the global solution is replaced. . T t This represents the temperature control parameter, and its dynamic update process satisfies... , αThis indicates the cooling rate, thus avoiding getting stuck in local optima.

[0087] Once the number of consecutive non-improvement iterations exceeds the predefined parameter... Or the number of iterations exceeds the maximum feasible number. Or the runtime exceeds the maximum available time. The algorithm then stops. Refer to the detailed process. Figure 10 Algorithm 2-4.

[0088] Step 4: Expert Review and Approval. Using expert experience or decision support systems, determine whether to approve the use of the plan based on its feasibility, efficiency, and alignment with strategic objectives.

[0089] A payload resource planning system for UAV swarms based on multi-attribute encoding and parallel variable neighborhood search; The system includes a target task module, a platform and resource module, a payload configuration module, and an approval module. The target task module is used to determine target characteristics and task requirements: the command center combines reconnaissance information and expert experience to define a set of target task attributes, including the resource requirements, expected benefits and task priority of each task. The platform and resource module is used to determine the attributes of unmanned platforms and available resources: clarify the task capabilities of different types of unmanned platforms and the relevant attributes of available resources, and archive the interaction relationships between different elements; The load configuration module is used to call the load configuration algorithm to generate a configuration scheme: the load configuration algorithm uses multi-attribute encoding to represent the configuration scheme, constructs an initial scheme based on task requirements, and then uses an adaptive variable neighborhood multi-population parallel evolution method to solve the problem, generating a task configuration scheme that meets the optimization objective and constraints and uploading it. The approval module is used for expert review and approval: based on the feasibility, efficiency and alignment with the objectives of the proposed solution, experts use their experience or decision support systems to decide whether to approve its use.

[0090] The performance of the proposed method in solving the resource allocation problem of heterogeneous UAV swarms is verified in the embodiments. First, simulation test cases with different characteristic parameters are constructed, and then compared and analyzed with various algorithms to verify the solution capability of the proposed method under different task scales. The algorithm is implemented using Matlab2023a, and the hardware device is a computer with an AMD Ryzen 9 5950X processor (3.4 GHz), 32 GB of memory, and Windows 10 operating system.

[0091] Based on the above, a simulation instance generator is set up. This generator creates corresponding verification instances by inputting basic parameters such as UAV attribute V, resource attribute S, and target attribute T. Unmanned platform parameters. In each test instance, the number of platforms of each category is randomly generated while keeping the total number of platforms constant, i.e., satisfying |V sur |+|V wea |+|V val |+|V com | = |V| and ensure that the number of each category is not lower than the preset minimum value, as shown in the fifth column of Table 1, to guarantee the diversity and representativeness of the instances. Since the payload performance of some unmanned platforms may vary after mission execution due to differences in consumption levels and quality control, a truncated normal distribution is used. The effective payload was simulated, where [ a , b [This is the cutoff interval.] To reflect the differences in platform loads, the effective load attachment points of different platforms of the same type are within a certain range. The data is randomly generated within the specified range, and the platform's unit cost is also within the specified range. The data is randomly generated internally. Furthermore, to reflect the differences in damage resistance among different platform types, truncated normal distributions with different distributions are used. An approximate model is used to ensure that performance differences between platforms of the same type are reasonably reflected. Furthermore, the platform type can be specified within a range. Randomly generated. Table 1 provides a reference range for the generation information of different types of unmanned platforms participating in the mission. The relevant data can be increased or decreased according to the actual situation.

[0092]

[0093] Table 1 Available UAV Platform Parameters Mission Requirement Parameters: Assume that in a certain operation, our side has obtained prior information about enemy targets through reconnaissance, including their value, resource requirements, and mission type, and has set corresponding damage requirements based on expert experience. Simultaneously, mission priorities are set for different targets according to their importance and operational needs. The resource requirements for different mission types are randomly generated as positive integers within a set range, i.e., satisfying... It should be noted that the task type for a single target is randomly generated, which means it does not perform all types of tasks. Considering that targets in adversarial environments typically possess certain countermeasures, it is assumed that the threat level posed by the target to the UAV can be represented by a truncated normal distribution with different means and variances. A simulation was performed. The damage threshold was set within a specified range. Randomly generated within the specified range. Task priorities are assumed to conform to a certain interval. The distribution is uniform. The expected reward for the task is within a specified range. Randomly generated. Resource redundancy for a single task is also within a specified range. Randomly generated. The generated task attributes and resource requirements for a single objective are shown in Table 2.

[0094]

[0095] Table 2 Task attributes of the target to be executed Payload resource parameters: For different resource categories participating in the task, it is assumed that they can be within a specified range. Randomly generated. The quantity of each resource also falls within a predefined range. The data is generated randomly, and its weight and cost are simulated using different truncated normal distributions. The truncated intervals and data distribution characteristics are set by experts to meet specific needs.

[0096] The probability of damage to a target by offensive resources is uniformly distributed over a specified interval. The simulation's initialization scope can be set based on historical data evaluated by experts. Assuming there are sufficient available resources for the task, different resources can be expanded to adapt to different task requirements. Table 3 provides initialization examples of relevant attributes.

[0097]

[0098] Table 3 Available Load Resource Parameters In addition, to verify the resource allocation effect of the proposed method under different task requirements, three types of task scenarios, namely large, medium and small, were set up according to the number of targets, cluster size and resource type, and it was assumed that the corresponding resources were sufficient. The relevant parameters are shown in Table 4.

[0099]

[0100] Table 4 Parameter settings for different task sizes Simulation parameter settings. The key parameters in the simulation are summarized in Table 5. Considering the issue of excessively long solution times as the problem dimension increases, a maximum runtime was set. = 1500s, exceeding this threshold will cause the program to terminate and output the configuration result. To reduce the impact of randomness and improve the reliability of the results, each algorithm is executed 20 times for each test instance. To verify the effectiveness of the proposed algorithm, it was further compared with different solution methods. In addition, to verify the effectiveness of the parallel strategy, serial search was also tested. It should be noted that the characteristics of the research problem in this invention and existing literature are somewhat different, which also leads to differences in the encoding method of the solution. To ensure fairness, all methods adopt the above encoding method and feasible solution construction method and draw on their corresponding feasible solution search methods.

[0101]

[0102] Table 5 Simulation Parameters Evaluation metrics for the scheme. On the one hand, considering that the application background of this algorithm is for the offline planning stage before task execution; on the other hand, considering the significant differences in computation time depending on the device used, runtime is not considered or limited. To verify that the resource allocation scheme can meet the multiple constraints of this invention, the following evaluation metrics are defined: Resource matching degree. This indicator measures the gap between the non-attack-related resource requirements and the configured resources for the target's pending mission. Its definition is shown in Formula 20. It can be seen that the closer this indicator is to 1, the higher the resource matching degree. For attack-related resources, the ratio of the configured damage probability to the predetermined value is used for judgment.

[0103] Configuration Cost. Calculating the task cost of a configuration scheme directly using Formula 16 is an intuitive way to reflect the economy of the scheme and the stability of the solution method. Because of the priority relationship between target tasks, resource allocation tends to prioritize more important tasks to obtain higher rewards. This process may lead to configuration schemes violating constraints. Therefore, the convergence curve of the average task cost from multiple repeated trials and the presence of outliers in the task benefit distribution can be used to determine whether the obtained scheme meets the corresponding task constraints.

[0104] (20) To verify the rationality of this invention, multiple test instances were first generated under a small-scale task requirement using the parameter instance generation method of the embodiment, and their configuration results were analyzed as shown in Table 6. This table shows that different objectives have different task types and resource requirements (three columns on the left), and the four columns on the right provide detailed configuration results obtained using the method described in this paper.

[0105] Taking the resource allocation results for target 4 as an example, it can be seen that its resource requirements all meet the type constraints in Table 4, and its damage threshold is set to be greater than 85%, which can weaken the target's combat capability to a certain extent. The allocation results for assessment and communication resources show that they are both twice the required resources, satisfying the task completion constraint in Formula 6. In addition, the damage probability effect of the platform carrying strike resources on the target is 0.9071, which, compared to the predetermined damage threshold of 0.9051, also satisfies the constraint in Formula 5. At the same time, the scheme redundancy for each type of task is also greater than the set safety margin of 0.8, which are 0.9166, 0.9345 and 0.8421 respectively, satisfying the constraint in Formula 12. The total number of unmanned platforms participating in target 4 is 9. According to Formula 13 and combined with the number of neighbors in Table 5, it can be seen that at least 2 platforms are required to perform communication tasks, and the obtained scheme meets the task requirements. The resource allocation results for the other targets also show that the obtained schemes meet multiple constraints.

[0106]

[0107] Table 6. Verification of the rationality of resource allocation schemes under small-scale instances. This invention first analyzes the impact of UAV swarm countermeasures technology on the payload resource allocation stage and, based on a clear understanding of mission requirements, clarifies the fundamental principles that payload resource allocation should follow. Building upon this, a multi-level payload resource allocation solution framework is constructed to address the problems of low allocation efficiency and poor mission utility caused by the increased number of UAV platforms, differences in mission capabilities, and diversification of target requirements, leading to a larger solution space. This framework comprehensively considers various coupled constraints such as the payload capabilities of different UAV platforms, the interaction between mission objectives and resources, matching requirements, and the total amount of available resources. With the goal of maximizing mission benefits and minimizing mission costs, a multi-factor coupled bivariate nonlinear integer programming model is constructed. To address the NP-hard nature of this problem, a multi-attribute encoding strategy and a mission-demand-oriented heuristic allocation scheme initialization method are further designed to improve payload resource allocation efficiency. Furthermore, a parallel evolutionary mechanism and an adaptive variable neighborhood search strategy are designed based on a genetic algorithm. Utilizing experience accumulated during iteration, the search direction is adaptively guided, thereby enhancing the algorithm's exploration capability in the solution space and obtaining a payload resource allocation scheme that meets mission requirements.

[0108] Simulation results show that the multi-level payload resource configuration scheme proposed in this invention can achieve an effective balance between mission benefits and configuration costs. The test results under different mission scales also demonstrate its good scalability, providing an efficient reference scheme for payload resource configuration of UAV platforms.

[0109] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0110] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0111] The above provides a detailed description of the UAV swarm payload resource planning method based on multi-attribute encoding and parallel variable neighborhood search proposed in this invention. The principles and implementation methods of this invention have been explained. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for planning payload resources in UAV swarms based on multi-attribute encoding and parallel variable neighborhood search, characterized in that: The method specifically includes the following steps: Step 1. Determine target characteristics and mission requirements: The command center, combining reconnaissance information and expert experience, defines a set of target mission attributes, including the resource requirements, expected benefits, and mission priority for each mission; Step 2. Determine the attributes of unmanned platforms and available resources: Clarify the task capabilities of different types of unmanned platforms and the relevant attributes of available resources, and archive the interaction relationships between different elements; Step 3. Call the load configuration algorithm to generate a configuration scheme: The load configuration algorithm uses multi-attribute encoding to represent the configuration scheme, constructs an initial scheme based on task requirements, and then uses an adaptive variable neighborhood multi-population parallel evolution method to solve it, generating a task configuration scheme that meets the optimization objective and constraints and uploading it; Step 4. Expert review and approval: Using expert experience or decision support systems, determine whether to approve the use of the plan based on its feasibility, efficiency, and alignment with the objectives.

2. The planning method according to claim 1, characterized in that: In step 1, Define target set For each target to be executed T i Its relevant attributes during the resource allocation phase can be represented by the set in Formula 1: (1) in This indicates the task priority of the objective. This represents a vector of rewards that can be obtained by performing different types of tasks on the current objective; This indicates the different task types that need to be performed to achieve the current goal. This indicates the predetermined damage threshold for the current target. Indicates whether the target has been destroyed; Indicates completion of the objective The corresponding set of expected resource vectors required for each task. Indicates the types of resources present in the resource pool; This indicates the degree of threat encountered from the target when performing the corresponding task.

3. The planning method according to claim 2, characterized in that: In step 2, Collection of drone platforms For any unmanned platform Its relevant attributes during the resource allocation phase are represented by the set in Formula 3: (3) Indicates platform Whether or not to participate in the execution of a specific objective or task; Indicates platform type Indicates platform The task target number to be executed. This determines the priority of the task. For task type, For the cost of using the platform, The maximum number of resources that can be carried is, The maximum load weight is; This indicates the platform's resistance to damage. Indicates platform The types and quantities of resources currently carried; The available resource set S is assumed to form a resource pool, consisting of selectable reconnaissance, strike, assessment, and communication resources. The corresponding resource categories and resource quantities are represented as follows: and .

4. The planning method according to claim 3, characterized in that: In step 3, the load configuration algorithm is implemented based on a load resource configuration optimization model, the construction of which includes: Define the decision matrix of the drone swarm as follows , x ij For binary decision variables, 1 represents the platform. Execution Objectives Task, 0 indicates not to be executed; Define a set of constraints, including task completion constraints, performance constraints, resource constraints, and scheme redundancy constraints. The objective function is defined based on a set of constraints, aiming to minimize the cost of drone deployment and resource costs while maximizing mission benefits.

5. The planning method according to claim 4, characterized in that: In step 3, the multi-attribute encoding representation configuration scheme specifically involves: constructing... The matrix is ​​a two-dimensional matrix, with each row representing the payload configuration status of a drone platform. The matrix fields include platform participation status, drone index, platform type, target number, target priority, task type, types and quantities of various resources, damage probability, available payload space, and total cost. The target and resource information for platforms that do not participate in missions is set to zero, indicating that the drone has not been assigned any mission.

6. The planning method according to claim 5, characterized in that: In step 3, the specific steps of constructing the initial solution based on task requirements include: Initialize the all-zero encoding matrix, configured platform flag vector, target requirement completion flag vector, and target construction sequence list; Traverse the target construction sequence list. For targets that are not completed and have not reached the maximum number of constructions, randomly select a platform from the available platforms, update the platform status, target and task type information of the encoding matrix, construct a resource vector by combining the resource pool, the maximum number of platform mounts and load constraints, and update the resource pool and target requirements. Construction stops when all objectives are achieved, no platform is available, or no resources are available. This process is repeated multiple times to generate subpopulations and the complete population set.

7. The planning method according to claim 6, characterized in that: In step 3, the specific process of multi-population parallel evolution includes: The total population is divided into multiple subpopulations, and each subpopulation independently performs selection, crossover, and mutation evolution operations; Construct a gene pool to store the fitness, subpopulation index, and individual index of all individuals in the subpopulation; Based on information sharing mechanisms, neighborhood structure design and adaptive neighborhood selection, solution reception and judgment mechanisms and chromosome repair strategies, after normalizing fitness, high-quality individuals are selected by roulette wheel to form a replacement set. Each subpopulation randomly selects a corresponding number of individuals to form the replaced set. The replacement set is used to update the subpopulation, and finally a task configuration scheme that satisfies the optimization objective and constraints is generated and uploaded.

8. A UAV swarm payload resource planning system based on multi-attribute encoding and parallel variable neighborhood search, characterized in that: The system executes the UAV swarm payload resource planning method according to any one of claims 1 to 7; The system includes a target task module, a platform and resource module, a payload configuration module, and an approval module. The target task module is used to determine target characteristics and task requirements: the command center combines reconnaissance information and expert experience to define a set of target task attributes, including the resource requirements, expected benefits and task priority of each task. The platform and resource module is used to determine the attributes of unmanned platforms and available resources: clarify the task capabilities of different types of unmanned platforms and the relevant attributes of available resources, and archive the interaction relationships between different elements; The load configuration module is used to call the load configuration algorithm to generate a configuration scheme: the load configuration algorithm uses multi-attribute encoding to represent the configuration scheme, constructs an initial scheme based on task requirements, and then uses an adaptive variable neighborhood multi-population parallel evolution method to solve the problem, generating a task configuration scheme that meets the optimization objective and constraints and uploading it. The approval module is used for expert review and approval: based on the feasibility, efficiency and alignment with the objectives of the proposed solution, experts use their experience or decision support systems to decide whether to approve its use.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images