A method for dynamic allocation of collaborative tasks among multiple drone swarms

Abstract

The present invention discloses a method for dynamically allocating collaborative tasks of a multi-UAV swarm, including step S1, establishing a collaborative task allocation model; step S2, optimizing the task allocation strategy of the collaborative task allocation model of step S1 based on a selection optimization method; making a circle with a certain radius with each task as the center, adjusting the selection of each task to the Agent within the circle to achieve strategy optimization, and the selection mechanism ensures that each iteration moves the allocation strategy of the Multi Agent system in a more optimal direction, and after a limited number of iterations, the Multi Agent system will obtain the optimal allocation strategy. The selection optimization method of the present invention uses optional UAVs within a certain range as optimization objects, reduces the scale of possible solutions, and improves the optimization speed. The optimal selection algorithm ensures that the allocation strategy continues to approach the optimal allocation strategy as the iteration process progresses. It meets the requirements of fast and dynamic task allocation, and quickly obtains suboptimal strategies under specific circumstances to achieve collaborative task allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method for dynamically allocating collaborative tasks among a swarm of unmanned aerial vehicles. Background Art

[0002] There are high requirements and many factors to consider in the process of multi-swarm collaborative task allocation. In the actual battlefield environment, tasks enter the system independently and randomly, and it is difficult to predict the time, location, type, etc. of the tasks, which brings a series of problems to task allocation:

[0003] 1. Real-time performance: In the dynamic task allocation problem, the changes in factors such as environment, members, and tasks require the use of effective task allocation methods to achieve rapid decision-making. Otherwise, the allocation algorithm will take too long and will delay the opportunity. However, in the actual task allocation process, the battlefield situation changes rapidly, and the emergence of new tasks is difficult to predict. Real-time response to these situations places high demands on the task allocation algorithm.

[0004] 2. Scale limitation: The application of drone swarms in future wars will be large-scale. However, the increase in the number of drones and targets will cause the scale of the task allocation algorithm to grow at an exponential rate, limiting the scale of dynamic task allocation and bringing huge challenges to solving a satisfactory allocation strategy.

[0005] 3. Coordination requirements: The goal of large-scale drone swarms is to achieve multi-machine coordination. The coordination between multiple drones can reflect the advantages of drones in executing tasks, better exert the performance of drones, improve the success rate of mission execution, and reduce costs. There is currently no appropriate evaluation index for coordination, and coordination is only considered from the aspects of grouping, time sequence of task execution, and grouping of weapons carried.

[0006] 4. Hybridity: The task allocation system should be able to manage different types of members, who may differ in software structure, hardware composition, etc., and may also complete different tasks. This requires the task allocation system to be an open and extensible system.

[0007] The factors that need to be considered in the multi-UAV dynamic task allocation model are:

[0008] 1. Threats: There are two types of threats when drones are performing missions: known threats and unknown threats. Known threats are ground-based air defense firepower, radar, and air threats that have been identified before the mission. Unknown threats are new and unforeseen threats that emerge during the drone mission. The drone mission system must take corresponding measures for different threats, including threat assessment, reporting threats to other drones and the ground command center, and avoiding threats or launching attacks.

[0009] 2. Obstacles: mainly refer to terrain obstacles, which are determined before the mission is executed. Obstacles need to be avoided when the task is assigned. Low-altitude penetration brings new problems to the flight safety of UAVs: after entering the enemy's defense zone, flying too high will increase the probability of being threatened, and flying too low will increase safety risks.

[0010] 3. Own forces: When making dynamic task allocations, one must take into account one's own helicopters, etc., to achieve coordination among all combat forces.

[0011] In addition, in the problem of swarm collaborative task allocation, the strategy set that maximizes the total system benefit is defined as the optimal strategy. The optimal strategy means that all drones perform tasks in the optimal allocation method to achieve the coordination of all drones. The optimal strategy can be solved in special problems through the monotonicity of the state space, using linear and nonlinear programming methods, etc. However, in the collaborative task allocation of multiple drones, multiple constraints are coupled with each other and have complex relationships. There is no monotonicity between multiple variables, so it is difficult to obtain the optimal strategy using the above methods.

[0012] Due to the changing battlefield situation, there is no fixed allocation strategy to follow in task allocation. The decision made at each moment will have an impact on the total benefit of the system. The optimal strategy is generally obtained through intelligent optimization algorithms, and system structure optimization can also help to obtain the optimal strategy.

[0013] The system has an optimal strategy. Since the total system time is finite, the state space is discrete and finite, the number of drones and tasks is also finite, and the space of optional strategy sets in the system will also be finite. Therefore, there must be a strategy set in the system whose performance is not inferior to other possible solutions, and this solution is the optimal solution.

[0014] Although the optimal strategy must exist, the scale of possible solutions is huge, and it is extremely difficult to find the optimal strategy. The method of using traversal to compare all possible solutions in the possible solution space to obtain the optimal solution cannot meet the requirements of fast and dynamic task allocation. At the same time, it is not necessary to obtain the global optimal strategy in certain situations. Quickly obtaining the suboptimal strategy is the key to achieving collaborative task allocation.

[0015] In view of the above-mentioned deficiencies in the prior art, the present invention is proposed. Summary of the invention

[0016] In view of the above problems, the object of the present invention is to provide a method for dynamically allocating collaborative tasks among multiple UAV swarms.

[0017] To achieve the purpose of the invention, the technical solution provided by the invention is: a method for dynamically allocating collaborative tasks of a multi-UAV swarm, comprising the following steps:

[0018] Step S1, establishing a collaborative task allocation model;

[0019] Based on the Multi-Agent system, each UAV is regarded as an agent, and tasks are assigned and executed by giving the agent autonomous capabilities;

[0020] The collaborative task allocation model is described as consisting of five groups:

[0021] {Time,Task,Agent,Policy_Set,Objective_Function} (Formula 1)

[0022] Among them, Time is time, Task is task, Agent is drone, Policy_Set is policy set, and Objective_Function is evaluation function;

[0023] Step S2, optimizing the task allocation strategy of the collaborative task allocation model of step S1 based on the selection optimization method;

[0024] Draw a circle with a certain radius with each task as the center, and adjust the selection of agents within the circle for each task to achieve strategy optimization. The selection mechanism ensures that each iteration moves the Multi Agent system allocation strategy in a better direction. After a limited number of iterations, the Multi Agent system will obtain the optimal allocation strategy.

[0025] in,

[0026] The task strategy optimization of the selection optimization method in step S2 is specifically as follows:

[0027] Assume that there are M agents and N tasks in the Multi Agent system;

[0028] Step S21: Task T k The position is the center of the circle, with r k Draw a circle with radius ; the radius of the circle must satisfy: there are at least n min Agents and the radius of each circle is greater than r min .

[0029] Step S22: generating an initial allocation strategy;

[0030] For each task in the Multi-Agent system, an agent is randomly selected and the decision variable at this time is represented as D t (0), calculate the profit value:

[0031]

[0032] in k∈N, and satisfies: When i≠j and M<N, there will be (N - M) tasks not assigned for execution;

[0033] Step S23: Randomly select a task Adjust this task to and satisfy: a k ≠a′ k , and represent the allocation strategy at this time as D t (1);

[0034] Step S24: Policy rationality check: Determine whether there exist in D t (1): i, j ∈ N, k≠i, if so, Adjust to to make D t (1) satisfy: i≠j, i, j ∈ N, and represent the allocation strategy at this time as D t (2), and the policy benefit value is:

[0035]

[0036] Step S25:: For D t (2), for the remaining tasks other than k and i x ∈ N, x≠k, i, sequentially select the Agent within the circle where it is located, and represent the newly adjusted strategy as D t (3), i ∈ N, to make

[0037]

[0038] satisfy: v3≥v2;

[0039] Step S26: Update the optimal strategy of the Multi - Agent system. Define the optimal strategy as v max . If v max <v3, then v max =v3;

[0040] Step S27: If v max meets the requirements of the Multi - Agent system, the optimization process ends, otherwise, go to Step S23.

[0041] The beneficial effects of the present invention include:

[0042] The selection and optimization method of the present invention takes the optional drones within a certain range as the optimization object, reduces the scale of possible solutions, and improves the optimization speed. The optimal selection algorithm ensures that the allocation strategy continuously approaches the optimal allocation strategy with the iteration process. It meets the requirements of fast and dynamic task allocation, and can quickly obtain a sub - optimal strategy in specific situations to achieve collaborative task allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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 or the description of the prior art 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.

[0044] Figure 1 This is a schematic diagram of the calculation of the encounter time and position between the UAV and the mission;

[0045] Figure 2 Schematic diagram of the impact of the number of drones and the number of missions on the scale of possible solutions;

[0046] Figure 3 Schematic diagram of the impact of the number of drones and the number of targets on the possible solution scale;

[0047] Figure 4 Schematic diagram of the effect of task type on the scale of possible solutions;

[0048] Figure 5 Flowchart of the strategy optimization steps for selecting the optimization method. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings.

[0050] The multi-UAV swarm collaborative task allocation method of the present invention comprises the following steps:

[0051] Step S1, establishing a collaborative task allocation model;

[0052] The multi-UAV collaborative task allocation model of the present invention is based on the Multi Agent system, which regards each UAV as an Agent and allocates and executes tasks by giving the Agent autonomous capabilities. There are two types of Agents in the system: decision-making Agents and member Agents. The decision-making Agent has the ability to make task allocation decisions, obtains system status information and makes task allocation decisions for member Agents; member Agents collaboratively execute the assigned tasks and feedback their status.

[0053] Each agent in the collaborative task allocation model has a different initial state. The Multi-Agent system allocates tasks according to task attributes and agent states, and completes state information exchange and decision-making instructions through data links.

[0054] The collaborative task allocation model is described as follows:

[0055] Based on the theory of Multi-Agent Markov decision process and referring to the modeling method of Markov theory, the model is established to consist of the following five groups:

[0056] {Time,Task,Agent,Policy_Set,Objective_Function} (Formula 1)

[0057] The parts are described as follows:

[0058] 1. Time

[0059] The Multi Agent system time is a finite value, with finite discrete time points t0, t1, t2, ..., t e Indicates that t0, t e are the start and end times respectively.

[0060] 2. Task

[0061] In the collaborative task allocation model, the drone's execution of a certain task or attack on a certain target is collectively referred to as a task. The task type has an important impact on the scale of task allocation. The more task types there are, the more the order of task execution that needs to be considered when allocating tasks will increase the scale of possible solutions dramatically.

[0062] The increase in task types will dramatically increase the scale of possible solutions, seriously affecting the efficiency of dynamic task allocation. To simplify the problem modeling, a single task type is used, and the tasks assigned to each agent are regarded as tasks of the same type. When each target enters the Multi Agent system, all tasks are decoupled into different and independent tasks through the task evaluation system, and different initial states are set for each task. By decoupling tasks and controlling the time and order of their entry into the task queue, the problem of dealing with multiple task types and execution order can be avoided, effectively reducing the scale of possible solutions and improving efficiency.

[0063] At time t, the number of new tasks in the Multi Agent system is n(t), which obeys the distribution θ(n(t)), n(t)∈Θ, and satisfies:

[0064]

[0065] Where P(·) represents the probability and D(·) is the correlation function. All tasks in the Multi Agent system at time t are expressed as:

[0066] [T1,T2,K,T N ] (Formula 3)

[0067] T n The state at time t is:

[0068]

[0069] in, Respectively represent the time t T n The position and velocity of are two-dimensional vectors, ζ n Execute task T for Agent n The expected consumption, including threat cost, weapon consumption, etc., is n When entering the Multi Agent system, it is decided. n To simplify the modeling process, it is assumed that each task has a randomly determined speed and direction when entering the Multi-Agent system, and it remains unchanged during the execution of the task. n Execute task T for Agent n The reward value, and w n > n >0. The size of the reward value reflects the importance of each task. A larger reward value indicates that the task is important, and the Multi Agent system will gain more benefits by executing the task.

[0070] The task queue Φ refers to the list of tasks that need to be executed currently:

[0071] Φ={T1,T2,K,T k ,...}, k=1,2,3,... (Formula 5)

[0072] 3. Drone (Agent)

[0073] Each drone with the ability to perform tasks and make autonomous decisions is regarded as an agent, expressed as:

[0074] a m ,m=1,2,...,M (Formula 6)

[0075] The Agent is set as follows: the Agent is a rigid body, the ground coordinates are inertial coordinates, the ground is considered a plane, and the gravity acceleration does not change with height. During the execution of the task, the state of each Agent changes over time, and its state information includes:

[0076]

[0077] in, for a m At time t, the state are position and velocity information respectively, for a m Attack capability at time t. for a mAssign task status at time t. Indicates that the Agent is currently not assigned a task. Indicates that the Agent has currently been assigned T n .

[0078] set up:

[0079] (1) All agents have the same speed when performing tasks and it is a fixed value. When no task is assigned, the agent hovers at the current position at a fixed speed and is considered to have unchanged position.

[0080] (2) All agents consume the same amount of time per unit;

[0081] (3) Each agent can only perform one task at a time;

[0082] (4) Different agents have the same benefit value when completing the same task, but different consumption values;

[0083] (5) The agent is regarded as a point mass and the turning radius is not considered, so the agent's communication capability is limited and the communication range is limited;

[0084] (6) The problem of collision avoidance between agents is not considered;

[0085] (7) Task allocation adopts a single-step planning mechanism;

[0086] (8) The route cost of an agent performing a task is proportional to the flight distance of the agent.

[0087] 4. Policy_Set

[0088] At time t, the number of agents and tasks in the Multi Agent system are denoted as m and n respectively. The optional task allocation strategy can be expressed as: D(t) = {T1', T'2, K, T' m} T (Formula 8)

[0089] Where D(t) is an m×1 matrix, T i '∈(T1,T2,K,T n ). Considering the mission and the number of drones, the scale of possible solutions is:

[0090]

[0091] in:

[0092]

[0093] (Equation 9) satisfies the following constraints:

[0094] (1) Each agent can only perform one task at any time, or it can perform no tasks;

[0095] (2) Each task can be executed by only one agent at any time, or not at all.

[0096] It can be seen from (Equation 9) that the possible solution scale of the Multi Agent system increases sharply with the increase of m and n.

[0097] A single-step planning mechanism is adopted, that is, each task is assigned to an agent only one task, and the task of each agent is dynamically adjusted according to the changes in the battlefield situation during the task allocation process. The set of strategies in the multi-agent system during the entire task allocation process is defined as a strategy set, which is expressed as:

[0098] Ω={D(t0),D(t1),D(t2),K,D(t e )} (Formula 11)

[0099] The strategy set reflects the task execution status of each agent during the entire task execution process and is the basis for evaluating the collaboration between agents.

[0100] 5. Evaluation function (Objective_Function)

[0101] The evaluation function refers to the objective function in task allocation, which needs to consider factors such as the flight distance of the drone, the benefit value of the attack target, the weapon consumption value, and the threat cost. For example, when the drone is required to complete the task in the shortest time, the objective function is set to the shortest flight distance; when the drone is required to consume the least in performing the task, the minimum consumption value is set as the goal. The above factors are coupled with each other, and a compromise needs to be made between the factors. Usually, the task allocation evaluation function is a collection of factors, and the weight of each factor is distinguished by weight.

[0102] The objective function is the benefit value obtained by all agents in the multi-agent system during the whole task allocation process. The task allocation aims to maximize the objective function value. The flight distance and task consumption value are introduced into the objective function through the time discount factor and task consumption discount factor.

[0103] The profit value of the Multi Agent system at time t in decision D(t) and state S(t) is defined as:

[0104]

[0105] where β t is the time discount factor, 0≤β t ≤1. β tThe smaller it is, the faster the benefit value of the Multi Agent system decreases over time. t = 1, the impact of time on the return value is not considered, t =0, the profit value is not considered. m t is the time consumed in the task allocation decision process and the time it takes for the agent to move to the target position after the decision is made. Generally, the decision time is short and can be ignored. The flight time is proportional to the distance. △t can be expressed as: △ m t=d t (m,n) / V m , where d t (m,n) is a at time t m Fly to T n Flight distance at encounter, V m for a m Rate.d t (m,n) is determined by the agent and task position, speed, and direction. t The consumption for executing tasks, including processing (Nn L ) tasks, the Agent's attack power consumption is n The communication cost and decision cost during task allocation are as follows: the communication cost is determined during task allocation and has nothing to do with the execution process; the decision cost is related to the possible solution scale and the allocation algorithm used. δ is the task consumption discount primer, β δ ≥0, β δ =0 means that task consumption is not considered. The communication cost only considers the communication cost when obtaining the agent status to complete the task assignment, and the communication cost between agents when executing the task is not considered.

[0106]

[0107] ξ is a penalty function. When there are n L When a task is not assigned,

[0108]

[0109] It represents the sum of the reward values ​​of unassigned tasks, η is the penalty function factor, η≥0. η=0 means that the impact of the penalty function on the benefit value is not considered. The larger η is, the heavier the penalty for unassigned tasks is, ensuring that the Multi Agent system completes as many tasks as possible.

[0110] In the collaborative task allocation model, both the agent and the task are in motion. When the agent and the task meet, the flight distance is not the straight-line distance between them. t The calculation diagram of (m,n) is as follows Figure 1 As shown:

[0111] Set the initial Agent position to (x a ,y a ), the absolute value of velocity is v a , direction is θ; the task position is (x T ,y T ),and The agent encounters the task at time t, and the encounter point satisfies the equation:

[0112]

[0113] Eliminating the variable θ, we get the equation:

[0114]

[0115] The condition for equation (16) to have a solution is:

[0116]

[0117] In order for the solution to be meaningful, it must also satisfy: t≥0.

[0118] It turns out that:

[0119]

[0120] t=max(t1,t2), when t≥0, t is a valid solution, when t<0, the equation has no solution.

[0121] From (Equation 18), we can get the heading and required flight time when the Agent performs the task, and the path length is:

[0122]

[0123] In summary, the revenue of the Multi Agent system at time t is:

[0124]

[0125] The total benefit Γ of the Multi Agent system during the entire task allocation process is:

[0126]

[0127] The complexity of the task allocation strategy increases with the increase of the task scale and the number of drones. The scale of possible solutions for multi-drone collaborative task allocation is determined by (Equation 9). Assuming the number of agents in the Multi Agent system is m and the number of tasks is n, the effect of the increase in the number of agents and tasks on the scale of possible solutions is as follows: Figure 2 As shown:

[0128] It is found through calculation that when m=n=20, the possible solution scale Num=1.73×10 21 It is extremely difficult to find the optimal strategy in such a large solution space. To reduce the scale of possible solutions, grouping agents and tasks in a Multi-Agent system is an effective method. For example, when the number of tasks is 20, the 20 agents are divided into 4 groups and then the tasks are assigned. At this time, the scale of possible solutions will become: Num = 4 × (3.13 × 10 19 )=1.25×10 20 , the possible solution scale is only 7.2% of that before grouping. The more groups there are, the more conducive it is to reduce the possible solution scale. Therefore, using a distributed model to group Agents is an effective way to reduce the complexity of the problem.

[0129] To simplify the model, a single-task type model is used, and each task is assigned only once to reduce the scale of possible solutions. To discuss the impact of task type on the scale of possible solutions, assume that when the number of agents is m and the number of tasks is n, N m The minimum possible solution size under this task type is:

[0130]

[0131] Each task is divided into N m tasks, the number of tasks to be assigned is nN m , each task has m ways of allocation, the possible solution scale of the Multi-Agent system is determined by (Equation 9), and the possible solution scale of the Multi-Agent system is greater than the value calculated by (Equation 22). The influence of the number of agents and tasks on the possible solution scale is analyzed by (Equation 22), as shown in Figure 3 As shown, where N m =1.

[0132] Task Type N m The impact on the size of possible solutions is as follows Figure 4 As shown:

[0133] Depend on Figure 3 It can be seen that N m The increase in will also cause the possible solution space to grow at an exponential rate, which is the same as the impact of the number of tasks and the number of agents on the possible solution space.

[0134] Step S2, optimizing the task allocation strategy of the collaborative task allocation model of step S1 based on the selection optimization method;

[0135] The scale of possible solutions is related to the number of candidate agents for each task. Reducing the number of candidate agents for each task can reduce the scale of possible solutions. Through the analysis of the optimal task allocation strategy, it is found that under the optimal task allocation strategy, each task is basically executed by the agents within a certain range around it. On the one hand, the shorter route between the task and the agent is conducive to reducing costs. On the other hand, the task execution delay and threat cost brought by the shorter route are smaller.

[0136] When using the selection optimization method to optimize the task allocation strategy, a circle with a certain radius is drawn with each task as the center, and the selection of agents within the circle where each task is located is adjusted to achieve strategy optimization. The selection mechanism ensures that the allocation strategy of the Multi-Agent system moves towards a better direction in each iteration. After a finite number of iterations, the Multi-Agent system will obtain the optimal allocation strategy.

[0137] As Figure 5 shown, the strategy optimization steps of the selection optimization method of the present invention are as follows:

[0138] It is assumed that there are M agents and N tasks in the Multi-Agent system.

[0139] Step S21: Taking the position where task T k is located as the center, draw a circle with a radius of r k . The radius of the circle is selected to meet the following conditions:

[0140] 1. There are at least n min agents in each circle;

[0141] 2. The radius of each circle is greater than r min .

[0142] Step S22: Generate the initial allocation strategy: Randomly select an agent for each task in the Multi-Agent system, and represent the decision variable at this time as D t (0), and calculate the benefit value:

[0143]

[0144] where k ∈ N, and satisfy: i ≠ j. When M < N, (N - M) tasks will not be assigned for execution.

[0145] Step S23: Randomly select a task Adjust this task to and satisfy: a k ≠ a′ k , and represent the allocation strategy at this time as D t (1).

[0146] Step S24: Policy rationality check: Determine whether there exist t i, j ∈ N within (1) such that k ≠ i. If so, adjust to to make D t satisfy (1): i ≠ j, i, j ∈ N. Denote the allocation policy at this time as D t (2), and the policy revenue value is:

[0147]

[0148] Step S25:: For D t (2), for the remaining tasks other than k and i x ∈ N, x ≠ k, i, sequentially select the Agent within the circle where it is located, and denote the newly adjusted policy as D t (3), for i ∈ N, make

[0149]

[0150] satisfy: v3 ≥ v2.

[0151] Step S26: Update the optimal policy of the Multi - Agent system. Define the optimal policy as v max . If v max < v3, then v max = v3.

[0152] Step S27: If v max meets the requirements of the Multi - Agent system, the optimization process ends; otherwise, go to Step S23.

[0153] The selection optimization method selects the optional Agents within a certain range as the optimization objects, reducing the scale of possible solutions and improving the optimization speed. The optimal selection algorithm ensures that the allocation policy continuously approaches the optimal allocation policy with the iterative process.

[0154] By using the selection optimization method and through a finite number of iterations, the Multi - Agent system will obtain the optimal allocation policy. At each moment, the number of tasks and the number of Agents in the Multi - Agent system are finite values, and the number of all optional allocation policies is a finite value. When r k is large enough so that the circle centered on each task can cover all the Agents in the Multi - Agent system, the rotation selection algorithm requires selecting the optimal policy in each selection process. This process ensures that after a finite number of iterations, the Multi - Agent system will obtain the optimal allocation policy.

[0155] Selecting an optimization algorithm can reduce the size of possible solutions. Assume that the number of agents in the Multi Agent system is m, a i , i=1,2,K,m, the number of tasks is T n , and T n ≤N. The number of agents distributed around each task is always x, and 1≤x≤m. Then the possible solution scale of the optimization method is:

[0156]

[0157] In the selection optimization method, the selection radius has an important influence on the optimal strategy obtained. When the selection radius is infinite, the optional agents for each task are all agents in the Multi-Agent system, which makes the possible solution scale of the selection optimization method the same as the strategy optimization based on the genetic algorithm. At this time, the selection optimization method can obtain the global optimal solution of the Multi-Agent system. When the selection radius is reduced, the possible solution scale of the Multi-Agent system is reduced, which may cause the optimal agent of some tasks in the Multi-Agent system to be outside the selection circle, and the Multi-Agent system will not obtain the optimal allocation strategy. Therefore, in the specific implementation process, the selection of the selection radius needs to balance the possible solution scale and strategy performance.

[0158] The described embodiments are only a part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

Claims

1. A method for dynamically allocating collaborative tasks among multiple drone swarms, characterized in that: The following steps are involved: Step S1, establishing a collaborative task allocation model; Based on the Multi-Agent system, each UAV is regarded as an agent, and tasks are assigned and executed by giving the agent autonomous capabilities; The collaborative task allocation model is described as consisting of five groups: {Time,Task,Agent,Policy_Set,Objective_Function}(Formula 1) Among them, Time is time, Task is task, Agent is drone, Policy_Set is policy set, and Objective_Function is evaluation function; Step S2, optimizing the task allocation strategy of the collaborative task allocation model of step S1 based on the selection optimization method; Draw a circle with a certain radius with each task as the center, and adjust the selection of agents in the circle for each task to achieve strategy optimization. The selection mechanism ensures that each iteration makes the allocation strategy of the Multi Agent system move in a better direction. After a limited number of iterations, the Multi Agent system will obtain the optimal allocation strategy. The task strategy optimization of the selection optimization method in step S2 is specifically as follows: Assume that there are M agents and N tasks in the Multi Agent system; Step S21: Task T k The position is the center of the circle, with r k Make a circle with radius; Step S22: generating an initial allocation strategy; For each task in the Multi-Agent system, an agent is randomly selected and the decision variable at this time is represented as D t (0), calculate the profit value: where k ∈ N and satisfies: When i ≠ j and M < N, there will be (N - M) tasks not assigned for execution; Step S23: Randomly select a task Adjust the task to And satisfy: a k ≠a′ k , the allocation strategy at this time is represented as D t (1); Step S24: Checking the rationality of the strategy: Determining D t (1) Whether i, j∈N exists in k≠i, if it exists, Adjust to Make D t (1) Satisfy: i≠j, i,j∈N, the allocation strategy at this time is represented as D t (2), the strategy return value is: Step S25: For D t (2) For tasks other than k and i x∈N,x≠k,i,select The agent in the circle is represented by the newly adjusted strategy D t (3), i∈N, so that Satisfies: v3 ≥ v2; Step S26: Update the optimal policy of the Multi Agent system, and define the optimal policy as v max , if v max < v3, then v max = v3; Step S27: If v max If the Multi Agent system requirements are met, the optimization process ends; otherwise, step S23 is performed.

2. A method for dynamically allocating collaborative tasks among multiple drone swarms according to claim 1, comprising the following steps: In step S21, the radius of the circle must satisfy the following requirement: there are at least n min Agents and the radius of each circle is greater than r m i n .

Images