Intelligent target task allocation method for UAV swarm communication and decision-making based on link dynamics

Abstract

The present invention proposes an intelligent target task allocation method for UAV cluster communication and decision-making based on link dynamics to cope with the challenges brought by link dynamics. As the UAV network structure becomes increasingly complex, the instability of the link affects the interaction stability between intelligent agents. This method migrates tasks affected by dynamic factors to other intelligent agents through grouping and layering strategies, reduces the number of task receptions, reduces the risk of dynamic factor propagation, and completes the target task allocation work better and more intelligently. The specific operations include modeling drones as intelligent agents, modeling drone groups of different levels as network layers, grouping intelligent agent tasks into three categories, and selecting the migrated intelligent agent based on the risk and cost of intelligent agent acceptance and task completion. Compared with the traditional centralized task migration method, the present invention effectively reduces the impact of dynamic factors and improves the task completion rate.

Description

Technical Field

[0001] The present invention relates to the field of autonomous collaboration of unmanned aerial vehicle (UAV) cluster systems in complex environments, and in particular to an intelligent target task allocation method for UAV cluster communication and decision-making based on link dynamics. Background Art

[0002] In recent years, with the continuous development and breakthroughs in theories and technologies related to intelligent systems and complex systems, perception and judgment, distributed collaboration, artificial intelligence, and algorithmic warfare, intelligent systems have demonstrated characteristics such as unmanned operation, swarming, and autonomy. As a major form of future swarm intelligent systems, drone swarms can achieve single-platform behavioral decision-making and multi-platform task collaboration, exhibiting cluster emergent properties and showing great application prospects. The mission environment of drone swarms is characterized by high elasticity, rapid situational changes, incomplete sensor information, and unstable communication structures. In this complex and dynamic environment, how drone swarms can implement explainable intelligent decision-making and reasoning based on uncertain situational information, complete intelligent target task allocation, and determine efficient and reliable task collaboration methods is crucial to ensuring swarm safety and improving operational effectiveness. Drone swarm formations require real-time decision-making and coordinated actions. This means that during actual flight, individual drones must continuously collect, process, and transmit information, making intelligent decisions based on this information to achieve coordinated actions. In large-scale drone swarms, drones may be connected by various physical and information network links, which are highly complex and diverse. Therefore, a problem with a single drone in the swarm network can cause the entire drone swarm system to collapse.

[0003] With the rapid development of drone networks, their network structure has become increasingly complex. The control, information, and service flows between drones are increasingly intertwined. To address this complex interaction, traditional specialized protocols, hardware, and operating systems are being replaced by more general-purpose Internet protocols and hardware. This has led to increased vulnerability of links to external factors, highlighting the dynamic nature of links during target task allocation, impacting network stability. Link dynamics, which may be induced during task migration, can gradually propagate within and across network layers. This necessitates that task migration strategies simultaneously consider multiple objectives, including agent failure scenarios, the risk of link dynamics propagation, the risk of overload at both agent and network layers, and the cost of task completion. The coupled impact of these objectives significantly increases the complexity of the problem. Traditional task migration methods that consider dynamic factors often overlook the potential for dynamic factors to propagate during task migration and exert coupled effects on agents and links. Furthermore, these methods typically focus only on task migration within a single network and rarely address the heterogeneous propagation capabilities of link dynamics across different network layers within multiple networks. Centralized solution methods have difficulty handling the huge decision space brought about by the coupling of link dynamics and agent dynamics in multiple network scenarios.

[0004] Large-scale drone swarms typically employ a hybrid decision-making structure, a complex, nonlinear, and multi-layered network structure. This complexity strengthens the relationships between various links within a swarm, but also increases their vulnerability to changes in the external environment. Considering the link dynamics of large-scale drone swarms, the collaborative interaction links between agents exhibit dynamic changes, making it difficult for the network to effectively cope with load imbalances at the agent and network levels, which in turn impacts intelligent task allocation. In this scenario, the task allocation process can trigger the cascading propagation of link dynamics, even leading to cascading failures and significant losses to the swarm network. Therefore, it is crucial to develop a task migration model for intelligent task allocation. Factors contributing to link dynamics can propagate within and between network layers during task migration, requiring task migration strategies to simultaneously consider multiple objectives. These objectives include agent failure scenarios, the risk of link dynamics propagating with task migration, the risk of overload at the agent and network levels, and the cost of task completion. Therefore, task migration strategies must minimize the risk of link dynamics propagation, reduce system failure rates, and mitigate the risk of overload at the agent and network layers while ensuring task completion rates. Furthermore, they must consider the cost of task migration to achieve a balance and optimization of multiple objectives. Task migration not only optimizes task execution within large-scale drone swarms but also mitigates the impact of the cascading propagation of link dynamics. Summary of the Invention

[0005] Technical Problem: The present invention proposes a novel task grouping and hierarchical migration algorithm in the context of large-scale drone swarms. By reducing the number of tasks agents must accept, this algorithm mitigates the risk of dynamic factors propagating across multiple industrial networks, caused by the cascading coupling of link dynamics with agent dynamics. The algorithm first models each drone in a large-scale drone swarm as an agent, and models the drone network at different levels of the swarm as network layers. Because the dynamics of each link are subject to constraints imposed by link dynamics, the impact of tasks within the same task group on link dynamics is correspondingly attenuated, reducing the probability of link dynamics being triggered by grouped task migration compared to migrating these tasks one at a time. To reduce the cost of task completion, increase the benefits of task allocation, and mitigate the risk of dynamic factors propagating with task migration, tasks are grouped and assigned to corresponding agents in batches. This multi-type grouping and migration approach effectively addresses the increased dimensionality of the solution to the problem of link dynamics and agent overload risk coupling, caused by the propagation of link dynamics across network layers and agents during the task allocation process in large-scale drone swarms.

[0006] Technical solution: In the process of task execution and deployment of large-scale drone clusters, the collaborative interaction links between intelligent agents will be affected by different external factors at different times. Dynamic changes will occur in the process of task deployment and collaborative interaction between intelligent agents, making it possible for the intelligent agents to no longer be able to interact through the link. This will result in the cluster network being unable to effectively perform intelligent target task allocation for the load imbalance between intelligent agents and the network layer. It is necessary to design a hierarchical and grouped task migration strategy to reduce the number of times an intelligent agent receives tasks by grouping tasks, thereby reducing the risk of dynamic factors propagating in the drone network due to the dynamic nature of task migration. Migrate the task groups that need to be migrated due to dynamic factors to other intelligent agents. Under the premise that the most tasks in the system are executed, the cost of task execution is reduced, while avoiding the cascading propagation of the dynamic nature of the link in the cluster network during the task migration process. The main technical solutions of this task migration method are as follows:

[0007] In large-scale drone swarm scenarios, dynamic factors may affect multiple agents on the same network layer at the same time. Before migration, it is necessary to consider grouping tasks on different agents on the same network layer. First, it is necessary to group tasks according to g m Transfer to agent a i Cost Agent a i The cost of migrating its original tasks and Implementation Task Force m The cost of the task and Task Group g m Transferred to agent a i The proportion of tasks that can be completed To comprehensively consider the benefits of task group migration Ben (·). Since the smaller the task group, the higher the possibility of being merged, a priority queue will be constructed according to the size of the task group, and then the two task groups g will be judged in turn with a greedy mindset. m and g n Is the task migration benefit after the merger better than the benefit before the merger? If the profit after merging is higher, the two task groups are merged. The two task groups are iteratively judged whether they can be merged to achieve the goal of grouping tasks of different agents in the same network layer.

[0008] In large-scale drone clusters, links in different network layers have different abilities to propagate link dynamic factors. By migrating across network layers, tasks can be migrated to network layers with poor link dynamic factor propagation capabilities, thereby reducing the propagation of dynamic factors in the system. In contrast, the method of grouping tasks within the same network layer alone makes it difficult to fully utilize the intelligent entities with poor link dynamic factor propagation capabilities and light loads on the network layer. Therefore, it is necessary to design a cross-network layer grouping algorithm. First, it is necessary to comprehensively consider the load conditions, dynamic conditions, and the ability to propagate dynamic factors of the network layer to evaluate the risk level of each network layer. And find the network layer with the greatest risk. and the network layer with the least risk Then, all possible task grouping and merging methods in the two network layers are traversed, and a determination is made as to whether the resulting benefit from merging the two task groups is greater than the pre-merger benefit. If the resulting benefit is greater, the task group in the highest-risk network layer is merged into the corresponding task group in the lower-risk network layer; otherwise, no merging occurs. A determination is then made in turn as to whether tasks can be merged between the highest- and lowest-risk network layers in the network layers to be measured. After this determination is made, the two network layers determined are removed from the set to be measured, and the next round of determination is continued. This continues until all network layers have been determined, thus completing the task grouping of different agents across network layers.

[0009] After completing the task grouping within the same network layer and across network layers, the task migration can be performed. First, traverse all task groups and construct the agent set A to which the task group is migrated. m , and then we need to combine the number of tasks expected to be executed in the task group |g m |·R [l] and the number of tasks performed on the agent to which it is expected to be transferred |Q m|, when |g m |·R [l] >|Q m |, the task migration algorithm is executed. Otherwise, the tasks in the task group are abandoned to reduce the cost of completing the task. Using the previous task grouping as a foundation, the task migration algorithm is executed, ultimately completing task migration, mitigating the impact of dynamic factors and ensuring task completion in large-scale drone swarms.

[0010] The specific technical solutions are as follows:

[0011] A method for intelligent target task allocation for UAV swarm communication and decision-making targeting link dynamics is characterized by the following specific steps: first, each UAV in the UAV swarm is modeled as an intelligent agent. When the swarm is too large, UAVs at different levels are modeled as different network layers. Second, in order to reduce the frequency of agents accepting tasks and reduce the risk of dynamic factor propagation due to task migration, the tasks of the agents are grouped into three categories: first, task grouping within the same agent; second, task grouping on different agents but on the same network layer; and third, task grouping across different network layers and different agents. Finally, based on the task grouping, the most suitable agent is selected for task group migration according to the risks and costs of the agents accepting tasks and completing tasks.

[0012] As a further improvement of the present invention, each task t in the task set T in the UAV cluster k Can use a binary <size(t k ),pos(t k )> to indicate; among them, size(t k ) indicates the execution of task t k The cost required for the agent to perform tasks with different loads is different; pos(t k ) represents the agent that performs the task. At the same time, a task can only be performed by one agent, and the agents in the drone cluster represent drones that can independently complete a given task. Each agent a i Use triples Indicates that v i Represents agent a i The size of the task that can be performed at unit cost; Q i Represents agent a i The task queue being executed, i.e. Q i ={t1,t2,…,t k}; Represents agent a i The set of network layers to which it belongs. In a large-scale drone swarm, each intelligent agent can belong to a different network layer.

[0013] As a further improvement of the present invention, a variable d with a value between 0 and 1 is used. ij Represents agent a i With agent a j The probability of link dynamic factors appearing between ij The larger the value of , the more likely it is to generate link dynamics during task migration. ij =1, it means that agent a i With agent a j Link dynamics will inevitably occur when migrating tasks between them; using vector D i Represents agent a i The dynamics of all connected links, i.e.

[0014] As a further improvement of the present invention, since link dynamics will be propagated to other agents during the task migration process in a large-scale UAV cluster, the propagation of this link dynamics is expressed as:

[0015]

[0016] in Then it means task t k From agent a i Transfer to agent a j ,otherwise R [l] R represents the propagation capability level of link dynamic factors along with task migration through link type l. [l] The smaller the value, the more intelligent the agent a i During the task migration process, the lower the probability of link dynamics propagation when receiving tasks through link type l; Spread(·) is the link dynamics propagation probability function, with a value of 0-1; this formula shows that the propagation process of link dynamics factors is related to the presence of situations in the surrounding agents that may cause link dynamics factors, whether tasks are migrated between agents, and the cost of task migration; Agent a i The more dynamic the links connected to the surrounding agents are, the more likely agent a is to i The surrounding links are affected by the spread of dynamic factors and the probability of link dynamics is higher; and the spread of this link dynamic factor between agents is triggered by the task dynamic process of task migration. If agent a i and agent a j If there is no task migration between them, then this dynamic factor will not be propagated; the cost c of the migration task is ijThe process of seeking quotient indicates that the higher the cost of task transfer between agents, the more difficult it is to propagate the dynamic factors caused by task transfer.

[0017] As a further improvement of the present invention, in the scenario where links in a large-scale UAV cluster are dynamic, a task grouping allocation or migration method is adopted, so that the agent can migrate or receive a group of tasks through the link at a time; the dynamics brought by each group of tasks can be attenuated according to the grouping method w; the tasks are migrated to the agent a according to the grouping method W. i For a j The influence function of the probability of dynamicity of connected links is defined as follows:

[0018]

[0019] Where 0<ψ(·)<1, the value of the function damp increases with the number of tasks in the task group. The increase of monotonically decreases, and transfers to agent a i Task Force The function ψ is defined as Where σ is a given attenuation factor; W = {w1, w2, ..., w |w|} represents the set of all task grouping methods, each task grouping method w i Each corresponds to the corresponding task group

[0020] As a further improvement of the present invention, in the UAV network with dynamic links, in addition to the tasks that are abandoned on the dynamic agents and the tasks that are not executed due to the risk of overload, there will also be situations where the link dynamics caused by task migration leads to the inability to complete task migration; therefore, the task completion ratio in the system after executing the task migration strategy π is Redefine as:

[0021]

[0022] in Indicates the task that was abandoned on the agent, Indicates that the task cannot be completed due to the dynamic nature of the link. Indicates tasks that were not executed due to overload risk.

[0023] As a further improvement of the present invention, the number of times an agent receives tasks is reduced by grouping tasks, thereby reducing the risk of dynamic factors spreading in the drone network due to the dynamic nature of task migration. Before performing grouped task migration, tasks need to be grouped in a hierarchical manner. In the task grouping stage in a large-scale drone cluster scenario, tasks are divided into the following three categories according to the distribution of agents where the tasks are located: (1) tasks from agents with the same dynamic factors; (2) tasks from agents with different dynamic factors at the same network layer; (3) tasks from agents with different dynamic factors at different network layers. For these three situations, the task grouping step is also divided into the following three stages: (1) task grouping stage on the same agent; (2) task grouping stage on different agents at the same network layer; (3) task grouping stage on different agents across network layers.

[0024] As a further improvement of the present invention, the task grouping stage on the same agent is specifically as follows: in the scenario of link dynamics, task migration is triggered by the emergence of factors on the agent that can cause link dynamics, and such factors will also cause the dynamics of the agent, that is, dynamic factors; tasks on the agent with dynamic factors all need to be migrated, and these tasks have the same impact on the agent and link dynamics after executing the task migration process, and have natural isomorphism; therefore, the tasks performed on each agent with dynamic factors are first divided into a separate task group; these task groups are represented as G = {g1, g2, ..., g |G|};

[0025] The task grouping stage of different agents on the same network layer is specifically as follows: considering that in large-scale drone cluster scenarios, dynamic factors will affect multiple agents on the same network layer at the same time, it is necessary to group the tasks of different agents on the same network layer; first, the agents on each network layer are divided into affected sets according to whether they are affected by dynamic factors. and normal run collection Task migration in different situations will bring different benefits to the UAV cluster. m ∈G migrates to agent a in the normal operation set i The resulting task migration benefit is defined as follows:

[0026]

[0027] It contains task group g m Transfer to agent a i Cost Contains agent a i The cost of migrating one's own tasks in order to accept task grouping and Implementation Task Forcem The cost of the task and Task Group g m Transferred to agent a i The proportion of tasks completed later Based on this, the maximum benefit obtained by migrating the task groups is determined. Through continuous iterative calculations, the benefits obtained by merging different task groups into one task group are compared with the benefits obtained by not merging the task groups, and the final task grouping is selected.

[0028] The task grouping stage on different agents across the network layer is specifically as follows: considering that the links of different network layers have different abilities to propagate link dynamic factors along with task migration; for the network layer with poor link dynamic factor propagation ability, more tasks are migrated therein, so as to reduce the propagation of dynamic factors in the system; in order to utilize the agents with poor link dynamic factor propagation ability and light load on the network layer, it is necessary to group the tasks on the agents across the network layer; when grouping, different network layer A [l] Risk Level RN [l] It is defined as follows:

[0029]

[0030] Calculate the risk level of different network layers. This risk level indicator comprehensively measures the load, dynamic situation, and ability to propagate dynamic factors of the network layer. Then, search for grouping combinations that can increase migration benefits after task grouping and merging, and merge the corresponding groups. Group the tasks of two different network layers. The benefits of a merger are determined as follows:

[0031]

[0032] in Used to represent task grouping g m Accept g when the maximum migration benefit is achieved m The intelligent agent, Used to represent task grouping g n Accept g when the maximum migration benefit is achieved n 's intelligent agent.

[0033] As a further improvement of the present invention, after grouping tasks on the same agent, between the same network layer, and between different network layers, task migration is performed based on this; all task groups are traversed to construct the agent set A to which the task group is migrated. m , where if the number of tasks expected to be executed in the task group |g m |·R [l] Less than the number of tasks performed on the agent to which it is expected to be transferred |Q m|, then give up the task in the task group to reduce the cost of task completion; then for the agent set A m The tasks on the task execution task migration, clear the agent set A m The load is to receive the task group, and also to avoid A after receiving the task group that may cause link dynamics. m When the original task on the agent is re-migrated, the link connected to it becomes dynamic, which leads to the expansion of the influence range of dynamic factors. Finally, each task group b i Transfer to the corresponding agent a m and updates the agent's task execution and the cost of task migration.

[0034] Beneficial effects:

[0035] (1) Reduce the risk of cascading diffusion of dynamic factors during the task migration process. In the adaptive task migration method, taking into account the overload risk and task cost, the number of times the agent accepts tasks is reduced by migrating tasks in groups, thereby reducing the risk of dynamic factor propagation caused by the cascade coupling of link dynamic factors with the dynamics of agents in large-scale UAV clusters.

[0036] (2) High task completion rate Regardless of the number of tasks, the adaptive task migration method takes into account the impact of link dynamics on the cascade propagation of dynamic factors brought about by large-scale UAV clusters. On the premise of minimizing the increase in task completion cost, it solves the problem of reduced task completion rate caused by the unlimited diffusion of dynamic factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flowchart showing the grouping of tasks on different agents on the same network layer in a large-scale drone swarm.

[0038] Figure 2 It is a flowchart showing the task grouping on different agents at different network layers in a large-scale drone swarm.

[0039] Figure 3 It is a flowchart showing the task migration process in a large-scale drone swarm.

[0040] Figure 4 It is the main principle diagram of the method of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0042] like Figure 4As shown, the present invention provides an adaptive task migration method for link dynamics in the context of UAV cluster communication and decision-making, in order to cope with the challenges brought by link dynamics. As the UAV network structure becomes increasingly complex, the instability of the link affects the interaction stability between agents. This method migrates tasks affected by dynamic factors to other agents through grouping and layering strategies, reduces the number of task receptions, and reduces the risk of dynamic factor propagation. The specific operations include modeling drones as agents, modeling drone groups at different levels as network layers, grouping agent tasks into three categories, and selecting the migration agent based on the risk and cost of agent acceptance and task completion. Compared with the traditional centralized task migration method, the present invention effectively reduces the impact of dynamic factors and improves the task completion rate. Specifically including:

[0043] (1) Based on the load of itself and other agents and the cost of executing tasks, the agents in a large-scale UAV swarm migrate tasks to the target agent, taking into account the cost of task migration. This reduces the overall cost of task completion and increases the task completion rate. In the initial stage, the task model, agent model, link dynamics model, link dynamics propagation impact model, link dynamics impact model during group task migration, and task completion rate model are defined. The model is described as follows:

[0044] Mission model definition: Each mission t in the mission set T in a large-scale UAV swarm k Use a two-tuple <size(t k ),pos(t k )> to express. Among them, size(t k ) indicates the execution of task t k The load size of different loads requires different costs for the same agent to execute tasks. Assuming that each task is atomic and cannot be divided into subtasks, the load of each task cannot be divided; pos(t k ) represents the agent that executes the task. Due to the atomicity of the task, a task can only be executed by one agent at a time.

[0045] ·Agent model definition: The agents in a large-scale drone swarm represent drones that can independently complete a given task. Each agent a i Use triples Indicates. Where v i Represents agent a i The size of the task that can be performed at unit cost; Q i Represents agent a i The task queue being executed, i.e. Q i ={t1,t2,…,t k}; Represents agent a i The set of network layers to which it belongs. In an actual multi-layered large-scale drone cluster, each intelligent agent can belong to a different network layer.

[0046] Link dynamic model definition: The link uses a variable d with a value between 0 and 1. ij Represents agent a i With agent a j The probability of link dynamic factors appearing between ij The larger the value of , the more likely it is to generate link dynamics during task migration. ij =1, it means that agent a i With agent a j Link dynamics will inevitably occur when migrating tasks between them. i Represents agent a i The dynamics of all connected links, i.e.

[0047] Definition of the link dynamics propagation model: In large-scale drone swarms, link dynamics will propagate to other agents during task migration. The propagation of this link dynamics is represented as:

[0048]

[0049] in Then it means task t k From agent a i Transfer to agent a j ,otherwise R [l] R represents the propagation capability level of link dynamic factors along with task migration through link type l. [l] The smaller the value, the more intelligent the agent a i During the task migration process, the lower the probability of link dynamics propagation when receiving tasks through link type l; Spread(·) is the link dynamics propagation probability function, with a value between 0 and 1. This formula shows that the propagation process of link dynamics factors is related to the presence of situations in the surrounding agents that may cause link dynamics factors, whether task migration has been performed between agents, and the cost of task migration. Agent a i The more dynamic the links connected to the surrounding agents are, the more likely agent a is to i The surrounding links are affected by the spread of dynamic factors and the probability of link dynamics is higher; and the spread of this link dynamic factor between agents is triggered by the task dynamic process of task migration. If agent a i and agent a jIf there is no task migration between them, then this dynamic factor will not be propagated; the cost c of the migration task is ij The process of seeking quotient indicates that the higher the cost of task transfer between agents, the more difficult it is to propagate the dynamic factors caused by task transfer.

[0050] Task completion ratio model: In a UAV network with dynamic links, in addition to tasks that are abandoned by dynamic agents and tasks that are not executed due to overload risks, there are also situations where task migration cannot be completed due to link dynamics. Therefore, the task completion ratio in the system after executing the task migration strategy π is calculated. Redefine as:

[0051]

[0052] in Indicates the task that was abandoned on the agent, Indicates that the task cannot be completed due to the dynamic nature of the link. Indicates tasks that were not executed due to overload risk.

[0053] Definition of the model for link dynamics during grouped task migration: In scenarios where links in large-scale drone swarms are dynamic, we use grouped task allocation or migration to allow agents to migrate or receive a group of tasks at a time. The dynamics of each group of tasks can be attenuated by the grouping method w. Tasks are migrated to agent a according to the grouping method W. i For a j The influence function of the probability of dynamicity of connected links is defined as follows:

[0054]

[0055] Where 0<ψ(·)<1, the value of the function damp increases with the number of tasks in the task group. The increase of monotonically decreases, and transfers to agent a i Task Force The function ψ is defined as Where σ is a given attenuation factor. W={w1,w2,…,w |w|} represents the set of all task grouping methods, each task grouping method wi corresponds to the corresponding task group

[0056] (1) Task grouping stage on the same agent: In the scenario of link dynamics, task migration is triggered by the emergence of factors on the agent that can lead to link dynamics. Such factors will also lead to the dynamics of the agent, that is, dynamic factors. Tasks on the agent with dynamic factors all need to be migrated, and these tasks have the same impact on the agent and link dynamics after the task migration process, and have natural isomorphism. Therefore, the tasks performed on each agent with dynamic factors are first divided into a separate task group. These task groups are represented as G = {g1, g2, …, g |G|}.

[0057] (2) Task grouping stage for different agents on the same network layer: Considering that in large-scale drone cluster scenarios, dynamic factors will affect multiple agents on the same network layer at the same time. It is necessary to group the tasks of different agents on the same network layer. First, the agents on each network layer are divided into affected sets according to whether they are affected by dynamic factors. and normal run collection Task migration in different situations will bring different benefits to the UAV cluster. m ∈G migrates to agent a in the normal operation set i The resulting task migration benefit is defined as follows:

[0058]

[0059] It contains task group g m Transfer to agent a i Cost Contains agent a i The cost of migrating one's own tasks in order to accept task grouping and Implementation Task Force m The cost of the task and Task Group g m Transferred to agent a i The proportion of tasks that can be completed Based on this, we can determine the maximum benefit of task group migration, such as Figure 1 As shown, through continuous iterative calculation, the benefits obtained after merging different task groups into one task group are compared with the benefits of the unmerged task groups to select the final task grouping.

[0060] (3) Task grouping stage on different agents across network layers: Considering that the links of different network layers have different abilities to propagate link dynamic factors along with task migration. For network layers with poor link dynamic factor propagation ability, more tasks can be migrated to them to reduce the propagation of dynamic factors in the system. In order to utilize agents with poor link dynamic factor propagation ability and light load on the network layer, it is necessary to group tasks on agents across network layers. When grouping, different network layer A [l] Risk Level RN [l] It is defined as follows:

[0061]

[0062] Calculate the risk level of different network layers. This risk level indicator comprehensively measures the load, dynamics, and ability to propagate dynamic factors of the network layer, and traverses to find the grouping combination that can increase the migration benefit after the task grouping is merged, and perform the corresponding grouping merger. Group the tasks of two different network layers The benefits of a merger are determined as follows:

[0063]

[0064] in Used to represent task grouping g m Accept g when the maximum migration benefit is achieved m The intelligent agent, Used to represent task grouping g n Accept g when the maximum migration benefit is achieved n Agent. Figure 2 As shown in the figure, the network layers with the highest and lowest risks are sequentially determined to determine whether tasks can be merged. After the determination is completed, the two network layers that have been determined are removed from the set to be determined, and the next round of determination is continued. This is done until all network layers have been determined, thus achieving the goal of grouping tasks of different agents across network layers.

[0065] (4) Task migration stage: After grouping tasks on the same agent, between the same network layer, and between different network layers, task migration is performed based on this. Figure 3 As shown, first traverse all task groups and construct the agent set A to which the task groups are transferred m , where if the number of tasks expected to be executed in the task group |g m |·R [l] Less than the number of tasks performed on the agent to which it is expected to be transferred |Q m |, then give up the task in the task group to reduce the cost of task completion; then for the agent set A mThe tasks on the task execution task migration, clear the agent set A m The load is to receive the task group, and also to avoid A after receiving the task group that may cause link dynamics. m When the original task on the agent is re-migrated, the link connected to it becomes dynamic, which leads to the expansion of the influence range of dynamic factors. Finally, each task group b i Transfer to the corresponding agent a m and updates the agent's task execution and the cost of task migration.

Claims

1. A method for intelligent target task allocation for UAV swarm communication and decision-making based on link dynamics, characterized by: Specific steps The approach is as follows: First, each drone in a drone swarm is modeled as an agent. When the swarm is too large, drones at different levels are modeled as different network layers. Second, to reduce the frequency with which agents accept tasks and mitigate the risk of dynamic factor propagation caused by task migration, the agents' tasks are grouped into three categories: tasks within the same agent, tasks on different agents at the same network layer, and tasks across different agents at different network layers. Finally, based on the task groupings, the most suitable agent is selected for task group migration, taking into account the risks and costs of task acceptance and task completion. Each task t in the task set T in the UAV cluster k Can use a binary <size(t k ),pos(t k )> to indicate; among them, size(t k ) indicates the execution of task t k The cost required for the agent to perform tasks with different loads is different; pos(t k ) represents the agent that performs the task. At the same time, a task can only be performed by one agent, and the agents in the drone cluster represent drones that can independently complete a given task. Each agent a i Use triples Indicates that v i Represents agent a i The size of the task that can be performed at unit cost; Q i Represents agent a i The task queue being executed, i.e. Q i ={t1,t2,…,t k }; Represents agent a i The set of network layers it belongs to. In a large-scale drone swarm, each agent can belong to a different network layer. Use a variable d with a value between 0 and 1 ij Represents agent a i With agent a j The probability of link dynamic factors appearing between ij The larger the value of , the more likely it is to generate link dynamics during task migration. ij =1, it means that agent a i With agent a j Link dynamics will inevitably occur when migrating tasks between them; using vector D i Represents agent a i The dynamics of all connected links, i.e.

2. The method for intelligent target task allocation for UAV cluster communication and decision-making based on link dynamics according to claim 1 is characterized by: Since link dynamics will propagate to other agents during the task migration process in a large-scale UAV swarm, the propagation of link dynamics is expressed as: in Then it means task t k From agent a i Transfer to agent a j ,otherwise R [l] R represents the propagation capability level of link dynamic factors along with task migration through link type l. [l] The smaller the value, the more intelligent the agent a i During the task migration process, the lower the probability of link dynamics propagation when receiving tasks through link type l; Spread(·) is the link dynamics propagation probability function, with a value of 0-1; this formula shows that the propagation process of link dynamics factors is related to the presence of situations in the surrounding agents that may cause link dynamics factors, whether tasks are migrated between agents, and the cost of task migration; Agent a i The more dynamic the links connected to the surrounding agents are, the more likely agent a is to i The surrounding links are affected by the spread of dynamic factors and the probability of link dynamics is higher; and the spread of this link dynamic factor between agents is triggered by the task dynamic process of task migration. If agent a i and agent a j If there is no task migration between them, then this dynamic factor will not be propagated; the cost c of the migration task is ij The process of seeking quotient indicates that the higher the cost of task transfer between agents, the more difficult it is to propagate the dynamic factors caused by task transfer.

3. The method for intelligent target task allocation for UAV cluster communication and decision-making based on link dynamics according to claim 2 is characterized by: In the scenario where links in a large-scale UAV cluster are dynamic, the task grouping or migration method is adopted to enable the agent to migrate or receive a group of tasks through the link at a time; the dynamics brought by each group of tasks is attenuated according to the grouping method W; the tasks are migrated to the agent a according to the grouping method W. i For a j The influence function of the probability of dynamicity of connected links is defined as follows: Where 0<ψ(·)<1, the value of the function damp increases with the number of tasks in the task group. The increase of monotonically decreases, and transfers to agent a i Task Force The function ψ is defined as Where σ is a given attenuation factor; W = {w1, w2, ..., w |w| } represents the set of all task grouping methods, each task grouping method w i Each corresponds to the corresponding task group 4. The method for allocating intelligent target tasks for UAV cluster communication and decision-making based on link dynamics according to claim 3 is characterized by: In a UAV network with dynamic links, in addition to tasks being abandoned by dynamic agents and tasks not being executed due to overload risks, there are also situations where link dynamics caused by task migration can lead to the inability to complete task migration. Therefore, the task completion ratio in the system after executing the task migration strategy π Redefine as: in Indicates the task that was abandoned on the agent, Indicates that the task cannot be completed due to the dynamic nature of the link. Indicates tasks that were not executed due to overload risk.

5. The method for intelligent target task allocation for UAV cluster communication and decision-making based on link dynamics according to claim 4 is characterized by: By grouping tasks, the number of times an agent receives tasks is reduced, thereby reducing the risk of dynamic factors spreading in the UAV network due to the dynamic nature of task migration. Before performing grouped task migration, tasks need to be grouped in a hierarchical manner. In the task grouping stage in a large-scale UAV cluster scenario, tasks are divided into the following three categories according to the distribution of agents where the tasks are located: (1) tasks on agents with the same dynamic factors; (2) tasks on agents with different dynamic factors at the same network layer; (3) tasks on agents with different dynamic factors at different network layers. For these three situations, the task grouping steps are also divided into the following three stages: (1) task grouping stage on the same agent; (2) task grouping stage on different agents at the same network layer; (3) task grouping stage on different agents across network layers.

6. The method for intelligent target task allocation for UAV cluster communication and decision-making based on link dynamics according to claim 5 is characterized by: Specifically, in the task grouping stage on the same agent, in the context of link dynamics, task migration is triggered by factors that can cause link dynamics in the agent. These factors also cause the dynamics of the agent, i.e., dynamic factors. Tasks on agents that experience dynamic factors all need to be migrated, and after executing the task migration process, these tasks have the same impact on the agent and link dynamics, thus exhibiting natural isomorphism. Therefore, we first divide the tasks performed on each agent with dynamic factors into a separate task group; these task groups are represented by G = {g1, g2, ..., g |G| }; The task grouping stage of different agents on the same network layer is specifically as follows: considering that in large-scale drone cluster scenarios, dynamic factors will affect multiple agents on the same network layer at the same time, it is necessary to group the tasks of different agents on the same network layer; first, the agents on each network layer are divided into affected sets according to whether they are affected by dynamic factors. and normal run collection Task migration in different situations will bring different benefits to the UAV cluster. m ∈G migrates to agent a in the normal operation set i The resulting task migration benefit is defined as follows: It contains task group g m Transfer to agent a i Cost Contains agent a i The cost of migrating one's own tasks in order to accept task grouping and Implementation Task Force m The cost of the task and Task Group g m Transferred to agent a i The proportion of tasks completed later Based on this, the maximum benefit obtained by migrating the task groups is determined. Through continuous iterative calculations, the benefits obtained by merging different task groups into one task group are compared with the benefits obtained by not merging the task groups, and the final task grouping is selected. The task grouping stage on different agents across the network layer is specifically as follows: considering that the links of different network layers have different abilities to propagate link dynamic factors along with task migration; for the network layer with poor link dynamic factor propagation ability, more tasks are migrated therein, so as to reduce the propagation of dynamic factors in the system; in order to utilize the agents with poor link dynamic factor propagation ability and light load on the network layer, it is necessary to group the tasks on the agents across the network layer; when grouping, different network layer A [l Risk level RN [l ] is defined as follows: Calculate the risk level of different network layers. This risk level indicator comprehensively measures the load, dynamic situation, and ability to propagate dynamic factors of the network layer. Then, search for the grouping combination that increases the migration benefit after the task grouping is merged, and merge the corresponding groups. Group the tasks of two different network layers. The benefits of a merger are determined as follows: in Used to represent task grouping g m Accept g when the maximum migration benefit is achieved m The intelligent agent, Used to represent task grouping g n Accept g when the maximum migration benefit is achieved n 's intelligent agent.

7. The method for intelligent target task allocation for UAV cluster communication and decision-making based on link dynamics according to claim 6 is characterized by: After grouping tasks on the same agent, between the same network layer, and between different network layers, task migration is performed based on this; Traverse all task groups and construct the agent set A to which the task groups are transferred m , where if the number of tasks expected to be executed in the task group |g m |·R [l] Less than the number of tasks performed on the agent to which it is expected to be transferred |Q m |, then give up the task in the task group to reduce the cost of task completion; then for the agent set A m The tasks on the task execution task migration, clear the agent set A m The load is to receive the task group, and also to avoid A after receiving the task group that may cause link dynamics. m When the original task on the agent is re-migrated, the link connected to it becomes dynamic, which leads to the expansion of the influence range of dynamic factors. Finally, each task group b i Transfer to the corresponding agent a m and updates the agent's task execution and the cost of task migration.

Images