Unmanned aerial vehicle cluster toughness task allocation method considering local information

Abstract

The invention belongs to the field of computer science and system engineering, and particularly relates to an unmanned aerial vehicle group toughness task allocation method based on consideration of local information. A task problem allocation module is constructed according to different task demand types, and a variable auction group mechanism is introduced in an allocation decision process, so that the unmanned aerial vehicle dynamically forms a local auction group in a local communication range, and task allocation only using local information is realized. Compared with an algorithm based on global information, the method has the advantages that the competitive distribution performance is kept, and meanwhile, the communication requirement is effectively reduced. On the basis, a task redistribution algorithm is designed for the external interference or fault condition of the unmanned aerial vehicle, the multi-dimensional damage state of the unmanned aerial vehicle is quantified by using damage factors and probability distribution vectors thereof, structural representation of performance degradation under various fault conditions is provided, and the reliability of the unmanned aerial vehicle is improved while the time and precision requirements are met. And tasks can be efficiently completed.

Description

Technical Field

[0001] This invention belongs to the fields of computer science and systems engineering, and specifically relates to a resilient task allocation method for unmanned aerial vehicle (UAV) swarms that considers local information. Background Technology

[0002] Drone swarms are increasingly being used across various fields due to their flexibility and efficiency. Effective task allocation is key to maintaining swarm performance, especially in dynamic and uncertain environments.

[0003] Existing methods often rely on global information, resulting in high communication overhead and weak anti-interference capabilities. Drone swarms are typically large-scale and often operate within limited communication range. In long-distance mission scenarios, it is difficult for all aircraft to share information such as the status and task priorities of each agent in real time. Traditional global information allocation methods are susceptible to interference and generate significant communication overhead, leading to decision-making delays; in large-scale swarms, these methods may even be unfeasible due to hardware limitations.

[0004] Numerous researchers have proposed many effective research methods for task allocation, area coverage, and path planning in UAV swarms. These methods improve the efficiency of UAV swarm task execution, and efficient and accurate task allocation is key to task completion. Therefore, researchers have proposed corresponding solution methods for different task allocation models, mainly divided into three categories: optimization-based methods, game theory-based methods, and market-based methods.

[0005] Numerous optimization-based task allocation methods have been proposed, with the Hungarian algorithm being a prime example. Originally proposed by Kuhn and later improved by Munkres, this algorithm has been widely applied to multi-agent task allocation problems, including imbalanced allocation and large-scale matching scenarios implemented through various extensions and hybrid methods. Other optimization techniques, such as the wolf pack algorithm and its improved variants, have also been introduced to address issues like local optima and allocation efficiency. While these optimization-based methods effectively solve allocation and scheduling problems, they still face challenges in large-scale dynamic environments, including slow convergence, limited scalability, and high computational resource consumption.

[0006] Game theory methods address multi-agent decision-making problems by introducing concepts from classical game theory. They have been applied to various optimization and allocation scenarios. For example, Wu et al. proposed a latent game theory approach for dynamic task allocation in multi-agent systems, guaranteeing at least 50% suboptimality under any Nash equilibrium, while Zhou et al. proposed a game-theoretic strategy for ship autonomous navigation with improved equilibrium accuracy. Although game-theoretic methods offer advantages such as problem simplification, structured modeling, support for dynamics, and multi-stage decision-making, some models rely on global information, which is often difficult to obtain in practical applications.

[0007] Market-based algorithms exhibit good stability and scalability. Auction algorithms are a typical market-based approach. These algorithms are particularly effective in finding resource allocation solutions, especially suitable for scenarios with a large number of participants. Furthermore, auction algorithms are flexible, capable of handling various resource types and adapting to different bidding strategies. Dimitri proposed an auction algorithm to solve the task allocation problem for drones. This algorithm simulates a real auction process, allowing drones to bid for target tasks and ultimately obtain the optimal task allocation result. Choi proposed a decentralized auction algorithm based on traditional auction methods. This algorithm relies on information sharing among drones to achieve task allocation, enhancing fault tolerance and making drone swarms more suitable for dynamic and flexible scenarios. However, many auction algorithms rely on global information. This requires drones to know the bidding situation of others before bidding, and information between drones is transmitted through multiple hops, which may lead to information delays or loss; if other drones do not make appropriate bids within a limited time, it may also lead to locally optimal allocation results.

[0008] The above research has yielded many results regarding task allocation and resilience. However, it should be noted that most of the algorithms in these studies exhibit obvious characteristics and inherent limitations in both aspects: (1) Most existing UAV swarm task allocation methods rely on global information or involve large-scale information exchange, while there is little research on distributed allocation methods that only utilize local information. (2) Existing task allocation algorithms rarely consider the interference effects that may cause UAV failures during task execution, and research combining task allocation with resilience considerations is also very limited. Summary of the Invention

[0009] In view of this, the present invention proposes a resilient task allocation method for UAV swarms that considers local information, in order to solve the task allocation problem of UAV swarms in limited communication and uncertain environments.

[0010] The technical solution adopted in this invention is as follows:

[0011] A drone swarm resilience task allocation method considering local information includes the following steps:

[0012] Step 1, Task Modeling: Model the overall task of the cluster as a set of subtasks, and for each subtask... Define its set of functional requirements Based on the number of elements in the functional requirements set Subtasks are categorized into single-function or multi-function requirements; simultaneously, for each drone in the cluster... Establish the set of functions it can provide ;

[0013] Step 2, Initialization and Value Assessment: For each functional requirement of all subtasks Set initial price Each drone independently assesses its own contribution to the sub-task based on its local information. Functional requirements Task execution value ,in Step 3, Iterative Allocation: Task allocation is completed through multiple rounds of iteration. Each round of iteration includes:

[0014] 3.1 Distributed Bidding: Each drone bids based on the value of its mission execution. Current price relative to each functional requirement Calculate the net revenue; each drone submits a bid to the feature requirement with the highest net revenue, with a bid price of [missing information]. ,in To increase the bid volume, For drones The highest net income, The second highest net income, ε is a predetermined non-negative adjustment factor;

[0015] 3.2 Allocation Decision: The auction center's drones collect bidding information within their local communication range and allocate sub-tasks based on the bids. Functional requirements The drone will be assigned to the highest bidder, and the winning bid will be updated to the current price for that feature requirement. At any given time, there is only one auction center in the drone swarm: during the initial allocation, a drone is randomly selected as the auction center, and the drones with bidding capabilities covered by the local communication range of the auction center constitute the initial auction group; thereafter, the drone serving as the auction center completes the replacement of the auction center according to the preset dynamic transfer rules;

[0016] 3.3 Convergence Output: Repeat steps 3.1 to 3.2 until the task allocation method satisfies the ε-complementary relaxation condition. At this point, calculate the total value based on the function allocation results of each subtask. Evaluate the overall allocation effect and output the final allocation plan.

[0017] Furthermore, in step 3.2, when the drone serving as the auction center completes the change of the auction center according to the preset dynamic transfer rules, the transfer rules are as follows:

[0018] a. If there are drones within the local communication range of the current auction center that have not been assigned any tasks, the auction center remains unchanged;

[0019] b. If rule a is not met, then within the local communication range of the current auction center, select drones with auction coordination capabilities and choose a new auction center based on their external communication score. The external communication score refers to the number of drones that the drone can communicate directly with outside the local communication range of the current auction center via a single hop (signal strength not lower than a preset threshold). This indicator reflects the drone's ability to establish connections and disseminate information with drones outside the group.

[0020] The higher the external communication level, the stronger its ability to spread and connect external information; therefore, the drone with the highest external communication level is set as the new auction center; if there are multiple drones with the same external communication level, one of them is randomly selected; if there is no drone within the communication range that meets the conditions, the current auction center is temporarily retained, and the transfer operation is performed after a drone that meets the conditions enters the communication range.

[0021] Furthermore, the local communication range is a circular area centered on the drone serving as the auction center and extending outwards by a preset communication radius. When the drone at the auction center moves, or when a drone within the group moves and exceeds / enters the local communication range, the members of the auction group are dynamically updated.

[0022] Furthermore, the total value of the sub-tasks in step 3.3 Calculate using the following formula:

[0023] like And there are drones To be assigned to its unique function ,but ;

[0024] like And all its functional requirements ,but ;

[0025] If none of the above conditions are met, then ;

[0026] in, Indicates a subtask. This represents the corresponding set of functional requirements. This represents a subtask that meets a single-function requirement. This represents a subtask that meets multi-functional requirements; Indicates drone fulfill the task Medium functional requirements value, This indicates the amount of allocation indicators.

[0027] Furthermore, the method also includes step 4, a toughness redistribution process: when drone damage is detected, based on the damage factor... The task execution value is reduced, and the task it undertakes is released to a task release pool. If the released task is a multi-functional sub-task, other drones that are cooperating in executing the task are also released. Based on the reduced task execution value, a task reallocation algorithm is executed to reallocate suitable drones to the tasks in the task release pool.

[0028] Furthermore, the task reallocation algorithm includes the following steps in the process of reallocating drones to released tasks:

[0029] Determine the available set of drones A(t) within the local communication range of the auction center;

[0030] Randomly select a drone from set A(t). ;

[0031] Find all the functional requirements in the mission release pool for drones The highest value it brings;

[0032] If the highest value is higher than that of the drone The value of the current mission will be the drone Reassign to the functional requirements that bring the highest value.

[0033] Furthermore, the damage factor Its value follows a predetermined probability distribution, which consists of a damage state set D and a probability set. Definition; where, , ; Indicates the selection of damage factors The probability of and satisfies the normalization condition.

[0034] A drone swarm task allocation system is provided to implement the aforementioned drone swarm resilient task allocation method based on local information, characterized in that the system is configured as follows:

[0035] Implement mission modeling, value assessment, bidding decision-making, and communication functions in each drone;

[0036] Furthermore, the system is further configured to dynamically select a drone as the auction center, and to collect bidding information and make allocation decisions within the drone at the auction center.

[0037] By adopting the above technical solution, the present invention has the following beneficial effects:

[0038] 1. This invention designs a preliminary allocation algorithm. This algorithm constructs a task allocation module based on different task requirement types and introduces a variable auction group mechanism during its allocation decision-making process. This allows the UAV to dynamically form local auction groups within its local communication range, achieving task allocation using only local information. Compared with algorithms based on global information, this method effectively reduces communication requirements while maintaining competitive allocation performance.

[0039] 2. Based on the preliminary allocation algorithm, a task reassignment algorithm is designed to address external interference or faults in UAVs. This algorithm quantifies the multidimensional damage state of UAVs using damage factors and their probability distribution vectors, providing a structured representation of performance degradation under various fault conditions. This ensures efficient task completion while meeting time and accuracy requirements. Attached Figure Description

[0040] Figure 1 This is an example of the evolution of auction groups under the dynamic auction group mechanism.

[0041] Figure 2 This is a flowchart of the initial task allocation algorithm in an embodiment;

[0042] Figure 3 This is a flowchart of the task reallocation algorithm in an embodiment;

[0043] Figure 4 Performance comparison of the LI algorithm; where: LI is the initial allocation algorithm proposed in this invention; a is the average value map of the initial allocation algorithm proposed in this invention under different drone scales; b is the average value map of the CBAA algorithm under different drone scales; c is the average value map of the GA algorithm under different drone scales; d is the average value map of the SA algorithm under different drone scales.

[0044] Figure 5 Iteration and overall value of drone swarms ( ); where: a is a comparison of the average number of iterations between the initial allocation algorithm proposed in this invention and the fusion Hungarian algorithm under different local communication ranges when the number of UAVs is 15; b is a comparison of the average overall value of the two algorithms under the same scale and conditions.

[0045] Figure 6 Iteration and overall value of drone swarms (); where: a) is a comparison of the average number of iterations between the initial allocation algorithm proposed in this invention and the fusion Hungarian algorithm under different local communication ranges when the number of UAVs is 30; b) is a comparison of the average overall value of the two algorithms under the same scale and conditions.

[0046] Figure 7 Iteration and overall value of drone swarms ( ); where: a is a comparison of the average number of iterations between the initial allocation algorithm proposed in this invention and the fusion Hungarian algorithm under different local communication ranges when the number of UAVs is 60; b is a comparison of the average overall value of the two algorithms under the same scale and conditions.

[0047] Figure 8 Task redistribution and overall value changes under small-scale damage (2 UAVs); where: a is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 10; b is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 20; c is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 30; d is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 40.

[0048] Figure 9 Task redistribution and overall value changes under moderate damage (10 UAVs); where: a is a comparison of recovery for the three redistribution strategies when the local communication range of the UAV is 10; b is a comparison of recovery for the three redistribution strategies when the local communication range of the UAV is 20; c is a comparison of recovery for the three redistribution strategies when the local communication range of the UAV is 30; d is a comparison of recovery for the three redistribution strategies when the local communication range of the UAV is 40.

[0049] Figure 10 Task redistribution and overall value changes under large-scale damage (25 UAVs); where: a is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 10; b is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 20; c is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 30; d is a comparison of recovery of the three redistribution strategies when the local communication range of the UAV is 40.

[0050] Figure 11 This is an experimental analysis demonstrating the resilience of three redistribution strategies under different local communication ranges with a total system value requirement of 27.

[0051] Figure 12 This is an experimental analysis demonstrating the resilience of three redistribution strategies under different local communication ranges with a total system value requirement of 29; value

[0052] Figure 13 This is an experimental analysis demonstrating the resilience comparison of three redistribution strategies under different local communication ranges with a total system value requirement of 31; value

[0053] Figure 14 It is an experimental analysis demonstration Comparison of average toughness under different scenarios. Detailed Implementation

[0054] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0055] Step 1, Task Modeling: Model the overall task of the cluster as a set of subtasks, and define the functional requirements set for each subtask; based on the number of elements in the functional requirements set, classify the subtasks into single-functional requirement subtasks or multi-functional requirement subtasks; simultaneously, establish the functional set that each UAV in the cluster can provide. The specific implementation method for this step is as follows:

[0056] A drone swarm is a collection of drones Composition, comprising A drone, of which Indicates the first A drone, Each drone Equipped with Different functions, the set of which is denoted as ,in express The This function.

[0057] The overall tasks that the drone swarm needs to complete are divided into: A set of non-intersecting subtasks is represented as follows: ,in Indicates the first Sub-tasks ( Each subtask Corresponding to a set of functional requirements , This indicates the number of functionalities required for this subtask. According to... Based on the numerical values, subtasks are divided into the following two categories:

[0058] Category 1: When hour, For a single-function requirement subtask, denoted as Such subtasks only require one function and can be completed independently by any drone equipped with that function.

[0059] Category Two: When hour, For multi-functional subtasks, record Such subtasks require Different functions need to work together. Each drone provides one of the required functions and works together to complete the task.

[0060] Each drone can provide at most one function for a specific subtask. If the subtask... If any functional requirement cannot be met, the task cannot be completed. Set to the total number of all functional requirements for all tasks.

[0061] For any drone Kazuko's Mission ,definition This set represents drones Possess and satisfy subtasks All functional requirements. If , then drone Able to participate in subtasks ;like , then drone Unable to participate in this subtask.

[0062] Step 2, Initialization and Value Assessment: By introducing a variable auction group mechanism, UAVs dynamically form local auction groups within their local communication range, achieving task allocation using only local information while maintaining competitive allocation performance and reducing communication requirements. Before initial allocation, an initial price needs to be set for each functional requirement of all subtasks; based on the performance of each UAV, the task execution value of each UAV for different functional requirements within each subtask is assessed. The specific implementation method for this step is as follows:

[0063] Building a value matrix ,in For the number of drones, A collection of functional requirement types. This represents the total number of functional requirements for all subtasks.

[0064] make Indicates drone sub-tasks The value correlation between them. Due to drones Need to be from the set Subtask Provide specific functions, and further Refine into function-specific value drones For subtasks Provide functions The specific value of the function at that time.

[0065] Given the heterogeneity of drone performance, even for the same function and the same subtask Different drones The values ​​may also differ. At time t, the UAV task assignment matrix is ​​denoted as... .

[0066] Step 3, Distributed Bidding: Based on the variable auction group mechanism described in Step 2, drones within the auction group (excluding the auction center) act as bidders. They obtain the current price of each sub-task's functional requirement based on the initial price, and calculate the net profit based on the task execution value and the current price. They then bid for the sub-task's functional requirement with the highest net profit. The information conveyed by the bidding behavior can be used to identify the bidding drone, the bidding functional requirement, and the bid price. The bid price is the current price of that functional requirement plus a bid increment. The specific implementation method for this step is as follows:

[0067] Employing a market-based decision-making mechanism, each drone The target task with the highest net profit will be given priority in bidding:

[0068] ;

[0069] in, This refers to the functional requirements section f of task j, which is for the drone. The selected functional requirement that yields the greatest benefit. If ,but , The current price for the f-functionality requirement portion of task j.

[0070] Before bidding, each drone operator will evaluate the target mission. Whether to maintain the highest bid, otherwise abandon the current task and continue the search. The bidding rules are as follows:

[0071] Current price of this feature section Plus bidding increment ;in, ; Represents the highest net revenue in the drone bidding mission. , The value represents the second highest net revenue for drone i among all eligible sub-tasks. ε is a small adjustment factor to ensure bid differentiation; ε>0 avoids completely identical bids from different drones.

[0072] Step 4, Allocation Decision: A drone is dynamically selected for the auction center. This drone collects bidding information from other drones within its local communication range. Based on the collected information, it identifies the drone, sub-task functional requirements, and bid for each bid. The allocation of each functional requirement within the sub-task is updated according to the bid ranking, and the winning bid price is updated to the current price of that functional requirement, thus forming a task allocation scheme. The specific implementation method for this step is as follows:

[0073] In this method, any drone can serve as an auction center, with its local communication range defined as a local information exchange area. Drones only participate in auction activities within this area, and each drone within the area can directly receive bidding information from other drones. The dynamism of the local auction group is reflected as follows: Figure 1 As shown:

[0074] Assuming a drone The local communication range is The dashed line represents the local communication range of the drone, while the solid line represents the information exchange area, determining the members of the local auction group. In this example, at time... If UAV-1 serves as the center of the local information exchange area, then UAV-2, UAV-3, and UAV-4 within its local communication range constitute the auction group. At time... If drone-1 remains the center of the local information exchange area, due to the movement of drones, drone-2, drone-3, and drone-4 leave the information exchange area, and drone-5, drone-6, and drone-8 enter the local auction group. Alternatively, the center of the local information exchange area shifts from drone-1 to drone-3, at which point drone-1, drone-2, drone-3, and drone-6 form a new auction group.

[0075] Define binary assignment variables , Used to represent drones At any moment Allocation status: when Assigned to execute subtasks When the function f is used, Its corresponding value Included in the total task value; otherwise ,and Value is not included in the statistics.

[0076] Each drone can be assigned to a maximum of one functional requirement at any given time, meaning there can be at most one pair. Make For example, if Assigned as a multi-functional task Provide functions If it is not assigned to any other function of the same task, it cannot be assigned to any other task at the same time.

[0077] Update the attribution of each functional requirement component of the subtask based on the bid ranking, and determine the total value of subtask j. Defined as:

[0078] For single-function requirements If drones exist The only required function assigned to this task (ie, ),but ;otherwise .

[0079] For multifunctional needs ,like All functional requirements are allocated, that is, for each Drones are present in all of them. satisfy ,but Take the sum of the highest value provided by each function. If any function requirement is not assigned, then... The unified expression is:

[0080] ;

[0081] The winning bid is updated to the current price of the required functionality, forming a task allocation plan. Leveraging the dynamic nature of the variable auction group mechanism, drones within the local communication range of the auction center are allocated tasks based on their net revenue. Bidding, with prices following the " "As the rules escalate, non-optimal bidders are filtered out until a one-to-one match is achieved."

[0082] By introducing a variable auction group mechanism, the UAV dynamically forms local auction groups within its local communication range to achieve task allocation using only local information. This method effectively reduces communication requirements while maintaining competitive allocation performance.

[0083] Step 5, Iterative Convergence and Output: Repeat steps 3 and 4. Through multiple rounds of iterative distributed bidding and allocation decisions, the task allocation scheme is made to ultimately satisfy the ε-complementary relaxation condition. This final allocation scheme is then output, achieving near-global optimality of the overall task execution efficiency of the cluster. The specific implementation method for this step is as follows:

[0084] Under the variable auction group mechanism, steps 3 and 4 are repeated. The auction group is dynamically updated by the drone moving into / out of the local communication range or by the auction center shifting. The auction center shifting rules are as follows:

[0085] 1) At any given time, there is only one auction center in the drone swarm;

[0086] 2) If there are drones within the local communication range of the current auction center that have not been assigned any tasks, the auction center remains unchanged;

[0087] 3) If rule a is not met, then within the local communication range of the current auction center, select drones with auction coordination capabilities and choose a new auction center based on their external communication score. The external communication score refers to the number of drones that the drone can communicate directly with outside the local communication range of the current auction center via a single hop (signal strength not lower than a preset threshold). This indicator reflects the drone's ability to establish connections and disseminate information with drones outside the group.

[0088] The higher the external communication level, the stronger its ability to spread and connect external information; therefore, the drone with the highest external communication level is set as the new auction center; if there are multiple drones with the same external communication level, one of them is randomly selected; if there is no drone within the communication range that meets the conditions, the current auction center is temporarily retained, and the transfer operation is performed after a drone that meets the conditions enters the communication range.

[0089] The communication topology of dynamic auction groups possesses three characteristics: dynamism, local connectivity, and global coverage. Dynamicity refers to the real-time updates of the communication topology as the drone moves; local connectivity means that drones within the auction center's sensing radius Rᵢ form a star topology with the center and exchange task information via direct single-hop links; and global coverage means that after multiple rounds of updates, each drone in the cluster joins at least one auction group, achieving global coverage through local information propagation.

[0090] Overall value of the allocation plan Defined as the sum of the values ​​of all subtasks; its essence is to allocate results through cumulative summation. With corresponding functional specific value This is achieved through the product of , and its expression is:

[0091] ;

[0092] The core objective of task allocation is to realize overall value. Maximize. The objective function and constraints can be expressed as:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] Among them, constraint 1 requires the drone Must have participation in subtasks Capability; Constraint 2 requires that at any given time, a single drone Only one functional component can participate in a single mission; constraint 3 stipulates that for multi-functional missions, all functional requirements must be performed by different UAVs for the mission's value to be realized.

[0098] Based on the market decision-making mechanism in step 3, each drone They will try to bid for the target task that yields the greatest net benefit to them:

[0099] ;

[0100] In the formula, Indicates task of - Functional requirements section, drones Choose the option that yields the greatest benefit, if ,but , For the task middle - Current price for the feature requirement section.

[0101] Considering the case where the highest and second-highest net profits are equal, an ε-complementary relaxation condition (ε≥0) is introduced. The algorithm terminates when this condition is met and matching is feasible. The smaller ε is, the closer the allocation result is to the global optimum, and the result is independent of the allocation order. Finally, this allocation scheme is output as the final result, achieving an approximate global optimum for the overall task execution efficiency of the cluster.

[0102] This invention achieves initial task allocation for UAV swarms through five steps: task modeling, initialization and value matrix input, distributed bidding, allocation decision-making, and iterative convergence. By introducing a variable auction group mechanism, UAVs dynamically form local auction groups within their local communication range, achieving task allocation using only local information. Compared to algorithms based on global information, this method effectively reduces communication requirements while maintaining competitive allocation performance. To better illustrate its feasibility and effectiveness, it is named the Initial Allocation Algorithm. The pseudocode structure of the Initial Allocation Algorithm, with UAV x as the auction center, is as follows:

[0103]

[0104] When a drone swarm performs a mission, the priorities of each mission are clearly different. Higher-priority missions are usually assigned to drones with matching capabilities to increase the probability of mission success. However, if a drone performing a mission malfunctions, its high-priority mission may remain incomplete. From a global perspective of overall mission planning, such scenarios will lead to mission failure. Therefore, it is necessary to dynamically adjust the mission allocation strategy to ensure that high-priority missions in critical areas can be effectively reallocated and successfully completed. Based on this, this embodiment adds a step 6, the mission reallocation process: designing a mission reallocation algorithm; when a drone is damaged, the released mission is reassigned to other drones through the mission reallocation algorithm; and based on the reallocation result, the resilience metric of the drone swarm within the mission time interval is calculated to quantify the system's ability to maintain performance under damage interference. The specific implementation method of this step is as follows:

[0105] In task allocation scenarios, external interference may cause partial loss of UAV functionality, necessitating the modeling of all potential damage states of the UAV. :

[0106] ;

[0107] in, This represents the damage factor corresponding to each damaged drone. The smaller the value, the more severe the damage to the corresponding drone.

[0108] Considering the random nature of damage occurrence, the damage state of a UAV must follow a corresponding probability distribution law. The probability distribution of UAV damage state is defined as follows:

[0109] ;

[0110] in, Indicates the selection of damage coefficient The probability of and satisfying the normalization condition. For example, if This means selecting the damage factor. The probability is 70%, choose... The probability of failure is 20%, and the probability of complete failure is... The probability is 10%.

[0111] Based on the previous task allocation model, the value matrix of a drone swarm is defined as follows: For each task Each functional requirement drones Each corresponds to a value This value is a matrix. One of the elements. Definition For drones Value vector for all missions. Drones when facing damage. Damage factors Based on the probability distribution vector of the damage factor Confirmed. At this moment, the drone... The value to each task will be related to Multiply to quantify the performance degradation caused by the loss, i.e.:

[0112] ;

[0113] The new value matrix after damage is obtained in the following way:

[0114] ;

[0115] Based on the damage status of the drones, the functional requirements of each task are divided into two states: assigned and unassigned. Assigned means the drone has successfully received the task, while unassigned includes drones that have failed and need to be re-assigned. The initial state of the drone swarm and the newly assigned tasks are represented as follows:

[0116] ;

[0117] in, Indicates drone The running status, If and only if the task All functional requirements have been met; specifically defined as:

[0118] When the task by drone When successfully received by the drone ,otherwise When drones Damage factors hour, ;otherwise, .

[0119] Assuming a drone Assigned to task If drone A failure occurs, regardless of its damage factor. To determine the appropriate value, the task needs to be released for reallocation and updated in order to reassign it to a more suitable drone. Then, the drone and tasks The status update is as follows:

[0120] ;

[0121] Initial task assignment matrix Represented as:

[0122] ;

[0123] In matrix X, the drone For the task The allocation result is expressed as follows:

[0124] ;

[0125] Resilience metric: Task time The toughness metric is calculated as follows:

[0126] ;

[0127] in, This represents the overall resilience of the system at time t. ∈[0,1] represents the weighting coefficient. Represents resilience over time. Performance resilience can be represented by the following formulas, which can be used to calculate these performance indicators:

[0128] ;

[0129] in, This represents the overall value of the cluster at time t. This represents the minimum total value requirement of the system at time t. It is the optimal expected total value of the system at time t; The brackets are Iverson brackets, and the result is 1 when the equation inside the brackets is true, otherwise it is 0; the parameter α∈(0,1) represents the conversion coefficient between the system's time toughness and performance toughness.

[0130] Finally, considering drone malfunctions, a task reassignment algorithm is proposed to ensure efficient task completion while meeting time and accuracy requirements. The code structure of the task reassignment algorithm is as follows:

[0131]

[0132] Experimental verification:

[0133] This section aims to analyze and verify the effectiveness of the UAV swarm resilience task allocation method proposed in this invention. The related experiments were implemented in MATLAB 2017a and run on a personal computer configured with a Windows 10 operating system, an Intel(R) Core(TM) i7-8550U CPU (1.99 GHz), and 8GB of memory.

[0134] 1. Analysis of the preliminary allocation algorithm:

[0135] Meanwhile, the Hungarian algorithm integrated with the local information allocation framework was used as a comparative example to conduct a comparative experiment with the UAV swarm resilience task allocation method of this invention. The purpose of the comparison is twofold: (1) to verify whether the total benefit obtained by the proposed method is close to the global optimum obtained by CBAA under global information under the same value matrix and model; (2) to examine the stability of the algorithm when the swarm size increases and the local communication range changes by comparing its performance with that of the Hungarian algorithm integrated with the local information allocation framework.

[0136] Assume that each task has a different value for different drones. The task value matrix is ​​represented as follows: The values ​​in the matrix are randomly generated. In this experiment, the number of drones in the swarm is set to... , , , and The number of tasks is set to , , , and drones Local communication range This area is designated as a local information interaction zone. Since drones move continuously, this is abstracted as a process of continuously and randomly selecting drones to participate in information interaction. Assume the number of drones entering the local communication range is... The expansion is directly proportional to the increase. As shown in Table 2, the number of drones participating in the auction mechanism (denoted as ) The distribution follows a uniform distribution. To ensure statistical reliability, 50 independent experiments were conducted, and the results were statistically averaged. The overall results of this experiment are as follows: Figure 4 As shown in the middle (ad). From Figure 4 As can be seen from the algorithm, the task allocation result based on local information is consistent with the allocation result based on global information. This is because the auction group is continuously updated, and the overall task allocation of the cluster is gradually completed through multiple rounds of auctions initiated within the group. Compared with the greedy algorithm and the simulated annealing algorithm, the allocation method of this invention has better performance and allocation effect; compared with the CBAA algorithm, its core advantage is that it can rely entirely on local information to achieve task allocation.

[0137] To ensure a fair comparison with the initial allocation algorithm described in this paper, a local information allocation framework is combined with the classic Hungarian method (hereinafter referred to as the fused Hungarian algorithm). The fused algorithm follows an allocation mechanism based on local information. It first identifies drones within the current information exchange domain and categorizes them into sets. In this system, task allocation within each auction group is coordinated by the auction center. Under this mechanism, the value matrix within the group serves as input to the algorithm, calculating the best match for the current auction group in a single execution. Allocation results are updated periodically by the auction center. After allocation for the current group is complete, the process continues to the next auction group, while previous results are propagated through communication between auction centers.

[0138] Table 2 and Relationship

[0139] Based on the modified Hungarian algorithm described above, a comparative experiment was conducted to compare the performance of the proposed initial allocation algorithm and the fused Hungarian algorithm. In this experiment, the number of UAVs in the swarm was set to... This corresponds to small, medium, and large-scale drone swarms. The number of missions is... Local communication range Gradually increase. The drones entering the interaction area are evenly distributed; specific parameters are shown in Table 2. Each configuration is run independently 50 times, and the average value of the experimental results is taken. Figure 5 a and b in Figure 6 a and b in the text, and Figure 7 As shown in a and b in the figure.

[0140] The black curve represents the proposed method; the red curve represents the baseline of the fused Hungarian algorithm. In a fixed... In this case, with The increase in shows a monotonically decreasing trend, indicating that the algorithm converges faster when more drones participate in the auction. Under the same conditions... As the population size increases from 15 to 30 and then to 60, the number of iterations increases significantly. In terms of benefits, for Small / medium-sized clusters, all The corresponding curves almost overlap, indicating that the algorithm is effective. It is not sensitive to changes. For large-scale... The integration of Hungarian methods in large The regional distribution exhibited fluctuations, initially declining and then rebounding, while the proposed method remained relatively stable, demonstrating superior stability. This is because the fused Hungarian algorithm performs a locally optimal allocation only once per round for drones within the current auction group and directly writes the result into the allocation scheme. Since this algorithm cannot adjust the allocation result in real-time within the current round, the error is relatively large, and this allocation result may affect the allocation paths of subsequent tasks. As the total size of the drone swarm increases and the number of available allocation paths increases, path dependence is amplified, leading to increased fluctuations in the results. In contrast, Algorithm 1 uses a drone-by-drone allocation method, which can adjust the allocation scheme in real-time during each round, thus achieving higher allocation accuracy. Overall, with increasing scale, the proposed method consistently demonstrates faster convergence speed and higher stability.

[0141] 2. Analysis of task redistribution methods:

[0142] When drone swarms encounter interference, the performance of the proposed task reassignment method needs to be verified. For a more comprehensive comparison, in addition to the strategy... In addition to (Algorithm 2), an extra design was made. and and Conduct comparative experiments.

[0143] Strategy 2 Auction redistribution under dynamic ε-complementary relaxation:

[0144] The initial allocation adopts the same as The same auction algorithm, but with adaptively adjusted ε-complementary relaxation conditions. In the initial stage, a relatively large ε is set (e.g., ...). (where n is the total number of drones) to increase the bid increment. This accelerates task allocation. When the system enters the phase of diminishing marginal returns, ε is reduced, shifting from rapid allocation to more precise allocation (e.g., ...). For higher precision, This improves the accuracy of the solution. In each round of the auction, k undamaged drones are randomly selected to join the auction group; bids are generated based on the difference between the highest and second-highest payouts to update the allocation and price. This scheme is only adopted when the total value is not less than the value after damage and the total value of the previous allocation. The auction primarily uses localized bidding; once the current group is allocated, the auction results are transmitted through the auction center. compared to, While there is a sacrifice in speed, convergence accuracy and stability are significantly improved. It's important to note that ε directly adjusts the assignment speed and solution accuracy: a larger ε increases the assignment speed, while a smaller ε results in higher accuracy.

[0145] Strategy 3 ( ): The Hungarian algorithm integrated into the local information allocation framework. Following the same fusion process, but based on the drone's loss state, the matrix is... Replace with The corresponding task is then added to the set to be assigned. Subsequently, the Hungarian algorithm is invoked to obtain the optimal match after the damage. During the reallocation process, if the overall value of the current scheme is better than or equal to the overall value after the damage, the original allocation scheme is overwritten and updated; the iteration continues until all drones achieve a unique match, and the overall value remains stable across multiple iterations.

[0146] Since the damage to the drones is randomly selected, and each damage state is determined based on the probability distribution of the damage, it is difficult to ensure that the three strategies can be compared under the same damage conditions. To more fairly and accurately verify the performance of different strategies and ensure the repeatability of the experiment, this experiment fixes the random factors of drone damage and the damage states of each drone, and compares the three strategies under the same damage scenario, thereby clearly demonstrating the performance differences of each scheme.

[0147] Experimental settings: Size of the drone swarm Number of tasks Damage threshold Corresponding to small-scale, medium-scale, and large-scale damage respectively; local communication range Set to respectively and Assuming the number of drones entering the local communication range is... The expansion is directly proportional to the value; specific parameters are shown in Table 2.

[0148] After the initial task allocation is completed, the cluster starts from... The damage begins progressively: at each time interval, one drone is selected to become inoperable based on the damage distribution of failure events. After a period of time, the auction center calculates the total damage and detailed information; when the accumulated damage reaches a predetermined threshold... At that time, the drone stopped being damaged, and the cluster triggered a task redistribution.

[0149] like Figure 8 a to d in Figure 9 a to d in the middle, and Figure 10 As shown in figures a to d, different damage levels (2, 10, and 25 drones) and local communication ranges were observed. The overall value changes of the three recovery strategies follow a largely consistent pattern:

[0150] Strategy (Green line) has the fastest recovery speed, but lower overall value. Strategy (Black line) is superior in terms of overall value of recovery. Although the recovery speed is slow. Strategy (Red line) has the slowest recovery speed, although its final total value is similar to... Similar, but with greater fluctuations during the recovery process. This varies depending on the local communication range. The increase, The recovery speed did not change much, while The recovery speed has improved. In comparison, Recovery is actually slower over a larger local communication range.

[0151] In summary, It boasts the fastest recovery speed and is suitable for scenarios where recovery time is tight but the requirements for recovery effect and stability are not high. It is suitable for scenarios with ample time and high requirements for recovery effect and stability. The increase in local communication range can effectively compensate for its slow recovery speed. The recovery effect is close to However, it has the slowest recovery speed and the effect fluctuates greatly.

[0152] To investigate the impact of task redistribution strategies on group resilience, the strategies were tested under different total value requirements. , and Each experiment was repeated 50 times. To more accurately characterize the features of each strategy, the number of damaged drones was kept consistent, and the drones that failed were set to be random, with their damage state determined by the damage probability distribution.

[0153] Assuming the size of the drone swarm is The number of tasks is Damage threshold The total task duration is 200, and the system's minimum total value requirement is... The values ​​are set to 27, 29, and 31, representing low, medium, and high value requirements, respectively. Optimal expected total value. The value is 33.92, representing the optimal task allocation scheme under normal operating conditions for all UAVs. By recording the overall value of the UAV swarm at each time step, the resilience changes of the three strategies within different local communication ranges were analyzed. The relationship between the local UAV set A and the local UAV set A is shown in Table 2.

[0154] like Figure 11-13 As shown, the performance of the three strategies is compared under low, medium, and high overall value demand scenarios: Strategy It exhibits good performance in low-to-medium demand scenarios, but its average resilience is inferior to the strategy under low demand and large local communication range. As shown in Equation 14, this phenomenon is mainly because resilience calculations simultaneously encompass both time resilience and performance resilience. When the local communication range is large, Compared to Its time resilience advantage is gradually weakening. Under low total value demand, It better meets system requirements, and the recovery effect is slightly better. .like Figure 13 As shown, under high demand, The performance of the device is significantly reduced, the recovery effect is poor, and the stability is insufficient.

[0155] Strategy It performs stably across various demand scenarios, and especially under high-demand conditions, it can provide better performance than... More stable recovery results. As the local communication range increases, The convergence speed is significantly accelerated. However, in high-demand scenarios with severe damage, i.e., where damaged nodes are of high value and most nodes are severely damaged or completely destroyed, it may affect... The recovery effect was insufficient, causing the system to be unable to meet the current high demands. Such anomalies are relatively rare; generally speaking... It performs exceptionally well in high-demand scenarios.

[0156] Strategy Recovery stability was poor across all demand scenarios, although its recovery performance was better than average under high-demand conditions. However, the recovery process fluctuates significantly, lacks overall stability, and the final recovery benefit is usually lower than expected. Therefore, it is not suitable for high-demand scenarios.

[0157] In summary, the strategy Suitable for low to medium demand scenarios with small local communication range, it offers relatively fast recovery speed, but the recovery effect is generally average; Strategy Suitable for low- or high-demand scenarios with a relatively large local communication range, especially providing more stable and higher recovery performance under high-demand conditions; Strategy It is suitable for scenarios with low requirements for recovery speed, but due to insufficient stability, it should be used with caution.

[0158] 3. Analysis of reallocation strategies under different redundancy levels and multi-functional task requirements

[0159] To better reflect the actual combat scenarios of UAV swarms, the experiment further incorporated factors such as UAV redundancy configuration and multi-functional task requirements. In actual mission execution, redundant UAVs are typically deployed to enhance swarm robustness and ensure continuous mission execution even when UAVs are damaged. Furthermore, although the overall mission of a UAV swarm can be divided into multiple non-overlapping sub-tasks, some tasks inherently require multiple UAVs to collaborate. The successful execution of such multi-functional tasks depends on the synchronous cooperation of multiple UAVs. If any UAV participating in the mission is damaged, all collaborating UAVs must release their current mission and re-enter the allocation process to obtain the optimal solution. The experiment set up two groups of damaged UAVs, six in each group, corresponding to damage scenarios A and B: Scenario A involves UAVs participating in single-functional tasks being damaged, and Scenario B involves UAVs participating in multi-functional tasks being damaged. The impact of redundancy configuration and damage type on system resilience was evaluated through repeated experiments. The experimental parameters were set as follows: Number of tasks Three of them are multi-functional missions: one mission requires three functions, and the other two missions each require two functions. The baseline drone swarm size... Damage threshold Local communication range The task duration is 50, and the minimum total value requirement is... Optimal expected total value Local communication range With local drone ensemble The correspondence is shown in Table 2.

[0160] To evaluate the redundancy effect, the number of redundant drones was set to vary from 1 to 10. Considering the randomness of the allocation process, each set of parameters was run randomly 50 times to obtain the average resilience value. (Based on strategy) Using a representative scheme, the recovery capabilities of different redundancy levels under two damage scenarios were analyzed, and the results are as follows: Figure 14 As shown ( Axis represents redundant configuration, such as Total number of drones The number of tasks remains unchanged; The axis represents the average toughness value from 50 experiments.

[0161] Experimental results show that the average resilience of the UAV swarm increases with the increase of the number of redundant UAVs, and redundancy configuration has a positive impact on system resilience: as the number of redundant UAVs increases, system resilience improves significantly until the number of redundant UAVs exceeds the number of damaged UAVs. The resilience improvement rates differ between the two damage scenarios: in scenario A, resilience increases rapidly when redundancy initially increases, but gradually saturates as the number of redundant UAVs continues to grow, indicating that the marginal effect of additional redundancy gradually decreases; in scenario B, however, resilience maintains a stable upward trend throughout the redundancy range, indicating that in multi-functional task scenarios, redundancy configuration can still play an effective role even at higher levels. This is because damage to UAVs in multi-functional tasks affects multiple cooperating nodes, leading to more severe degradation of system performance, and redundancy configuration can effectively compensate for such coupling losses.

[0162] In summary, redundancy plays a crucial role in enhancing the recovery and resilience of drone swarms. Especially in scenarios with complex task coupling and high damage levels, reasonable redundancy deployment can significantly improve system robustness and ensure mission continuity under adverse conditions.

[0163] It is understood that this invention is described through comparative experiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of this invention. Furthermore, under the teachings of this invention, these features and experiments can be modified to suit specific circumstances and materials without departing from the spirit and scope of this invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are protected by this invention.

Claims

1. A method for resilient task assignment of a UAV swarm considering local information, characterized in that, The method comprises the following steps: Step 1, task modeling: model the overall task of the swarm as a set of subtasks, and for each subtask define its set of functional requirements ; according to the number of elements in the set of functional requirements classify the subtask as a single-functional requirement subtask or a multi-functional requirement subtask; at the same time, for each unmanned aerial vehicle in the swarm establish its set of functions that it can provide ; Step 2, initialization and value assessment: for each functional requirement of all sub-tasks Set initial price ; each UAV independently assesses its own value to the sub-task based on its local information functional requirements of the sub-task task execution value of the sub-task wherein ; Step 3, iterative allocation: completing task allocation through multiple rounds of iteration, each round of iteration comprising: Step 3.1, Distributed Bidding: Each drone bids based on its mission execution value. Current price for each feature requirement Calculate the net revenue; each drone submits a bid to the feature requirement with the highest net revenue, with the bid price being... ,in To increase the bid volume, For drones The highest net income, The second highest net income, ε is a predetermined non-negative adjustment factor; Step 3.2, Allocation Decision: The auction center's drone collects bidding information within its local communication range and allocates sub-tasks according to the bids. Functional requirements The drone will be assigned to the highest bidder, and the winning bid will be updated to the current price for that feature requirement. At any given time, there is only one auction center in the drone swarm: during the initial allocation, a drone is randomly selected as the auction center, and the drones with bidding capabilities covered by the local communication range of the auction center constitute the initial auction group; thereafter, the drone serving as the auction center completes the replacement of the auction center according to the preset dynamic transfer rules; Step 3.3, convergent output: repeat steps 3.1 to 3.2 until the task allocation formula satisfies the case-epsilon complementary relaxation condition, at which point the total value of each sub-task is calculated based on the functional allocation results of each sub-task , evaluate the overall allocation effect, and output the final allocation scheme. And output the final allocation scheme.

2. The method of claim 1, wherein, In step 3.2, the UAV serving as the auction center completes replacement of the auction center according to a preset dynamic transfer rule, and the transfer rule is as follows: a. If there is a UAV in the local communication range of the current auction center that has not been allocated any task, the auction center remains unchanged; b. If rule a is not met, filter the UAVs with auction coordination capabilities in the local communication range of the current auction center, and select a new auction center according to the external communication degree thereof; wherein the external communication degree refers to the number of UAVs outside the boundary that can communicate directly through one hop, with the local communication range of the current auction center as the boundary at the time of calculation; then the UAV with the largest external communication degree is set as the new auction center; if there are multiple UAVs with the same external communication degree, one of them is randomly selected; if there is no UAV meeting the condition in the communication range, the current auction center is temporarily retained, and the transfer operation is performed again when a UAV meeting the condition enters the communication range.

3. The method of claim 1, wherein, The local communication range is a circular area with the UAV serving as the auction center as the center and a preset communication radius as the distance; when the UAV serving as the auction center is transferred or the movement of the UAVs in the group causes the local communication range to be exceeded / entered, the members of the auction group are dynamically updated.

4. The method of claim 1, wherein, the total value of the subtasks in step 3.3 is calculated by the following formula: If , and there is a drone assigned to its unique function needs , then ; If , and all of its functional requirements , then ; If any of the above conditions are not met, then ; wherein, represents a subtask, represents a corresponding set of functional requirements, represents a single functional requirement subtask, represents a multiple functional requirement subtask; represents a drone satisfies a task a functional requirement a value, represents an assigned index quantity.

5. The method of claim 1, wherein, The method further comprises step 4, a resilience re-allocation process: When a UAV damage is monitored, according to the damage factor Decays its task execution value, and releases its undertaken task to a task release pool; If the released task is a multi-functional requirement sub-task, other UAVs that are currently executing the task are also released; Based on the decayed task execution value, a task re-allocation algorithm is executed to re-allocate appropriate UAVs to the tasks in the task release pool.

6. The method of claim 5, wherein, In the task re-allocation algorithm, the process of re-allocating UAVs to the released tasks comprises the following steps: Determine the available UAV set A(t) in the local communication range of the auction center; randomly selecting a drone from the set A(t) ; Finding all functional requirements in the mission release pool can bring the highest value to the UAV design; if the highest value is higher than the value of the current task, then the drone is reassigned to a functional requirement that brings the highest value. is reassigned to a functional requirement that brings the highest value.

7. The method of claim 5, wherein, said impairment factor , which values follow a predetermined probability distribution, said probability distribution being defined by a set of impairment states D and a set of probabilities ; wherein, , ; denotes the probability of selecting an impairment factor , and satisfies the normalization condition.

Images