Unmanned aerial vehicle task allocation method and device for adaptive matching mode switching

Abstract

The invention provides an unmanned aerial vehicle task allocation method and device for self-adaptive matching mode switching, and the switching of matching modes mainly comprises two parts: (1) switching of dynamic and static value calculation functions according to historical task conditions; and (2) matching and switching task types, analyzing task allocation types according to the tasks, and switching to a corresponding rapid task allocation algorithm. The invention provides a task allocation static revenue function and dynamic value function switching method which is used for constructing a matching value matrix. The invention provides a static revenue function definition method and a dynamic value function definition method which are used for determining the value of a task allocation scheme. And accurate matching of the task and the unmanned aerial vehicle under different conditions is realized. The invention provides a rule for task allocation algorithm selection under the condition that the number of unmanned aerial vehicles and the number of tasks are different, determines the allocation relationship between the unmanned aerial vehicles and the tasks based on a bipartite graph allocation algorithm and a consistent auction algorithm, and realizes rapid solution of an allocation scheme.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle (UAV) cluster combat task allocation, and relates to the design of static value functions, dynamic value functions and matching algorithm selection rules for UAV task allocation, and in particular to a method and device for UAV task allocation with adaptive matching mode switching. Background Art

[0002] With the development of swarm intelligence technology, the scale of drone swarms has expanded from a few to thousands.

[0003] When performing tasks in a highly dynamic combat space, there is a contradiction between solution time and matching accuracy.

[0004] Therefore, it is necessary to select a suitable matching mode to achieve accurate and fast matching. Through reasonable resource allocation, the combat efficiency can be maximized. Summary of the invention

[0005] The present invention provides a method and device for unmanned aerial vehicle task allocation with adaptive matching mode switching, which realizes adaptive switching of unmanned aerial vehicle task matching modes in a high dynamic combat environment.

[0006] A first aspect of the present invention provides a method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching, comprising:

[0007] When receiving the task set M, determine whether there is historical task execution data of drone group A performing the same type of tasks;

[0008] If so, the state value function is trained by reinforcement learning method according to the historical task execution data, and the dynamic benefit matrix of the drone group A executing the task set M is obtained according to the trained state value function. According to the dynamic benefit matrix, the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm is used to allocate the task set M to the drone group A;

[0009] If not, the static benefit function of the drone group A executing the task set M is calculated. According to the task type of each task in the task set M and the sufficiency of drone resources, the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm is used to obtain the allocation plan when the static benefit function takes the maximum value as the drone task allocation plan; the task types include: point tasks and surface tasks.

[0010] Optional, static return function

[0011] In the formula, R am The profit of drone a completing mission m, v importance is the task importance; J = {J 1 ,J2 ,...,J N}To complete the set of drone capability requirements, at least including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, and mobility capability; j is the capability weighting coefficient, satisfying a j represents the numerical expression of the capability j of drone a; m j represents the numerical expression of the ability j required to complete task m; θ is the ability verification factor. For a certain ability j, if it is required to meet a j ≥m j , UAV a can perform mission m, then when a j ≥m j When , θ=1, otherwise θ=0;

[0012] X am Indicates whether UAV a completes task m, which is a decision variable and is expressed as:

[0013] Optionally, according to the task type of each task in the task set M and the sufficiency of drone resources, a bipartite graph allocation algorithm, a consistency auction algorithm or a consistency alliance auction algorithm is used to obtain an allocation scheme when the static benefit function takes the maximum value as the drone task allocation scheme, including:

[0014] For the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution;

[0015] For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution;

[0016] For the task set M, when there are face tasks in the task set M, the consistency alliance auction algorithm is called to solve the task allocation plan.

[0017] Optionally, a reinforcement learning method is used to train the state value function based on historical task execution data, including:

[0018] Based on historical mission execution data, obtain the mission completion status of a certain UAV performing a certain mission in the time series, as well as the evaluation of the completion status, and establish a reinforcement learning problem model;

[0019] According to the reinforcement learning problem model, obtain the state value function;

[0020] Train the state-value function through reinforcement learning methods;

[0021] After the state value function training is completed, the matching value V of assigning a task m to a drone a can be calculated. πd (S am ), where S am It represents the state space composed of the corresponding attributes of drone a and task m. By traversing the task set M and drone group A, the dynamic benefit matrix of drone group A executing task set M can be obtained.

[0022] The reinforcement learning problem model includes: state space, action space, state transfer and immediate reward; the state space includes: current time segment information, drone location attribute information, payload attribute information, weapon attribute information and platform capability information, task spatial geographic information, task type information and task capability requirement information; platform capability information includes: reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability and maneuverability;

[0023] The action space includes: the drone is assigned a task and the drone is not assigned a task in the current time slice;

[0024] Instant Rewards γ is the discount factor, which takes a decimal value between 0 and 1; T is the entire time series after discretization T = {t 1 ,...,t all}, t∈T is any moment in the time series, is a discrete quantity, R evaluate is the numerical evaluation of a certain UAV completing a certain task in the allocation scheme, T execute It is the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task; the numerical evaluation includes: task execution time, resource utilization, energy consumption, damage effect, and task priority.

[0025] Optionally, according to the dynamic profit matrix, a bipartite graph allocation algorithm, a consistency auction algorithm or a consistency alliance auction algorithm is used to allocate the task set M to the drone group A, including:

[0026] According to the dynamic profit matrix, for the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution;

[0027] For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution;

[0028] For the task set M, when there are face tasks in the task set M, the consistent alliance auction algorithm is called to solve the task allocation solution;

[0029] Reference weight and consistency of edges in bipartite graph allocation method Drone value w in alliance auction algorithm am =R am +β(V πd (S am )); where R am is the static return function, and β is the weight factor.

[0030] A second aspect of the present invention provides a UAV task allocation device with adaptive matching mode switching, comprising:

[0031] A judgment module is used to determine whether there is historical task execution data of the drone group A performing the same type of task when receiving the task set M;

[0032] The first allocation module is used to train the state value function by using the reinforcement learning method according to the historical task execution data when there is historical task execution data of the drone group A performing the same type of tasks, obtain the dynamic benefit matrix of the drone group A performing the task set M according to the trained state value function, and allocate the task set M to the drone group A according to the dynamic benefit matrix by using the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm;

[0033] The second allocation module is used to calculate the static benefit function of drone group A executing task set M when there is no historical task execution data of drone group A executing the same type of tasks. According to the task type of each task in task set M and the sufficiency of drone resources, a bipartite graph allocation algorithm, a consistency auction algorithm or a consistency alliance auction algorithm is used to obtain the allocation plan when the static benefit function takes the maximum value as the drone task allocation plan; task types include: point tasks and surface tasks.

[0034] Optional, static return function

[0035] In the formula, R am The profit of drone a completing mission m, v importance is the task importance; J = {J 1 ,J 2 ,...,J N}To complete the set of drone capability requirements, at least including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, and mobility capability; j is the capability weighting coefficient, satisfying a j represents the numerical expression of the capability j of drone a; m j represents the numerical expression of the ability j required to complete task m; θ is the ability verification factor. For a certain ability j, if it is required to meet a j ≥mj , UAV a can perform mission m, then when a j ≥m j When , θ=1, otherwise θ=0;

[0036] X am Indicates whether UAV a completes task m, which is a decision variable and is expressed as:

[0037] Optionally, the second allocation module is specifically used for:

[0038] For the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution;

[0039] For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution;

[0040] For the task set M, when there are face tasks in the task set M, the consistency alliance auction algorithm is called to solve the task allocation plan.

[0041] Optionally, the first allocation module is specifically used to:

[0042] Based on historical mission execution data, obtain the mission completion status of a certain UAV performing a certain mission in the time series, as well as the evaluation of the completion status, and establish a reinforcement learning problem model;

[0043] According to the reinforcement learning problem model, obtain the state value function;

[0044] Train the state-value function through reinforcement learning methods;

[0045] After the state value function training is completed, the matching value V of assigning a task m to a drone a can be calculated. πd (S am ), where S am It represents the state space composed of the corresponding attributes of drone a and task m. By traversing the task set M and drone group A, the dynamic benefit matrix of drone group A executing task set M can be obtained.

[0046] The reinforcement learning problem model includes: state space, action space, state transfer and immediate reward; the state space includes: current time segment information, drone location attribute information, payload attribute information, weapon attribute information and platform capability information, task spatial geographic information, task type information and task capability requirement information; platform capability information includes: reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability and maneuverability;

[0047] The action space includes: the drone is assigned a task and the drone is not assigned a task in the current time slice;

[0048] Instant Rewards γ is the discount factor, which takes a decimal value between 0 and 1; T is the entire time series after discretization T = {t 1 ,...,t all}, t∈T is any moment in the time series, is a discrete quantity, R evaluate is the numerical evaluation of a certain UAV completing a certain task in the allocation scheme, T execute It is the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task; the numerical evaluation includes: task execution time, resource utilization, energy consumption, damage effect, and task priority.

[0049] Optionally, the first allocation module is specifically used to:

[0050] According to the dynamic profit matrix, for the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution;

[0051] For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution;

[0052] For the task set M, when there are face tasks in the task set M, the consistent alliance auction algorithm is called to solve the task allocation solution;

[0053] Reference weight and consistency of edges in bipartite graph allocation method Drone value w in alliance auction algorithm am =R am +β(V πd (S am )); where R am is the static return function, and β is the weight factor.

[0054] The present invention proposes a method and device for unmanned aerial vehicle task allocation with adaptive matching mode switching. The switching of matching modes mainly includes two parts: (1) switching of dynamic and static value calculation functions according to historical task conditions; (2) switching of task type matching, switching to the corresponding fast task allocation algorithm according to the task analysis task allocation type. The present invention provides a method for switching between a static benefit function and a dynamic value function for task allocation, which is used to construct a matching value matrix. A static benefit function definition method and a dynamic value function definition method are provided for determining the value of a task allocation scheme. Accurate matching of tasks and unmanned aerial vehicles under different circumstances is achieved. The present invention provides rules for selecting a task allocation algorithm when the number of unmanned aerial vehicles and the number of tasks are different, and determines the allocation relationship between unmanned aerial vehicles and tasks based on a bipartite graph allocation algorithm and a consistency auction algorithm, thereby achieving a rapid solution to the allocation scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the construction of the matching value matrix. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] The features and illustrative embodiments of various aspects of the present invention will be described in detail below. In the detailed description below, many specific details are proposed to provide a comprehensive understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention. The present invention is by no means limited to any specific settings and methods proposed below, but covers any improvements, replacements and modifications of structures, methods, devices without departing from the spirit of the present invention. In the accompanying drawings and the following description, known structures and technologies are not shown to avoid unnecessary ambiguity in the present invention.

[0058] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other, and the embodiments can refer to and quote each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0059] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0060] The task allocation method for matching mode switching proposed in the present invention adjusts the resource matching mode used in different task situations according to the task type and battlefield resource distribution, so as to realize the rapid matching of resources and task execution units. In the case of no historical task completion data, the static benefit function is switched to, and different optimization algorithms are selected to solve the matching scheme in the corresponding situations of different resources and tasks; in the case of certain historical task completion data, the dynamic value function is learned according to the historical task completion. After the dynamic value matrix is ​​constructed, different optimization algorithms are selected to solve the matching scheme in the corresponding situations of different resources and tasks. When the bipartite graph solution algorithm is used for solution, the edge weight calculation method of the existing binary allocation algorithm is improved.

[0061] The present invention mainly includes the following main technical points:

[0062] (1) The matching value matrix between execution units and tasks is generated by combining the static benefit function and the dynamic reward function.

[0063] The task matching value matrix describes the degree of matching between tasks and task-performing drones. The task matching value matrix comprehensively considers the characteristics of various task types and the properties of available drones, such as the current position of the drone, the maximum speed, the endurance of the drone, and the relative position to the task. Key indicators are selected to measure the effectiveness of the matching scheduling scheme, such as task execution time, resource utilization, energy consumption, damage effect, task priority, etc. These indicators are converted into a comprehensive evaluation function through mathematical models to quantify the comprehensive effectiveness of different schemes, thereby supporting the intelligent matching algorithm to evaluate and compare various matching scheduling schemes. According to the task characteristics and task priorities, an objective reference for the quantification of effectiveness is provided for matching scheduling in different scenarios to optimize the decision-making of resource allocation and task scheduling.

[0064] (2) Allocate tasks based on the established matching value matrix.

[0065] For the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution;

[0066] For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution;

[0067] For the task set M, when there are face tasks in the task set M, the consistency alliance auction algorithm is called to solve the task allocation plan.

[0068] Exemplarily, the static benefit function definition method includes the following contents:

[0069] Assuming the number of drones is A, the mission set is M, the static profit function is:

[0070]

[0071] In the formula, R am Static value function for drone a to complete task m:

[0072]

[0073] In the formula, v importance is the task importance; J = {J 1 ,J 2 ,...,J N}To complete the set of drone capability requirements, including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, mobility capability, etc.;λ j is the capability weighting coefficient, satisfying a j represents the numerical expression of the capability j of drone a; m j represents the numerical expression of the ability j required to complete task m; θ is the ability verification factor. For a certain ability j, if it is required to meet a j ≥m j , UAV a can perform mission m, then when a j ≥m j When , θ=1, otherwise θ=0;

[0074] X am Indicates whether UAV a completes task m, which is a decision variable and is expressed as:

[0075]

[0076] Exemplarily, the dynamic value function definition method includes the following:

[0077] For the case of tasks with a partial history of successful execution, the task allocation problem is defined as a sequential decision problem, that is, round orders are allocated in a longer time period T, and the goal is to maximize the total revenue within this time period T.

[0078] The drone task allocation problem is modeled as an MDP (Markov Decision Process) problem, which contains four basic elements: state space, action space, reward function and state transition matrix. Solving the MDP problem is to determine which task a drone should be assigned to in a certain state to maximize the benefit.

[0079] In the UAV task allocation problem, the MDP is defined as follows:

[0080] 1) State space: Numerical description of the UAV state and mission state. The Cartesian product of the UAV state and mission state is the set of all states S. The state includes: 1) current time segment information; 2) UAV location attribute information, payload attribute information, weapon attribute information and platform capability information (including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, maneuverability, etc.); 3) task spatial geographic information, task type information and task capability requirement information.

[0081] 2) Action space: There are two behaviors in the task allocation problem. The first is that drone a is assigned a task m, then the drone performs the task and completes the task after a period of time. The second behavior is that in this time slice, drone a is not assigned any task, it will idle for a time slice and enter the next round of task allocation.

[0082] 3) The reward function expression is as follows:

[0083]

[0084] Where γ is the discount factor, which takes a decimal value between 0 and 1; T is the entire time series after discretization T = {t 1 ,...,t all}, t∈T is any moment in the time series, is a discrete quantity, Δt is the time discretization threshold; R evaluate It is the numerical evaluation of a certain UAV completing a certain task in the allocation plan (including task execution time, resource utilization, energy consumption, damage effect, and task priority). execute is the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task.

[0085] 4) State transfer: State transfer occurs along with behavior. After UAV a performs mission m, the position state of the UAV itself changes and the platform capabilities change.

[0086] 5) Solve the dynamic value function V πd (s), a multi-layer neural network is used to represent the function, and the function parameters are updated using an update algorithm based on TD (Temporal-Dierence). The update formula is as follows:

[0087]

[0088] Among them, s' is the next state, T executeis the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task. For unassigned actions, T execute =1, and R γ =0.

[0089] Exemplarily, the construction and solution of the matching value matrix includes the following:

[0090] The task allocation problem assumes that there is an existing drone group A that needs to complete a set of tasks M, W. am It represents the fitness edge weight of UAV a for task m. Task allocation needs to determine the allocation relationship between UAVs and tasks, so as to maximize the total fitness of each round of task allocation results.

[0091] Figure 1 Schematic diagram of the construction of the matching value matrix.

[0092] The lines between drones and tasks represent the appropriate values ​​of different drones performing different tasks. By calculating the static value benefits and dynamic value benefits, the matching value matrix can be obtained:

[0093]

[0094] The reference weight and bid value w of the modified edge am =R am +β(V πd (S am )); where R am is the static return function, and β is the weight factor.

[0095] The above description is only a further embodiment 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 solutions and concepts of the present invention within the scope disclosed by the present invention, which belong to the protection scope of the present invention.

Claims

1. A method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching, characterized in that: include: When receiving the task set M, determine whether there is historical task execution data of drone group A performing the same type of tasks; If so, the state value function is trained by reinforcement learning method according to the historical task execution data, and the dynamic benefit matrix of the drone group A executing the task set M is obtained according to the trained state value function. According to the dynamic benefit matrix, the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm is used to allocate the task set M to the drone group A; If not, the static benefit function of the drone group A executing the task set M is calculated. According to the task type of each task in the task set M and the sufficiency of drone resources, the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm is used to obtain the allocation plan when the static benefit function takes the maximum value as the drone task allocation plan; Task types include: point tasks and area tasks.

2. The method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching according to claim 1, characterized in that: Static profit function In the formula, R am The profit of drone a completing mission m, v importance is the task importance; J = {J1, J2, ..., J N }To complete the set of drone capability requirements, at least including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, and mobility capability;λ j is the capability weighting coefficient, satisfying a j represents the numerical expression of the capability j of drone a; m j represents the numerical expression of the ability j required to complete task m; θ is the ability verification factor. For a certain ability j, if it is required to meet a j ≥m j , UAV a can perform mission m, then when a j ≥m j When , θ=1, otherwise θ=0; X am Indicates whether UAV a completes task m, which is a decision variable and is expressed as:

3. The method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching according to claim 1, characterized in that: According to the task type of each task in the task set M and the sufficiency of UAV resources, a bipartite graph allocation algorithm, a consistent auction algorithm or a consistent alliance auction algorithm is used to obtain an allocation scheme when the static profit function takes the maximum value as the UAV task allocation scheme, including: For the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution; For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution; For the task set M, when there are face tasks in the task set M, the consistency alliance auction algorithm is called to solve the task allocation plan.

4. The method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching according to claim 1, characterized in that: The state value function is trained using reinforcement learning methods based on historical task execution data, including: Based on historical mission execution data, obtain the mission completion status of a certain UAV performing a certain mission in the time series, as well as the evaluation of the completion status, and establish a reinforcement learning problem model; According to the reinforcement learning problem model, obtain the state value function; Train the state-value function through reinforcement learning methods; After the state value function training is completed, the matching value V of assigning a task m to a drone a can be calculated. πd (S am ), where S am It represents the state space composed of the corresponding attributes of drone a and task m. By traversing the task set M and drone group A, the dynamic benefit matrix of drone group A executing task set M can be obtained. The reinforcement learning problem model includes: state space, action space, state transfer and immediate reward; the state space includes: current time segment information, drone location attribute information, payload attribute information, weapon attribute information and platform capability information, task spatial geographic information, task type information and task capability requirement information; platform capability information includes: reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability and maneuverability; The action space includes: the drone is assigned a task and the drone is not assigned a task in the current time slice; Instant Rewards γ is the discount factor, which takes a decimal value between 0 and 1; T is the entire time series after discretization T = {t1,...,t all }, t∈T is any moment in the time series, which is a discrete quantity, R evaluate is the numerical evaluation of a certain UAV completing a certain task in the allocation scheme, T execute It is the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task; the numerical evaluation includes: task execution time, resource utilization, energy consumption, damage effect, and task priority.

5. The method for assigning unmanned aerial vehicle tasks with adaptive matching mode switching according to claim 4, characterized in that: According to the dynamic profit matrix, the task set M is allocated to the drone group A using the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm, including: According to the dynamic profit matrix, for the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution; For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution; For the task set M, when there are face tasks in the task set M, the consistent alliance auction algorithm is called to solve the task allocation solution; Reference weight and consistency of edges in bipartite graph allocation method Drone value w in alliance auction algorithm am =R am +α(V πd (S am )); where R am is the static return function, and α is the weight factor.

6. A UAV task allocation device with adaptive matching mode switching, characterized in that: include: A judgment module is used to determine whether there is historical task execution data of the drone group A performing the same type of task when receiving the task set M; The first allocation module is used to train the state value function by using the reinforcement learning method according to the historical task execution data when there is historical task execution data of the drone group A performing the same type of tasks, obtain the dynamic benefit matrix of the drone group A performing the task set M according to the trained state value function, and allocate the task set M to the drone group A according to the dynamic benefit matrix by using the bipartite graph allocation algorithm, the consistency auction algorithm or the consistency alliance auction algorithm; The second allocation module is used to calculate the static benefit function of the drone group A executing the task set M when there is no historical task execution data of the drone group A executing the same type of tasks, and adopt a bipartite graph allocation algorithm, a consistency auction algorithm or a consistency alliance auction algorithm according to the task type of each task in the task set M and the sufficiency of drone resources, to obtain the allocation plan when the static benefit function takes the maximum value as the drone task allocation plan; Task types include: point tasks and area tasks.

7. The UAV task allocation device with adaptive matching mode switching according to claim 6 is characterized in that: Static profit function In the formula, R am The profit of drone a completing mission m, v importance is the task importance; J = {J1, J2, ..., J N }To complete the set of drone capability requirements, at least including reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability, and mobility capability;λ j is the capability weighting coefficient, satisfying a j represents the numerical expression of the capability j of drone a; m j represents the numerical expression of the ability j required to complete task m; θ is the ability verification factor. For a certain ability j, if it is required to meet a j ≥m j , UAV a can perform mission m, then when a j ≥m j When , θ=1, otherwise θ=0; X am Indicates whether UAV a completes task m, which is a decision variable and is expressed as:

8. The UAV task allocation device with adaptive matching mode switching according to claim 6 is characterized in that: The second allocation module is specifically used for: For the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution; For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution; For the task set M, when there are face tasks in the task set M, the consistency alliance auction algorithm is called to solve the task allocation plan.

9. The UAV task allocation device with adaptive matching mode switching according to claim 6 is characterized in that: The first allocation module is specifically used for: Based on historical mission execution data, obtain the mission completion status of a certain UAV performing a certain mission in the time series, as well as the evaluation of the completion status, and establish a reinforcement learning problem model; According to the reinforcement learning problem model, obtain the state value function; Train the state-value function through reinforcement learning methods; After the state value function training is completed, the matching value V of assigning a task m to a drone a can be calculated. πd (S am ), where S am It represents the state space composed of the corresponding attributes of drone a and task m. By traversing the task set M and drone group A, the dynamic benefit matrix of drone group A executing task set M can be obtained. The reinforcement learning problem model includes: state space, action space, state transfer and immediate reward; the state space includes: current time segment information, drone location attribute information, payload attribute information, weapon attribute information and platform capability information, task spatial geographic information, task type information and task capability requirement information; platform capability information includes: reconnaissance capability, attack capability, endurance capability, generation capability, electronic jamming capability and maneuverability; The action space includes: the drone is assigned a task and the drone is not assigned a task in the current time slice; Instant Rewards γ is the discount factor, which takes a decimal value between 0 and 1; T is the entire time series after discretization T = {t1,...,t all }, t∈T is any moment in the time series, which is a discrete quantity, R evaluate is the numerical evaluation of a certain UAV completing a certain task in the allocation scheme, T execute It is the execution time of the task, including the cruising time of the UAV from the current position to the task position, and the time to complete the task; the numerical evaluation includes: task execution time, resource utilization, energy consumption, damage effect, and task priority.

10. The UAV task allocation device with adaptive matching mode switching according to claim 9, characterized in that: The first allocation module is specifically used for: According to the dynamic profit matrix, for the task set M, when there are only point tasks in the task set M and the number of drones is greater than the number of tasks, the bipartite graph allocation algorithm is called to solve the task allocation solution; For the task set M, when there are only point tasks in the task set M and the number of drones is less than the number of tasks, the consistency auction algorithm is called to solve the task allocation solution; For the task set M, when there are face tasks in the task set M, the consistent alliance auction algorithm is called to solve the task allocation solution; Reference weight and consistency of edges in bipartite graph allocation method Drone value w in alliance auction algorithm am =R am +β(V πd (S am )); where R am is the static return function, and β is the weight factor.

Images