Large-scale heterogeneous unmanned aerial vehicle multi-level dynamic task decision-making method based on coalition game

Abstract

The invention discloses a multi-level dynamic task decision-making method for a large-scale heterogeneous unmanned aerial vehicle based on a coalition game, and relates to the field of game theory, task decision-making and multi-agent collaborative task planning, and the method comprises the following steps: 1, building a dynamic task decision-making optimization model of the heterogeneous unmanned aerial vehicle under multiple constraints, determining constraint conditions and designing an evaluation function based on the unmanned aerial vehicle income; 2, converting the dynamic task decision process into a coalition game solving process, establishing a coalition game model, determining a preference relationship of the unmanned aerial vehicle based on coalition revenue, designing a coalition switching rule of the unmanned aerial vehicle, giving a form of a coalition game stable solution, and determining a convergence condition of the game; and step 3, designing a multi-level dynamic task decision algorithm based on the established coalition game model. According to the method, a complex dynamic task decision process is decomposed into a plurality of stages, global revenue maximization is realized, and the method has good expandability for heterogeneous unmanned aerial vehicle clusters.

Description

Technical Field

[0001] The present invention relates to the technical fields of game theory, task decision and multi-agent collaborative task planning, and in particular to a multi-level dynamic task decision method for large-scale heterogeneous unmanned aerial vehicles based on alliance game. Background Art

[0002] With the advancement of aerospace technology, drones are increasingly used in agriculture, transportation and military fields. The development of multi-agent systems has further enhanced the ability of large drone swarms to work collaboratively in dangerous and uncertain environments. In a multi-drone system, mission decision-making is crucial to the successful execution of the mission. Its essence is a combinatorial optimization problem with multiple constraints, including the capabilities, energy, safety and flight costs of heterogeneous drones. Under these constraints, how to design fast and dynamic strategies to achieve autonomous mission decision-making for heterogeneous multi-drones is a major challenge.

[0003] Task decision methods in multi-UAV systems are mainly divided into two frameworks: centralized and distributed. Centralized task decision methods use a single computing unit to analyze the battlefield environment and enemy information to make unified task decisions for multiple UAVs. According to different solution methods, they can be further divided into optimization-based and heuristic-based centralized methods. Optimization methods usually include integer programming, combinatorial optimization, and graph theory, which can clearly and effectively solve simple task decision problems. However, these methods often perform poorly when dealing with complex dynamic collaboration between UAVs, making them less efficient in large-scale tasks. Heuristic methods, such as genetic algorithms and particle swarm optimization algorithms, can well balance real-time and optimality and provide timely and practical decision solutions. Although heuristic methods perform well in computational efficiency, they are sometimes limited in versatility and scalability due to scenario-specific assumptions and constraints, and may converge to local optimal solutions rather than global optimal solutions.

[0004] Unlike centralized task decision-making methods, in a distributed task decision-making framework, each drone acts as a decision-making subject and shares environmental data, task information, and operating status with other drones. In a distributed framework, task decisions are based on iterative exchange of information, which enhances the flexibility and adaptability of the algorithm. Common distributed task decision-making methods include market-based and game-theory-based methods. Market-based methods simulate economic transactions by conceptualizing tasks as auctionable commodities. Drones participate in auctions and negotiations to optimize overall system performance in scenarios where multiple tasks are auctioned. Unlike market mechanisms, game theory is an analytical method used to study the optimal decision-making problem of multiple participants in a cooperative or competitive environment. Game-theory-based methods have advantages in solution efficiency, scalability, and communication costs, and are suitable for large-scale task decision-making, but have high requirements on the convergence of the algorithm. At the same time, the redundancy of drones will affect the results of task decision-making.

[0005] At present, large-scale mission decision-making still has the following problems: the increasing complexity of tasks and environments has put higher requirements on the efficiency and adaptability of decision-making methods. For heterogeneous UAVs, this complexity is further increased because the different capabilities of each UAV need to be considered; ignoring the current status of UAVs and tasks before making decisions will lead to a waste of computing and communication resources; the existing distributed mission decision-making methods based on market mechanisms and game theory require multiple communications between all UAVs, which puts a heavy burden on system communication. Summary of the invention

[0006] The purpose of this invention is to propose a multi-level dynamic task decision-making method for large-scale heterogeneous UAVs based on alliance game. Through directional communication between UAVs and the formation of a UAV alliance, task decision-making and dynamic adjustment of UAV resources are realized. The proposed decision-making method has excellent real-time performance and generalization ability.

[0007] The technical solution of the present invention is achieved in this way:

[0008] A multi-level dynamic task decision-making method for large-scale heterogeneous UAVs based on coalition game includes the following steps:

[0009] Step 1: Establish an optimization model for dynamic task decision-making of heterogeneous UAVs under multiple constraints, determine the constraints and design an evaluation function based on the UAV benefits, assign appropriate types and numbers of UAVs to tasks, establish an optimization model, and mathematically express the task decision-making problem of heterogeneous UAVs;

[0010] Step 2: Convert the dynamic task decision-making process into a coalition game solving process, establish a coalition game model, and determine the preference relationship of drones based on the coalition benefits. Based on this, design the coalition switching rules of drones, give the form of a stable solution to the coalition game, and determine the convergence conditions of the game;

[0011] Step 3: Based on the established alliance game model, a multi-level dynamic task decision strategy is designed; first, the UAVs and tasks involved in dynamic task decision-making are confirmed based on UAV requirements and task levels; second, the leader of dynamic tasks and some initial tasks is determined by the maximum weighted matching method; finally, after each task determines its leader, the remaining UAVs perform switching operations based on the preference relationship to respond to the task, the leader evaluates the response results of the UAVs, and resolves conflicts based on cost optimization, and finally obtains the task decision result.

[0012] Furthermore, the UAV needs to meet the following constraints when making dynamic mission decisions:

[0013] (a) UAV i and mission T j The type constraints are:

[0014]

[0015] Among them, i and j represent the drone and mission ID respectively; the binary variable p ij Indicates that when the type constraint is met, t_T j =t_V i Assign UAV i to task j under the condition of t_T j =t_V i Indicates that when the type of drone i is the same as the type of mission j, the drone can choose to perform the mission; l j ≤max(T_new) means that only tasks whose task level does not exceed the highest dynamic task level can participate in dynamic task decision making; Indicates that drones whose initial mission level does not exceed the highest dynamic mission level or drones whose status is idle or mission completed are eligible to participate in dynamic mission decision-making. represents the level of the initial mission k of UAV i;

[0016] (b) Capacity constraint: Since drones cannot perform multiple tasks at the same time, each drone can choose at most one task to perform, that is, satisfying:

[0017]

[0018] (c) Scale constraint: Considering that each task requires several drones to perform collaboratively, the number of drones assigned should not be less than the number of drones required for the task. At the same time, in order to prevent too many drones from selecting the same task, resulting in a waste of drone resources, an upper limit is set on the number of drones required for each task:

[0019]

[0020] Among them, J is the set of all tasks.

[0021] (d) Energy constraint: According to the actual situation, all UAVs need to retain enough energy to return to the base after completing the mission, which can be expressed as:

[0022]

[0023] Among them, e_V i Indicates the current remaining energy of drone i; e_T j represents the energy required to perform task j; g ij represents the flight cost of UAV i to reach task j and return; e_Th represents the minimum energy threshold of the UAV.

[0024] Furthermore, the profit evaluation function for assigning a suitable task to each drone includes the following steps:

[0025] (1) Establish an evaluation function for the UAV to perform tasks:

[0026]

[0027] Among them, r ij (T j ,|A j |) is the reward for drone i to perform task j; |A j | represents the number of drones participating in task j;

[0028] (2) The profit function r of drone i performing mission j ij (T j ,|A j |)The expression is as follows:

[0029]

[0030] Among them, u j is the actual value generated by task j in the entire task decision process;

[0031] (3) The actual value of the shared mission among drones performing the same mission u j :

[0032]

[0033] Among them, v j represents the preset value of task j and the actual value u obtained by the drone team from the task j Different from , this value remains unchanged throughout the task decision process; ε and η are both extremely small constants;

[0034] (4) The cost function expression for determining UAV i to perform task j is: bij =e_T j +g ij (8).

[0035] Furthermore, the method for establishing the alliance game model includes the following steps:

[0036] Step 1: Alliance definition and alliance partition construction:

[0037] An alliance refers to a group of drones assigned to the same mission. The drones in each alliance work together to perform a mission, and each drone can only choose one alliance.

[0038] In the task decision process, the coalition partition is defined as a disjoint coalition set Ω = {A0, A1, ..., A m+k},in represents a coalition of drones assigned to task j, satisfying and

[0039] Step 2: Establish a preference relationship based on global benefits:

[0040] Each participant drone formulates a strategy based on the preference relationship to decide whether to join or leave a certain alliance. j ,A j' , if the preference relation of drone i is A j ≥ i A j' , which means that drone i is more inclined to choose alliance A j Rather than A j' , or the willingness to join the two alliances is equal, the preference relationship of global benefits is established according to equations (6)-(7):

[0041]

[0042] in, Indicates equivalence relation; symbol > i Indicates a strict preference for one coalition; A j ∪{i} indicates that drone i participates in alliance A j ; A j' \{i} indicates that drone i leaves alliance A j' ;

[0043] Step 3: Alliance switching rules for drones:

[0044] Switching operations based on alliance partitions and preference relations is defined as Therefore, the alliance partitions from Ω{A0,A1,...,A m+k} is transformed into Ω'=(Ω\{A j ,A j'})∪{A j ∪{i},A j' \{i}}, the reward for this switching operation is calculated as follows:

[0045]

[0046] Step 4: Nash stable partition:

[0047] Alliance partition Ω={A0,A1,...,A m+k The condition for Nash stability is if and only if each drone i∈N prefers its current coalition and cannot increase the global payoff by switching operations. , Nash's stable alliance division means that drones can no longer be operated by switching Increase the total benefits of all alliances and reach a stable state. The Nash stable alliance partition obtained at this time is the solution to the game process and the final result of the task decision process.

[0048] Furthermore, based on the established alliance game model, a multi-level dynamic task decision-making method is designed as follows:

[0049] 1. Participant identification based on drone requirements and mission levels

[0050] According to formula (1), under the premise of meeting the low-level requirements such as drone safety and basic needs, participants are identified based on task level and drone status, which can be expressed as:

[0051]

[0052] 2. Determine the task leader based on the maximum weighted matching method

[0053] The leader is designated to store key mission information and iterative results in the decision-making process. These iterative results include the current alliance partition, drone revenue, and energy consumption. Drones interact with the leader to obtain this information and formulate strategies accordingly. The leader determination process is as follows:

[0054] First, the leader set is initialized to the empty set D e , based on the participant identification results, the payoff matrix is ​​updated as ; Secondly, based on the two-way matching method, the maximization task T is identified j If the optimal task of drone i* is T j , then designate it as T j Then, UAV i* is added to the leader set D e , and from the waiting queue Removed from the task T j From the list of pending machines Delete; finally update the profit matrix , to prevent i* and T j is selected again; this process accurately matches each task with its leader and establishes the initial alliance partition At this time, the task alliance participating in the reallocation only includes the designated leader; at the same time, all other drones participating in the reallocation are assigned to the virtual task alliance A0.

[0055] 3. UAV response based on alliance game

[0056] When each mission determines its leader, the remaining drones respond to the mission based on the switching operation; based on the initial alliance partition, the drones perform switching operations according to their preference relationships. To enhance the global benefits, until the alliance partition reaches Nash stability, an intermediate alliance is formed and the initial task decision result is obtained;

[0057] The drone obtains the current alliance partition from the leader and task information, including drone allocation, ID, execution cost, execution benefit, and current global benefit; then, the drone calculates the execution switching operation based on the preference relationship potential reward, choose the mission with the highest switching reward or stay in the current alliance; perform the switching operation After that, UAV i communicates its response result to the mission leader, triggering the alliance partition and proceed to the next stage;

[0058] 4. Conflict resolution based on cost optimization to ensure that the number of drones participating in the mission does not exceed the maximum number required by the mission;

[0059] According to the alliance division results of the previous stage , the mission leader will increase the number of drones in its alliance And the task limit max j Compare; if the number exceeds this limit, sort the drones in ascending order of execution cost and only keep those that do not exceed the limit max j The remaining drones will be reallocated to the Virtual Mission Alliance. ; If the number of drones for a mission is within the range [min j ,max j ], then the alliance The UAVs in are assigned to task j; the conflict resolution results are communicated between the leading UAVs, and the Nash stable alliance obtained after conflict resolution is taken as the final task decision result.

[0060] The beneficial effects of the present invention are as follows: (1) It solves the problem of low decision efficiency caused by the difficulty of solving large-scale heterogeneous UAV dynamic task decisions under multiple constraints and limited global communication. The present invention is based on the alliance game strategy, and the UAVs self-organize into independent alliances according to their own preferences, thereby improving the overall benefits of the alliance. At the same time, the task decision results are updated through directional communication between UAVs and leading aircraft, and between leading aircraft and leading aircraft, thereby reducing the number of interactions between UAVs.

[0061] (2) The conflict between the initial task and the dynamic task caused by the task change is solved. The multi-level control mechanism for the UAV requirements and task levels proposed in the present invention can first meet the low-level requirements of the UAV such as safety and basic requirements, and ensure that the high-level tasks in the initial task are given priority. At the same time, it ensures that the UAV can perform the task safely on the basis of meeting the requirements of the UAV.

[0062] (3) The present invention decomposes the complex dynamic task decision process into multiple stages, including the participant identification stage based on UAV requirements and task levels, the leader determination stage based on maximum weighted matching, the UAV response stage based on coalition game, and the conflict resolution stage based on cost optimization. The proposed multi-level dynamic task decision method improves the real-time performance of large-scale heterogeneous UAV task decision-making, while maximizing the global benefits, and has good scalability for heterogeneous UAV clusters. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 For the overall technical solution.

[0065] Figure 2 Decision time for the method with different numbers of drones.

[0066] Figure 3 The number of iterations of the method under different numbers of drones.

[0067] Figure 4 Decision time for methods with different dynamic task sizes. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0069] The present invention is further described below in conjunction with the accompanying drawings.

[0070] The overall technical solution of the implementation method of the present invention is as follows Figure 1 As shown, each part is described in detail below in conjunction with the accompanying drawings of the specification:

[0071] Scenario description: Assume that there are large-scale heterogeneous drones in a scenario, including reconnaissance, communication and strike drones, which have different reconnaissance, communication and strike capabilities. Drones perform tasks of their corresponding types according to their own capabilities. Considering that the urgency and importance of tasks vary, the corresponding values ​​and levels are also different. In addition, new tasks may appear in the environment at any time while drones are performing designated tasks. In this case, drones need to be reallocated to perform new tasks and ensure that global benefits are maximized.

[0072] The following defines the variables related to tasks and drones in the task decision scenario. Assume that there is a set of heterogeneous drones V = {V1, V2, ..., V n} and a set of initial tasks Task_ini={T1,T2,...,T m}, where n is the total number of drones and m is the total number of initial tasks in the scene. The newly added dynamic task is represented by Task_new = {T m+1 ,T m+2 ,...,T m+k}, where k represents the number of newly added dynamic tasks. At the same time, a virtual task T0 is established at the drone base to integrate drones that have not been assigned tasks.

[0073] 1. Task-related variables: All tasks in the scenario are represented by T = {T0, T1, ..., T m ,T m+1 ,...,T m+k Each task can be represented by a tuple <TI, Tt, Tp, Ts, Tu, Tv, Tl, ene_d, min_d, max_d>, where TI is the task ID; Tt is the task type, including reconnaissance, communication and attack; the task position Tp is a fixed parameter; Ts is the task importance, Tu is the task urgency, both of which are integers ranging from 50 to 100; Tv is the task value, which is obtained by weighting the task type Tt, task position Tp, task importance Ts and task urgency Tu; Tl is the task level: l j ≥90 tasks are level 3, 70≤1j <90 tasks are level 2, l j Tasks with energy consumption less than 70 are at level 1; ene_d is the energy consumption required to execute the task; min_d is the minimum number of drones required to complete the task; max_d is the maximum number of drones required to complete the task.

[0074] 2. Drone related variables: All drones in the scene are represented by V = {V1, V2, ..., V n Each drone can be represented by a tuple <VI, Vt, Vp, Ve, Vs>, where VI is the drone ID; Vt is the drone type, including reconnaissance, communication and attack; Vp is the drone position; Ve is the remaining energy of the drone; Vs is the drone status, including four types: 0 (idle), 1 (flying), 2 (performing a mission) and 3 (mission completed). Except for the drones that are performing tasks (Vs = 2) and are located at the locations where they perform the tasks, other drones are randomly distributed in the scene.

[0075] The specific parameter settings are shown in Table 1.

[0076] Table 1 Task decision scenario parameter settings

[0077]

[0078] Step 1: Establish a dynamic task decision optimization model for large-scale heterogeneous UAVs under multiple constraints and determine the constraints and evaluation functions.

[0079] Constraints

[0080] (1) Type Constraints

[0081] Assume that the mission types include reconnaissance, communication and attack, and the types of drones include reconnaissance, communication and attack. According to its own capabilities, each drone can only perform its corresponding type of mission, so drone i and mission T j The type constraint is expressed as:

[0082]

[0083] Among them, i and j represent the drone and mission ID respectively; the binary variable p ij Indicates that when the type constraint t_T is satisfied j =t_V i Assign UAV i to task j under the condition of t_T j =t_V i Indicates that when the type of drone i is the same as the type of mission j, the drone can choose to perform the mission; l j≤max(T_new) ensures that only tasks whose task level does not exceed the highest dynamic task level can participate in dynamic task decision making; Indicates that drones whose initial mission level does not exceed the highest dynamic mission level or drones whose status is idle or mission completed are eligible to participate in dynamic mission decision-making. Represents the level of the initial mission k of UAV i.

[0084] (2) Capacity constraints

[0085] Since drones cannot perform multiple tasks at the same time, each drone can choose at most one task to perform, that is, satisfying:

[0086]

[0087] (3) Scale constraints

[0088] Considering that each task requires several drones to perform collaboratively, the number of drones allocated should not be less than the number of drones required for the task. At the same time, in order to prevent too many drones from choosing the same task, resulting in a waste of drone resources, an upper limit is set for the number of drones required for each task:

[0089]

[0090] Among them, J is the set of all tasks.

[0091] (4) Energy constraints

[0092] According to the actual situation, all drones need to retain enough energy to return to the base after completing the mission, which can be expressed as:

[0093]

[0094] Among them, e_V i Indicates the current remaining energy of drone i; e_T j represents the energy required to perform task j; g ij represents the flight cost of UAV i to reach task j and return; e_Th represents the minimum energy threshold of the UAV.

[0095] (5) Evaluation function

[0096] Taking into account the mission value, the number of drones selected for the mission, and the cost of executing the mission, an evaluation function for drone mission execution is established to achieve global benefit maximization and efficient use of drone resources:

[0097]

[0098] Among them, r ij (Tj ,|A j |) is the reward for drone i to perform task j; |A j | represents the number of drones participating in task j. The profit function r of drone i performing task j ij (T j ,|A j |)The expression is as follows:

[0099]

[0100] Among them, u j is the actual value generated by task j in the entire task decision process. The actual value u of the shared task by drones performing the same task j :

[0101]

[0102] Among them, v j represents the preset value of task j and the actual value u obtained by the drone team from the task j Different from , this value remains unchanged throughout the task decision process; ε and η are both extremely small constants. The cost function expression of drone i performing task j is:

[0103] b ij =e_T j +g ij (8)

[0104] The multi-level dynamic task decision method of the present invention aims to assign a suitable task to each UAV and maximize the evaluation function (5) while satisfying the constraints (1)-(4) such as type, capability, scale and energy.

[0105] Step 2: Convert the dynamic task decision-making problem into a coalition game process and establish a coalition game model.

[0106] 1. Alliance definition and construction of alliance partitions:

[0107] In coalition game theory, participants cooperate by forming coalitions to enhance individual and overall benefits. A coalition refers to a group of drones assigned to the same task. The drones in each coalition work together to perform a task, and each drone can only choose one coalition.

[0108] In the task decision process, the coalition partition is defined as a disjoint coalition set Ω = {A0, A1, ..., A m+k},in represents a coalition of drones assigned to task j, satisfying and

[0109] 2. Preference Relationship

[0110] For participant drone i, the preference relation It defines which alliance it prefers to join in the two alliances. During the game, each participant drone formulates a strategy based on the preference relationship to decide whether to join or leave a certain alliance. j ,A j' , if the preference relation of drone i is A j ≥ i A j' , which means that drone i is more inclined to choose alliance A j Rather than A j' , or the willingness to join the two alliances is equal. According to formula (6)-formula (7), a preference relationship based on global benefits is established:

[0111]

[0112] in, Indicates equivalence relation; symbol > i Indicates a strict preference for one coalition; A j ∪{i} indicates that drone i participates in alliance A j ; A j' \{i} indicates that drone i leaves alliance A j' This preference relationship emphasizes that drones need to consider not only their own benefits but also the global benefits when choosing an alliance.

[0113] 3. Switching rules

[0114] Based on the coalition game model and preference relations (9), the switching operation is defined as Therefore, the alliance partitions from Ω = {A0, A1, ..., A m+k} is transformed into Ω'=(Ω\{A j ,A j'})∪{A j ∪{i},A j' \{i}}, the reward for this switching operation is calculated as follows:

[0115]

[0116] 4. Nash Stable Partition

[0117] Alliance partition Ω={A0,A1,...,A m+k The condition for Nash stability is if and only if each drone i∈N prefers its current coalition and cannot increase the global payoff by switching operations. Nash's stable alliance division means that drones can no longer be operated by switching Increase the total benefits of all alliances (5) and reach a stable state. The Nash stable alliance partition obtained at this time is the solution to the game process and the final result of the task decision process.

[0118] Step 3: Based on the established alliance game model, design a multi-level dynamic task decision-making strategy.

[0119] The multi-level dynamic task decision strategy divides the complex dynamic task decision process into four stages to enhance the real-time decision-making and reduce the communication cost. These four stages include: participant identification based on UAV requirements and task levels, leader determination based on maximum weighted matching, UAV response based on coalition game, and conflict resolution based on cost optimization.

[0120] 1. Participant identification based on drone requirements and mission levels

[0121] Drawing on Maslow's theory of human psychological needs hierarchy, the needs of drones are divided into multiple levels based on the priority queue mechanism: 1. Safety needs; 2. Basic needs, such as energy, time constraints, etc.; 3. Capability needs, such as communication capabilities, reconnaissance capabilities, etc.; 4. Cooperation needs, such as global benefits, etc. Among them, safety needs are the lowest level of needs, and low-level needs are the premise for meeting high-level needs. According to formula (1), on the premise of meeting low-level needs such as drone safety and basic needs, participants are identified based on task level, drone status, etc., expressed as:

[0122]

[0123] 2. Leader determination based on maximum weighted matching

[0124] The leader of dynamic tasks and some initial tasks is determined based on the maximum weighted matching method. The leader is designated to store key task information and iterative results in the decision-making process, including the current alliance partition, drone benefits, and energy consumption. The drone obtains this information and formulates strategies accordingly by interacting with the leader. The leader determination process is as follows:

[0125] First, the leader set is initialized to the empty set D e , based on the participant identification results, the payoff matrix is ​​updated as Secondly, based on the two-way matching method, the maximization task T is identified. j If the optimal task of drone i* is T j , then designate it as T j Then, drone i* is added to the leader set D e , and from the waiting queue Removed from the task T j From the list of pending machines Finally update the profit matrix , to prevent i* and T j This process accurately matches each task with its leader and establishes the initial dynamic alliance partition At this point, the task alliance participating in the reallocation only includes the designated leader. At the same time, all other drones participating in the reallocation are assigned to the virtual task alliance A0.

[0126] 3. UAV response based on alliance game

[0127] After each mission determines its leader, the remaining drones respond to the mission based on the switching operation. Based on the initial alliance partition, the drones perform switching operations according to their preference relations (9) To enhance the global benefit (5), until the alliance partition reaches Nash stability and obtains the initial task decision result.

[0128] The drone obtains the current alliance partition from the leader (τ is the current iteration number) and task information, including drone allocation, ID, execution cost, execution benefit, and current global benefit. Then, the drone calculates the execution switching operation according to the preference relationship (9) Potential rewards (10), choose the mission with the highest switching reward or stay in the current alliance. Perform the switching operation After that, UAV i communicates its response result to the mission leader, triggering the alliance partition and proceed to the next stage.

[0129] 4. Conflict resolution based on cost optimization

[0130] Considering that too many drones may choose the same task, resulting in unnecessary allocation and waste of drone resources, the mission leader needs to evaluate the response results of the drones and then perform conflict resolution based on cost optimization to ensure that the number of drones participating in the mission does not exceed the maximum number limit required for the mission.

[0131] According to the alliance division results of the previous stage , the mission leader will increase the number of drones in its alliance And the task limit max j If the number exceeds this limit, the drones are sorted in ascending order of execution cost, and only those that do not exceed the limit are retained. j The remaining drones will be reallocated to the Virtual Mission Alliance. If the number of drones for a mission is within the range [minj ,max j ], then the alliance The UAVs in are assigned to task j. The conflict resolution results are communicated between the leading UAVs, and the Nash stable alliance obtained after conflict resolution is is taken as the final task decision result.

[0132] Simulation verification:

[0133] In order to evaluate the effectiveness of the proposed strategy, this section first gives the scenario settings used by the present invention, and then gives the task decision results in this scenario, and simulates different numbers of drones and tasks to evaluate its effectiveness and adaptability. In addition, the proposed strategy is compared with the existing task decision algorithm to verify the rapidity and superiority of the present invention for large-scale drone task decision-making.

[0134] 1. Simulation environment and parameters

[0135] The simulation process is carried out in the Windows 11 operating platform, and the algorithm is implemented by programming in Python. Assume that there are 100 drones, 4 initial tasks and 1 new dynamic task, each task includes reconnaissance, communication and strike subtasks. The types and numbers of heterogeneous drones are set as follows: No. 1-35 are reconnaissance drones, No. 36-60 are communication drones, and No. 61-100 are strike drones. In the initial state of the scene, except for the idle drone (s_V i =0), the states of the remaining drones are randomly selected in the flight, mission execution and mission completion. i =2), the remaining drones (s_V i =1,3) are randomly distributed in the scene. The energy reserve of the drone is set between 60 and 100, and the minimum energy threshold e_Th=30.

[0136] 2. Simulation results of typical scenarios

[0137] In order to verify the effectiveness of the proposed task decision method, based on the above simulation parameter settings, the task decision results are shown in Table 2. The subtask levels of each task are T_l = {l0 = 0, l1 = 3, l2 = 3, l3 = 3, l4 = 2, l5 = 2, l6 = 2, l7 = 1, l8 = 1, l9 = 1, l 10 =2,l 11 =2,

[0138] l 12 =2,l 13 =2,l 14 =2,l 15=2}. It can be seen from Table 2 that high-level subtasks 1-3 do not participate in the reallocation, while the drones that complete the tasks in the initial alliance (s_V i =3) participate in dynamic mission decision making, and the remaining drones (s_V i =1,2) continue to execute the original task; the initial alliance A0, i.e. all drones executing subtasks 4-12, participate in the task decision and perform the switching operation based on the switching reward (10) , join a new alliance or stay in the original alliance to perform the initial task. In 10,000 simulations, the average number of iterations to reach Nash stability is 4.23, and the average running time is 0.0165 seconds.

[0139] Table 2 Dynamic task decision results

[0140]

[0141] 3. Algorithm performance verification

[0142] To evaluate the scalability and effectiveness of the proposed strategy, simulation experiments are conducted with different numbers of UAVs and missions. Figure 2-Figure 3 The effect of different numbers of drones on task decision time and iteration number is shown. As the number of drones increases, the frequency of conflicts among multiple drones choosing the same task increases, and more iterations are required to achieve stable partitioning, so the decision time also increases accordingly. Figure 4The decision time required for the proposed strategy to reach Nash equilibrium in different numbers of dynamic tasks is shown. As the number of dynamic tasks increases, the number of tasks that the drone can choose increases, which will result in more switching operations and result adjustments, resulting in an increase in decision time. Despite the increase in the number of drones and tasks, the proposed strategy always reaches convergence within a limited number of iterations, demonstrating its scalability and adaptability in dynamic task decision problems. In order to verify the superiority and real-time performance of the proposed strategy, it is compared with the GRAPE and CDTA algorithms respectively, and the GRAPE and CDTA algorithms are appropriately modified to adapt to the scenario. As shown in Tables 3 and 4, in scenarios with different numbers of drones, the decision time of the proposed strategy is 0.003s-0.006s faster than the CDTA algorithm, and 0.005s-0.008s faster than the GRAPE algorithm; for different numbers of dynamic tasks, the decision time of the proposed strategy can also be at least 0.005s faster than the GRAPE and CDTA algorithms, verifying the excellent real-time performance and fast convergence of the proposed task decision strategy in different numbers of drones and dynamic tasks. In order to verify the superiority of the proposed strategy, the average benefits of the three algorithms are compared based on the evaluation function. As shown in Table 5, the average benefit of the proposed strategy for UAV task execution is higher than that of the GRAPE algorithm, but lower than that of the CDTA algorithm. This difference occurs because the CDTA algorithm does not consider the safety and energy constraints of UAVs, while the GRAPE algorithm requires all UAVs to perform tasks, resulting in unnecessary allocation of UAVs. In addition, in the proposed strategy and the CDTA algorithm, the increase in the number of UAVs allows more UAVs with lower flight costs to be used to perform corresponding tasks, so that as the number of UAVs increases, the average benefit of UAVs will also increase.

[0143] Table 3 Comparison of running time of different algorithms and the proposed algorithm under different numbers of drones

[0144]

[0145]

[0146] Table 4 Comparison of running time of different algorithms and the proposed algorithm under different numbers of dynamic tasks

[0147]

[0148] Table 5 Comparison of average benefits between different algorithms and the proposed algorithm under different numbers of drones

[0149]

[0150] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A multi-level dynamic task decision-making method for large-scale heterogeneous UAVs based on alliance game, characterized by: The following steps are involved: Step 1: Establish a dynamic task decision optimization model for heterogeneous UAVs under multiple constraints, determine the constraints and design an evaluation function based on the UAV benefits, assign appropriate types and numbers of UAVs to the tasks, and mathematically express the task decision problem of heterogeneous UAVs; Step 2: Convert the dynamic task decision-making process into a coalition game solving process, establish a coalition game model, and determine the preference relationship of drones based on the coalition benefits. Based on this, design the coalition switching rules of drones, give the form of a stable solution to the coalition game, and determine the convergence conditions of the game; Step 3: Based on the established alliance game model, a multi-level dynamic task decision strategy is designed; first, the UAVs and tasks involved in dynamic task decision-making are confirmed based on UAV requirements and task levels; second, the leader of dynamic tasks and some initial tasks is determined by the maximum weighted matching method; finally, after each task determines its leader, the remaining UAVs perform switching operations based on the preference relationship to respond to the task, the leader evaluates the response results of the UAVs, and resolves conflicts based on cost optimization, and finally obtains the task decision result.

2. According to claim 1, a large-scale heterogeneous UAV multi-level dynamic task decision-making method based on alliance game is characterized in that: The UAV needs to meet the following constraints when making dynamic mission decisions: (a) UAV i and mission T j The type constraints are: Among them, i and j represent the drone and mission ID respectively; the binary variable p ij Indicates that when the type constraint is met, t_T j =t_V i Assign UAV i to task j under the condition of t_T j =t_V i Indicates that when the type of drone i is the same as the type of mission j, the drone can choose to perform the mission; l j ≤max(T_new) means that only tasks whose task level does not exceed the highest dynamic task level can participate in dynamic task decision making; Indicates that drones whose initial mission level does not exceed the highest dynamic mission level or drones whose status is idle or mission completed are eligible to participate in dynamic mission decision-making. represents the level of the initial mission k of UAV i; (b) Capacity constraint: Since drones cannot perform multiple tasks at the same time, each drone can choose at most one task to perform, that is, satisfying: (c) Scale constraint: Considering that each task requires several drones to perform collaboratively, the number of drones assigned should not be less than the number of drones required for the task. At the same time, in order to prevent too many drones from selecting the same task, resulting in a waste of drone resources, an upper limit is set on the number of drones required for each task: Among them, J is the set of all tasks. (d) Energy constraint: According to the actual situation, all UAVs need to retain enough energy to return to the base after completing the mission, which can be expressed as: Among them, e_V i Indicates the current remaining energy of drone i; e_T j represents the energy required to perform task j; g ij represents the flight cost of UAV i to reach task j and return; e_Th represents the minimum energy threshold of the UAV.

3. According to claim 2, a large-scale heterogeneous UAV multi-level dynamic task decision-making method based on alliance game is characterized in that: The evaluation function of the task decision result includes the following steps: (1) Establish an evaluation function for the UAV to perform tasks: Among them, r ij (T j ,|A j |) is the reward for drone i to perform task j; |A j | represents the number of drones participating in task j; (2) The profit function r of drone i performing mission j ij (T j ,|A j |)The expression is as follows: Among them, u j is the actual value generated by task j in the entire task decision process; (3) The actual value of the shared mission among drones performing the same mission u j : Among them, v j represents the preset value of task j and the actual value u obtained by the drone team from the task j Different from , this value remains unchanged throughout the task decision process; ε and η are both extremely small constants; (4) The cost function expression for determining UAV i to perform task j is: b ij =e_T j +g ij (8).

4. According to claim 3, a large-scale heterogeneous UAV multi-level dynamic task decision-making method based on alliance game is characterized in that: The method for establishing the alliance game model comprises the following steps: Step 1: Definition of alliance and construction of alliance partitions: An alliance refers to a group of drones assigned to the same mission. The drones in each alliance work together to perform a mission, and each drone can only choose one alliance. In the task decision process, the coalition partition is defined as a disjoint coalition set Ω = {A0, A1, ..., A m+k },in represents a coalition of drones assigned to task j, satisfying and j≠j'; Step 2: Establish a preference relationship based on global benefits: Each participant drone formulates a strategy based on the preference relationship to decide whether to join or leave a certain alliance. j ,A j' , if the preference relation of drone i is This means that UAV i is more inclined to choose Alliance A j Rather than A j' , or the willingness to join the two alliances is equal, the preference relationship of global benefits is established according to equations (6)-(7): in, Indicates equivalence relations; symbols Indicates a strict preference for one coalition; A j ∪{i} indicates that drone i participates in alliance A j ; A j' \{i} indicates that drone i leaves alliance A j' ; Step 3: Alliance switching rules for drones: Switching operations based on alliance partitions and preference relations is defined as Therefore, the alliance partitions from Ω={A0,A1,...,A m+k } is transformed into Ω'=(Ω\{A j ,A j' })∪{A j ∪{i},A j' \{i}}, the reward for this switching operation is calculated as follows: Step 4: Nash stable partition: Alliance partition Ω={A0,A1,...,A m+k The condition for Nash stability is if and only if each drone i∈N prefers its current coalition and cannot increase the global payoff by switching operations. Nash's stable alliance division means that drones can no longer be operated by switching Increase the total benefits of all alliances and reach a stable state. The Nash stable alliance partition obtained at this time is the solution to the game process and the final result of the task decision process.

5. According to claim 4, a large-scale heterogeneous UAV multi-level dynamic task decision-making method based on alliance game is characterized in that: Based on the established alliance game model, the multi-level dynamic task decision-making method is designed as follows:

1. Participant identification based on drone requirements and mission levels According to formula (1), under the premise of meeting the low-level requirements such as drone safety and basic needs, participants are identified based on task level and drone status, which can be expressed as:

2. Determine the task leader based on the maximum weighted matching method The leader is designated to store key mission information and iterative results in the decision-making process. These iterative results include the current alliance partition, drone revenue, and energy consumption. Drones interact with the leader to obtain this information and formulate strategies accordingly. The leader determination process is as follows: First, the leader set is initialized to the empty set D e , based on the participant identification results, the payoff matrix is ​​updated as Secondly, based on the two-way matching method, the maximization task T is identified. j If the optimal task of drone i* is T j , then designate it as T j Then, UAV i* is added to the leader set D e , and from the waiting queue Removed from the task T j From the list of pending machines Delete; finally update the profit matrix To prevent i* and T j is selected again; this process accurately matches each task with its leader and establishes the initial alliance partition At this time, the task alliance participating in the reallocation only includes the designated leader; at the same time, all other drones participating in the reallocation are assigned to the virtual task alliance A0; 3. UAV response based on alliance game When each mission determines its leader, the remaining drones perform switching operations based on the preference relationship to respond to the mission; Based on the initial coalition partition, drones perform switching operations according to their preference relations. To enhance the global benefits, until the alliance partition reaches Nash stability, an intermediate alliance is formed and the initial task decision result is obtained; The drone obtains the current alliance partition from the leader and task information, including drone allocation, ID, execution cost, execution benefit, and current global benefit; The drone then performs the switching operation based on the preference relationship calculation. potential reward, choose the mission with the highest switching reward or stay in the current alliance; perform the switching operation After that, UAV i communicates its response result to the mission leader, triggering the alliance partition and proceed to the next stage; 4. Conflict resolution based on cost optimization to ensure that the number of drones participating in the mission does not exceed the maximum number required by the mission; According to the alliance division results of the previous stage The mission leader will increase the number of drones in its alliance And the task limit max j Compare; if the number exceeds this limit, sort the drones in ascending order of execution cost and only keep those that do not exceed the limit max j The remaining drones will be reallocated to the Virtual Mission Alliance. If the number of drones for a mission is within the range [min j ,max j ], then the alliance The UAVs in are assigned to task j; the conflict resolution results are communicated between the leading UAVs, and the Nash stable alliance obtained after conflict resolution is taken as the final task decision result.

Images