A method for attention grouping of uav assisted service caching and task offloading

Abstract

The application discloses an unmanned aerial vehicle (UAV) auxiliary service caching and task offloading method based on attention grouping, relates to the technical field of wireless network caching, and constructs a double-layer UAV aerial edge network system, including a quadcopter UAV cluster, a fixed-wing UAV and a ground user; the fixed-wing UAV and the quadcopter UAV cluster are used to jointly offload and calculate a task chain generated by the ground user; the quadcopter UAV cluster is grouped by using an attention mechanism, task subdivision is performed according to grouping, collaborative service caching and task offloading are performed, a double-layer UAV aerial edge network model is constructed and solved, and the minimum delay of the double-layer UAV aerial edge network system in processing a task is obtained. The problem that the task calculation and storage burden of the edge network system is too large is solved, and the overall performance of the edge network system is improved.

Description

Technical Field

[0001] This invention relates to the technical field of wireless network caching, and more particularly to a method for collaborative service caching and task offloading. Background Technology

[0002] To address the issues of long transmission latency in traditional centralized network architectures, edge networks utilize Mobile Edge Computing (MEC) technology to shift the core functions of the network from centralized data centers to the network edge, closer to users, thereby achieving lower network latency and higher data throughput.

[0003] Existing user requests are redundant in network services. Wireless edge caching strategies can effectively utilize this redundancy, significantly improving the utilization rate of network resources such as computing and communication. Based on the cached content, existing caching strategies can be divided into content caching and service caching. The former caches the computation results of computing tasks, while the latter caches data associated with computing tasks, such as databases. Service caching is a crucial part of edge networks. Due to the distributed nature of edge networks, user requests may be sent to any edge node in the network. Service caching allows a single node to handle multiple computing tasks. Designing efficient service caching strategies in edge networks is a key technical challenge.

[0004] In edge networks, task offloading typically occurs from terminal devices with limited computing power and battery life to edge servers with stronger computing power and more stable power supply. Conventionally, ground base stations can act as edge servers, assisting ground nodes in providing services. However, fixed base station deployments struggle to meet the needs of dynamically changing ground nodes. To compensate for the shortcomings of mobile edge computing based on fixed ground infrastructure, a drone-assisted mobile edge computing approach has emerged. Drones offer advantages such as rapid mobility, convenient deployment, and low cost, effectively assisting ground base stations in handling the dynamic computing demands of ground nodes. By carrying edge computing servers on drones, they further enhance the computing performance of ground users.

[0005] In edge networks, service caching and task offloading are complementary in edge design. Tasks on a node can only be offloaded to edge servers that already cache the relevant services. Service caching reduces data transmission latency and improves data access speed, while task offloading reduces the computational burden on user devices and improves data processing speed. Combining service caching and task offloading, edge networks can provide users with more efficient and lower-latency services. Since edge servers have limited storage capacity, service caching and task offloading strategies need to be designed wisely to maximize edge computing performance.

[0006] Existing research typically addresses the service caching and task offloading problem using traditional optimization methods. These methods require complete information from the network, which is difficult to achieve in real-world, dynamic, and distributed MEC networks. When faced with a complex task, most approaches divide it into task chains that can be executed in a specific order according to its inherent logic. Then, service chains are built at the service cache level, and task offloading is performed.

[0007] However, this type of caching and offloading strategy ignores the relationship between existing cached resources in the network environment and the task offloading strategy, and cannot fully utilize network resources. In edge network task offloading, edge nodes have limited computing and storage resources. An edge node cannot cache all the services required to support the entire edge network system. Moreover, in high computing load environments, many complex application processes consist of multiple different execution and storage components with a chained structure, and there are also multiple task processing modules. When a task cannot be processed by the edge server, it is usually sent to a remote center for processing. This increases the transmission latency for task completion in the entire edge network system, increases the task computing and storage burden on the system's edge nodes, and reduces system performance. Summary of the Invention

[0008] This invention addresses the problem in existing edge network systems where edge nodes suffer from excessive task computation and storage burdens, leading to long task completion times and poor performance. It proposes an attention-grouping method for caching and offloading tasks for UAV-assisted services. By offloading task chains generated by ground users to multiple quadcopter and fixed-wing UAVs for task computation, the invention reduces the task computation and storage burden on each edge node, improves system performance, and accelerates task completion time.

[0009] To achieve the objectives of this invention, the following technical solution is adopted:

[0010] A method for caching and offloading drone-assisted services based on attention grouping includes the following steps:

[0011] S1: Construct a two-layer UAV airborne edge network system, which includes: a fixed-wing UAV, I quadcopter UAVs and J ground users. The I quadcopter UAVs form a quadcopter UAV cluster, and the fixed-wing UAVs, the quadcopter UAV cluster and the ground users establish communication connections with each other.

[0012] S2: The fixed-wing UAV and the quadcopter UAV cluster work together to unload and process the task chain generated by the ground user. The task in the task chain is a computing task. The unloading process means unloading the computing task to the quadcopter UAV cluster or the fixed-wing UAV for computing.

[0013] S3: With the goal of minimizing the latency of the two-layer UAV aerial edge network system in processing tasks, attention mechanism is used to group the quadcopter drone swarm and further subdivide the tasks based on the grouping.

[0014] S4: Based on grouping and task subdivision, coordinate service caching and task offloading, construct and solve a two-layer UAV air edge network model, and obtain the minimum latency of the two-layer UAV air edge network system for task offloading and computation processing of the task chain.

[0015] The dual-layer UAV refers to a fixed-wing UAV layer and a quadcopter UAV swarm layer.

[0016] According to the above technical solution, a two-layer UAV aerial edge network system is constructed. Fixed-wing UAVs and quadcopter UAV swarms jointly offload and process tasks generated by ground users, reducing the task computation and storage burden on individual edge nodes in the edge network system. With the goal of minimizing the latency of task processing in the two-layer UAV aerial edge network system, an attention mechanism is used to group the quadcopter UAV swarm, and tasks are further subdivided based on these groups. Based on the grouping and task subdivision, collaborative service caching and task offloading are implemented to construct and solve a two-layer UAV aerial edge network model, obtaining the minimum latency for task offloading and computation processing of task chains in the two-layer UAV aerial edge network system. This effectively solves the problem of excessive task computation and storage burden in the edge network system, leading to long task completion times and poor performance, and improves the overall performance of the edge network system.

[0017] Furthermore, the specific process of grouping the quadcopter drone swarm using an attention mechanism and further subdividing the tasks based on the grouping in step S3 is as follows:

[0018] S31: Obtain the service cache information of the service chain and the quadcopter drone required by the task chain, and obtain the service cache matching degree between the task chain and the quadcopter drone based on the service chain and the service cache information.

[0019] S32: Based on the service cache matching degree between the quadcopter drones and the task chain, the quadcopter drone cluster is grouped using an attention mechanism;

[0020] S33: Calculate the attention score of each quadcopter drone group, update the quadcopter drone members in the attention group based on the attention score, and re-subdivide the initial task chain according to the new quadcopter drone members in the attention group to obtain a new serial-parallel hybrid task execution order chain.

[0021] The specific process of obtaining the service cache information of the service chain and the quadcopter drone required for the task chain in step S31, and obtaining the service cache matching degree between the task chain and the quadcopter drone based on the service chain and the service cache information, is as follows:

[0022] Let S be the task chain initiated by the ground user, and F be the service chain required by the task chain S; let UAV be the set of quadcopter drone clusters, and U be the service cache information of the quadcopter drones; let c be the service cache matching degree between the quadcopter drones and the task chain. ik ;

[0023]

[0024] Where S = {s1, s2, ..., s} k ,…,s K}, K = {1,2,…,k,…K} indicates that the current task chain contains K tasks, s k Let F = {f1, f2, ..., fk} be the k-th task in the task chain. k ,…,f K}, f k Indicates task s k Required services; UAV = {1,2,…,i,…,I}, where I is the number of quadcopter drones, U = [U1,U2,...,U i ,...,U I ], U i This represents all services cached by the i-th quadcopter drone. n i UAV i Number of cached services, UAV i Let i represent the i-th quadcopter drone.

[0025] The specific process of grouping the quadcopter drone cluster using an attention mechanism based on the service cache matching degree between the quadcopter drone and the task chain, as described in step S32, is as follows:

[0026] Suppose service f is cached k The set of the number of quadcopter drones is V kLet u be the set of quadcopter drones that have cached the corresponding service in service chain F. Determine the size of the variable in set u, and divide the quadcopter drone cluster into attention group A and attention group B based on the size of the variable in set u. The size of the variable in set u represents whether any quadcopter drone in set u has cached the corresponding service in service chain F. The process satisfies:

[0027] If the variable in set u is greater than or equal to 1, it means that one or more quadcopter drones are simultaneously caching service f. k Let's call this attention group A;

[0028] If the variable in set u is less than 1, it means that there is no service in the quadcopter drone cache service chain F, which is denoted as attention group B;

[0029] in, u k This indicates that service f is cached. k The number of quadcopter drones, u = {u1, u2, ..., u k ,…,u K}

[0030] Step S33, which involves calculating the attention score of each quadcopter drone group and updating the quadcopter drone members in the attention group based on the attention score, is as follows:

[0031] Let task s k Communication quality C between attention group A and attention group B k and computing power M k According to communication quality C k and computing power M k The specific process for calculating the attention scores of attention group A and attention group B is as follows:

[0032] Communication quality C k and computing power M k Transform into feature F C and feature F M This indicates that feature F is used. C and feature F M The attention weights are calculated, and the process satisfies the following:

[0033] Let the query vector be Q = [q1, q2]. Additive attention is used to calculate the attention weights, and a softmax function is added to normalize the obtained attention weights, thus yielding the communication quality C. k The attention weight is a Ci =softmax((Q·W) qc )+(f Ci ·W fc Computing power Mk The attention weight is a Mi =softmax((Q·W) qm )+(f Mi ·W fm ));

[0034] Task s is obtained by calculating attention weights. k Attention score A of the middle attention group A k Attention score B of attention group B k ;

[0035] in, f Ci = g(Ci), where g is the feature mapping function, f Mi = h(Mi), where h is the feature mapping function, q1 represents the degree of concern for communication quality, q2 represents the degree of concern for computing power, and W qc and W fc These are the learning parameters, where · represents the dot product, and W qm and W fm It is the learning parameter, A k =a Ci f Ci +a Mi f Mi B k =a Ci f Ci +a Mi f Mi ;

[0036] Based on attention score A k Quadrone drones with an attention score below 0.5 are eliminated, and the remaining quadcopters are arranged according to the service order of the service chain to form a new attention group A.

[0037] Based on attention score B k Quadrone drones with an attention score below 0.5 will be eliminated, and the remaining quadcopter drones will be used as service providers, forming a new attention group B.

[0038] The attention score ranges from 0 to 1, and the service provider refers to a quadcopter drone that is capable of providing services and is waiting to provide services.

[0039] The specific process described in step S33, which involves re-subdividing the initial task chain based on the quadcopter drone members in the new attention group to obtain a new serial-parallel hybrid task execution sequence chain, is as follows:

[0040] Based on the new attention groups A and B, let u′ be the set of quadcopter drones corresponding to the services in service chain F that the new attention group A caches, where u′=[u′1,u′2,…,u′…]. k ,…,u′ K ], u′ k This indicates that service f is cached in the new attention group A. k The number of drones;

[0041] By utilizing ground users, the initial task chain S is subdivided according to the quadcopter drone members in the new attention group A, resulting in a new serial-parallel hybrid task execution sequence chain.

[0042] According to the above technical solution, the matching degree between the service chain required by the task chain and the service cache information of the quadcopter is obtained. Based on the service cache matching degree, the quadcopter cluster is grouped using an attention mechanism, and the attention score of each group is calculated. Quadrotors with a score lower than 0.5 are eliminated based on the attention score, thereby updating the quadcopter members of the attention group. The initial task chain is then subdivided into new serial-parallel hybrid task execution order chains, which maximizes the utilization of quadcopters when executing tasks in the task chain, effectively saving resources for quadcopters and improving the overall performance of the edge network system.

[0043] Furthermore, the specific process of constructing and solving the two-layer UAV airborne edge network model based on grouping and task subdivision, and coordinating service caching and task offloading, as described in step S4, is as follows:

[0044] S41: The quadcopter drone in the new attention group A is used to process the tasks that cache the corresponding services required in the task chain S. The collaborative service caching and task offloading strategy will process any task in the task chain S on the quadcopter drone, fixed-wing drone cloud platform, or ground user server in the new attention group B; wherein, the fixed-wing drone is used as a computing cloud platform, i.e., a fixed-wing drone cloud platform.

[0045] S42: Calculate the processing latency of task chain S in the new attention group B, including the quadcopter UAV, fixed-wing UAV cloud platform, and ground user server. Based on the processing latency of task chain S, construct and solve a two-layer UAV airborne edge network model.

[0046] The collaborative service caching and task offloading strategy described in step S41, which processes any task in task chain S on a quadcopter drone cluster, fixed-wing drone cloud platform, or ground user server in the new attention group B, is as follows:

[0047] Let the position of the task in the task chain S be L. k When L k =0 indicates that the task is in the first half of the task chain S, when L k =1 indicates that the task is in the latter half of task chain S;

[0048] Let any task be s k Tasks k The latency limit is: t k ≤τ k Tasks k The computing power requirement is Tasks k Data upload time is

[0049] Establish a three-dimensional Cartesian coordinate system, and let the coordinate position of the ground user server be (x... l ,y l ,0), the position of the fixed-wing UAV is set to The position of the quadcopter drone is set to Tasks k The previous task k-1 The position is set to (x k-1 ,y k-1 ,z k-1 According to task s k The coordinates of the task k-1 The coordinates are used to calculate the distance D between the destinations of the two tasks executed consecutively. k The distance between the destinations of the two tasks is the distance between the destination of the previous task and the destination of the current task, and the process satisfies:

[0050] If task s k If executed on the ground user server, then the distance to D k =D k，local , If task s k When executed by a fixed-wing UAV, the distance is D k =D k，u0 , If task s k The quadcopter drones operating in the new attention group B are at a distance of d. k =D k，ui , Wherein, the distance D k d represents the distance d between the destinations of the two tasks executed consecutively. k ;

[0051] According to task s k Computing power requirements Distance D k And the position of the task in the task chain S L k Collaborative service caching and task unloading determine the tasks s in the task chain S. k Processing is performed on the cloud platform or ground user server of the quadcopter or fixed-wing drone in the new attention group B;

[0052] Among them, s k =(d k ,w k ,τ k ,f k ); d represents the size of the task input data, d k Indicates task s k The size of the input data; w represents the number of CPU cycles required to complete the task, w k Indicates completion of task s k The number of CPU cycles required, i.e., the computational complexity; τ k f represents the latency tolerance, i.e., the maximum processing latency. k Indicates task s k The corresponding service; w′ k Indicates task s k The number of CPU cycles executed per second by the server, w′ k =M k That is, the number of CPU cycles executed per second by the quadcopter drone is The number of CPU cycles executed per second by a fixed-wing UAV is The number of CPU cycles executed per second by the ground user server is r represents the communication transmission rate of the task, r k Indicates task s k The communication transmission rate is given by: B represents the bandwidth of the communication channel, N represents the noise power of the channel, S represents the signal power of the signal, P represents the transmission power of the device, and G represents the transmission gain of the signal.

[0053] According to task s k Computing power requirements Distance D k And the position of the task in the task chain S L k Collaborative service caching and task unloading determine the tasks s in the task chain S. k The specific process for processing on the quadcopter or fixed-wing drone cloud platform or ground user server in the new attention group B is as follows:

[0054] S411: Determine the computing power requirements of the ground user server like This indicates that the ground user server satisfies task s. kComputing power requirements The option to retain ground user servers is retained; if This indicates that the ground user server cannot fulfill task s. k Computing power requirements Remove the option to select ground user servers;

[0055] S412: Determine task s k Position L in task chain S k If L k =0 indicates task s k The first half of task chain S indicates that the quadcopter drone in the new attention group B has not yet downloaded the cached task s from the fixed-wing drone cloud platform. k Required services f k Remove the quadcopter drone selection from the new attention group B; if L k =1, indicating task s k The latter half of task chain S indicates that the quadcopter drone in the new attention group B has downloaded and cached task s from the fixed-wing drone cloud platform. k Required services f k The option of quadcopter drones will be retained in the new attention group B;

[0056] S413: Combine the remaining ground user server (after the judgments in steps S411 and S412) with the quadcopter drone in the new attention group B, and the fixed-wing drone cloud platform to integrate task s. k Computing power requirements and distance D k Select one of the following to perform the task: a ground user server, a quadcopter drone, or a fixed-wing drone cloud platform from the new attention group B. k ;

[0057] S414: Let the collaborative service caching and task offloading strategy be α. k α k =0 indicates task s k Processed on a fixed-wing UAV cloud platform; α k =i, where i = {1, 2, ..., I} represents task s k Processed on a quadcopter drone; α k =I+1 represents task s k Processed on the ground user server; where α k ∈{0}∪N;

[0058] Let the correlation factor be a, Indicates task s k Whether to process on a fixed-wing UAV, 1 for yes, 0 for no; Indicates task s k Whether to process on a quadcopter drone, 1 for yes, 0 for no; a k,local Indicates task s k Whether to execute on the ground user server, 1 for yes, 0 for no;

[0059]

[0060] Performing the same task k At the same time, the expression is satisfied:

[0061] The specific process of constructing a two-layer UAV aerial edge network model based on the processing latency of task chain S in step S42, which involves the computational task chain S across the quadcopter UAV cluster, fixed-wing UAV cloud platform, and ground user server, is as follows:

[0062] Computational tasks s k The latency processed on a quadcopter drone swarm satisfies the following conditions:

[0063] When u′ k =1, task s k Processed on a quadcopter drone, task s k The processing latency consists of task data upload time, task calculation time, and waiting delay time;

[0064] Data upload time:

[0065] Task calculation time:

[0066] Waiting delay time: σ k ;

[0067] Tasks k The delay is:

[0068] When u′ k When >1, task s k The task is divided into multiple parts and processed in parallel on multiple quadcopter drones. k The processing latency is the slowest of all parallel programs, and the process satisfies:

[0069]

[0070]

[0071] Tasks k The delay is:

[0072] Quadrone drone swarm processing tasks k The delay is:

[0073]

[0074] in, For the communication transmission rate between ground users and quadcopter drones, This represents the bandwidth of the communication channel between the ground user and the quadcopter drone, where P represents the transmit power of the ground user. w represents the signal transmission gain between the ground user and the quadcopter drone swarm. k This indicates the completion of the computation task s. k The number of CPU cycles required u′ represents the number of CPU cycles executed per second by a quadcopter drone swarm. k For task s k The number of parallel tasks it is split into;

[0075] Computational tasks s k The latency processed on the fixed-wing UAV cloud platform satisfies the following conditions:

[0076] When task s k Unloaded to the fixed-wing UAV cloud platform, task s k The processing latency is the task data upload time and the task calculation time;

[0077] Data upload time:

[0078] Task calculation time:

[0079] Task processing on fixed-wing UAVs k Delay:

[0080] in, This indicates the communication transmission rate between ground users and fixed-wing UAVs. This represents the bandwidth of the communication channel between the ground user and the fixed-wing UAV, where P represents the transmit power of the ground user. w represents the signal transmission gain between ground users and fixed-wing UAVs. k This indicates the completion of the computation task s. k The number of CPU cycles required This indicates the number of CPU cycles executed per second by a fixed-wing UAV.

[0081] Computational tasks s k The latency processed on the ground user server satisfies the following process:

[0082] Execute tasks on the ground user server kTasks k The processing latency consists of task computation time and waiting delay time;

[0083] Task calculation time:

[0084] Waiting delay time:

[0085] Ground user server processing tasks k Delay:

[0086] Among them, w k This indicates the number of CPU cycles required to complete the computation task. This represents the number of CPU cycles executed per second by the ground user server.

[0087] Based on task s k The processing latency of the quadcopter drone swarm, the fixed-wing drone cloud platform, and the ground user server is used to obtain the total processing latency of task chain S as follows:

[0088] Based on the total processing latency of task chain S, the two-layer UAV airborne edge network model is derived as follows:

[0089]

[0090] The particle swarm optimization algorithm is used to solve the two-layer UAV airborne edge network model, and the minimum latency for task offloading and computation of the two-layer UAV airborne edge network system is obtained:

[0091] minT.

[0092] According to the above technical solution, after the quadcopter drones in the new attention group A process and cache the tasks corresponding to the services required in task chain S, the collaborative service caching and task offloading strategy allows any task in task chain S to be processed on the quadcopter drones, fixed-wing drone cloud platforms, or ground user servers in the new attention group B. Simultaneously, the processing latency of task chain S on the quadcopter drones, fixed-wing drone cloud platforms, and ground user servers in the new attention group B is calculated. Based on the processing latency of task chain S, a two-layer UAV aerial edge network model is constructed and solved using the particle swarm optimization algorithm. The minimum latency for the two-layer UAV aerial edge network system to offload and compute tasks in the task chain is obtained: minT. Through the cooperation between the quadcopter drone cluster, fixed-wing drones, and ground users, the task chain S generated by the ground users is effectively offloaded and computed, reducing the burden of task computation and storage in the edge network system and improving the overall performance of the edge network system.

[0093] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0094] By constructing a two-layer UAV aerial edge network system, a swarm of fixed-wing UAVs and quadcopter UAVs jointly offloads and processes tasks generated by ground users, thereby reducing the task computation and storage burden on individual edge nodes in the edge network system. With the goal of minimizing the latency of task processing in the two-layer UAV aerial edge network system, an attention mechanism is used to group the quadcopter UAV swarm and further subdivide tasks based on these groups. Based on the grouping and task subdivision, collaborative service caching and task offloading are implemented. A two-layer UAV aerial edge network model is constructed and solved to obtain the minimum latency for task offloading and computation processing of task chains in the two-layer UAV aerial edge network system. Through the cooperation between the quadcopter UAV swarm, fixed-wing UAVs, and ground users, the task chain S generated by ground users is effectively offloaded and processed, reducing the task computation and storage burden on the edge network system and improving the overall performance of the edge network system. Attached Figure Description

[0095] Figure 1 A flowchart illustrating an attention-grouping method for caching and offloading unmanned aerial vehicle (UAV) assisted services, provided in an embodiment of this application;

[0096] Figure 2 A diagram illustrating the configuration of a two-layer UAV aerial edge network system provided in this application embodiment;

[0097] Figure 3 Task chain breakdown diagram provided for embodiments of this application;

[0098] In the diagram, U2G-1Link represents the communication connection between the quadcopter drone swarm and the ground user.

[0099] U2G-2Link: Communication connection between fixed-wing UAVs and ground users;

[0100] U2U Link: A communication link between quadcopter drone swarms and fixed-wing drones in a dual-layer UAV aerial network;

[0101] UAV trajectory: the flight path of a fixed-wing unmanned aerial vehicle;

[0102] UAV layer_1: The lower layer of the two-layer UAV aerial network (quadcopter drone swarm);

[0103] UAV layer_2: The upper layer of the dual-layer UAV aerial network (fixed-wing UAVs). Detailed Implementation

[0104] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0105] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0106] Example 1:

[0107] A method for caching and offloading drone-assisted services based on attention grouping, see [link to relevant documentation]. Figure 1 This includes the following steps:

[0108] S1: Construct a two-layer UAV airborne edge network system, which includes: a fixed-wing UAV, I quadcopter UAVs and J ground users. The I quadcopter UAVs form a quadcopter UAV cluster, and the fixed-wing UAVs, the quadcopter UAV cluster and the ground users establish communication connections with each other.

[0109] S2. The fixed-wing UAV and the quadcopter UAV cluster work together to unload and process the task chain generated by the ground user. The task in the task chain is a computing task. The unloading process means unloading the computing task to the quadcopter UAV cluster or the fixed-wing UAV for computing processing.

[0110] S3: With the goal of minimizing the latency of the two-layer UAV aerial edge network system in processing tasks, attention mechanism is used to group the quadcopter drone swarm and further subdivide the tasks based on the grouping.

[0111] S4: Based on grouping and task subdivision, coordinate service caching and task offloading, construct and solve a two-layer UAV air edge network model, and obtain the minimum latency of the two-layer UAV air edge network system for task offloading and computation processing of the task chain.

[0112] The dual-layer UAV refers to a fixed-wing UAV layer and a quadcopter UAV swarm layer.

[0113] For example, see Figure 2 The specific communication and operation within the dual-layer UAV aerial edge network system are as follows:

[0114] The two-layer UAV network architecture consists of an upper layer of fixed-wing UAVs and a lower layer of quadcopter UAV swarms. Quadrotor UAVs offer advantages such as flexible deployment and high maneuverability, but their resources are limited; therefore, they are used as the lower-layer, flexibly deployed task execution nodes. Fixed-wing UAVs provide wide-area coverage and have abundant resources, thus serving as the computing cloud platform for the edge network system. Compared to traditional cloud computing centers, they are closer to the user end, resulting in lower transmission latency. The ground user end is the task initiator and also caches the services required for task execution, making it one of the task executors. In summary, the two-layer UAV edge network system offers lower latency and faster system response compared to traditional edge-cloud systems.

[0115] In this embodiment, a two-layer UAV aerial edge network system is constructed. Fixed-wing UAVs and quadcopter UAV swarms jointly offload and compute tasks from task chains generated by ground users, reducing the computational and storage burden on individual edge nodes within the edge network system. To minimize the latency of task processing in the two-layer UAV aerial edge network system, an attention mechanism is used to group the quadcopter swarm, and tasks are further subdivided based on these groups. Based on the grouping and task subdivision, collaborative service caching and task offloading are implemented to construct and solve a two-layer UAV aerial edge network model. This yields the minimum latency for task offloading and computation processing of task chains by the two-layer UAV aerial edge network system. This effectively solves the problem of excessive computational and storage burden on edge network systems, leading to long task completion times and poor performance, thus improving the overall performance of the edge network system.

[0116] Example 2:

[0117] In this embodiment, the process of grouping quadcopter drone swarms using an attention mechanism and further subdividing tasks based on these groups, with the goal of minimizing the latency of the two-layer UAV aerial edge network system processing tasks as described in step S3 of embodiment one, is explained. The specific process is as follows:

[0118] S31: Obtain the service cache information of the service chain and the quadcopter drone required by the task chain; based on the service chain and the service cache information, obtain the service cache matching degree between the task chain and the quadcopter drone, the process satisfying:

[0119] Let S be the task chain initiated by the ground user, and F be the service chain required by the task chain S; let UAV be the set of quadcopter drone clusters, and U be the service cache information of the quadcopter drones; let c be the service cache matching degree between the quadcopter drones and the task chain. ik ;

[0120]

[0121] Where S = {s1, s2, ..., s} k ,…,s K}, K = {1,2,…,k,…K} indicates that the current task chain contains K tasks, s k Let F = {f1, f2, ..., fk} be the k-th task in the task chain. k ,…,f K}, f k Indicates task s k Required services; UAV = {1,2,…,i,…,I}, where I is the number of quadcopter drones, U = [U1,U2,...,U i ,...,u I ], U i This represents all services cached by the i-th quadcopter drone. n i UAV i Number of cached services, UAV i Let i represent the i-th quadcopter drone.

[0122] S32: Based on the service cache matching degree between the quadcopter drones and the task chain, the quadcopter drone cluster is grouped using an attention mechanism, and the process satisfies:

[0123] Suppose service f is cached k The set of the number of quadcopter drones is V k Let u be the set of quadcopter drones that have cached the corresponding service in service chain F. Determine the size of the variable in set u, and divide the quadcopter drone cluster into attention group A and attention group B based on the size of the variable in set u. The size of the variable in set u represents whether any quadcopter drone in set u has cached the corresponding service in service chain F. The process satisfies:

[0124] If the variable in set u is greater than or equal to 1, it means that one or more quadcopter drones are simultaneously caching service f. k Let's call this attention group A;

[0125] If the variable in set u is less than 1, it means that there is no service in the quadcopter drone cache service chain F, which is denoted as attention group B;

[0126] in, u k This indicates that service f is cached. k The number of quadcopter drones, u = {u1, u2, ..., u k ,…,u K}

[0127] Step S33, which involves calculating the attention score of each quadcopter drone group and updating the quadcopter drone members in the attention group based on the attention score, is as follows:

[0128] Let task s k Communication quality C between attention group A and attention group B k and computing power M k According to communication quality C k and computing power M k Calculate the attention scores for attention group A and attention group B, satisfying the following process:

[0129] Communication quality C k and computing power M k Transform into feature F C and feature F M This indicates that feature F is used. C and feature F M The attention weights are calculated, and the process satisfies the following:

[0130] Let the query vector be Q = [q1, q2]. Additive attention is used to calculate the attention weights, and a softmax function is applied to normalize these weights, ensuring the sum of all attention weights equals 1. This guarantees the effectiveness and rationality of the weights, and further yields the communication quality C. k The attention weight is a Ci =softmax((Q·W) qc )+(f Ci ·W fc Computing power M k The attention weight is a Mi =softmax((Q·W) qm )+(f Mi ·W fm ));

[0131] Task s is obtained by calculating attention weights. k Attention score A of the middle attention group A k Attention score B of attention group B k That is, the attention weights of attention groups A and B are summed separately to obtain the attention score A of attention group A. k Attention score B of attention group B k ;

[0132] in, f Ci = g(Ci), where g is the feature mapping function, f Mi= h(Mi), where h is the feature mapping function, q1 represents the degree of concern for communication quality, q2 represents the degree of concern for computing power, and W qc and W fc These are the learning parameters, where · represents the dot product, and W qm and W fm It is the learning parameter, A k =a Ci f Ci +a Mi f Mi B k =a Ci f Ci +a Mi f Mi ;

[0133] Based on attention score A k Quadrone drones with an attention score below 0.5 are eliminated, and the remaining quadcopters are arranged according to the service order of the service chain to form a new attention group A.

[0134] Based on attention score B k Quadrone drones with an attention score below 0.5 will be eliminated, and the remaining quadcopter drones will be used as service providers, forming a new attention group B.

[0135] The attention score ranges from 0 to 1, and the service provider refers to a quadcopter drone that is capable of providing services and is waiting to provide services.

[0136] Step S33 describes re-subdividing the initial task chain based on the quadcopter drone members in the new attention group to obtain a new serial-parallel hybrid task execution sequence chain, the process of which satisfies:

[0137] Based on the new attention groups A and B, let u′ be the set of quadcopter drones corresponding to the services in service chain F that the new attention group A caches, where u′=[u′1,u′2,…,u′…]. k ,…,u′ K ], u′ k This indicates that service f is cached in the new attention group A. k The number of drones;

[0138] See Figure 3 By utilizing ground users, the initial task chain S is subdivided into tasks based on the quadcopter drone members in the new attention group A, resulting in a new serial-parallel hybrid task execution sequence chain.

[0139] For example, we can use u′ to break down the task chain into tasks, i.e., the k-th task in attention group A has u kIf a quadcopter drone is deployed, then the k-th task is divided into u k A program that can be processed in parallel. For example... Figure 3 As shown, when u2 = 2, task s2 can be split into s 21 and s 22 ; when u k When ≤1, task s k No further subdivision is made.

[0140] In this embodiment, the matching degree between the service chain required by the task chain and the service cache information of the quadcopter is obtained. Based on the service cache matching degree, the quadcopter cluster is grouped using an attention mechanism, and the attention score of each group is calculated. Quadrotors with a score lower than 0.5 are eliminated based on the attention score, thereby updating the quadcopter members of the attention group. The initial task chain is then subdivided to obtain a new serial-parallel hybrid task execution order chain, which maximizes the utilization of quadcopters when executing tasks in the task chain, effectively saving resources for quadcopters and improving the overall performance of the edge network system.

[0141] Example 3:

[0142] In this embodiment, the following explanation is given regarding step S4 of Embodiment 1, which describes the construction and solution of a two-layer UAV airborne edge network model based on grouping and task subdivision, coordinating service caching and task offloading, to obtain the minimum latency of the two-layer UAV airborne edge network system for task offloading and computation processing of the task chain. The specific process is as follows:

[0143] S41: The quadcopter drone in the new attention group A processes the tasks that cache the corresponding services required in task chain S. A collaborative service caching and task offloading strategy ensures that any task in task chain S is processed on the quadcopter drone, fixed-wing drone cloud platform, or ground user server in the new attention group B. The fixed-wing drone serves as the computing cloud platform, i.e., the fixed-wing drone cloud platform. The specific process is as follows:

[0144] The specific process by which the collaborative service caching and task offloading strategy processes any task in task chain S on the quadcopter drone cluster, fixed-wing drone cloud platform, or ground user server in the new attention group B is as follows:

[0145] Let the position of the task in the task chain S be L. k When L k =0 indicates that the task is in the first half of the task chain S, when L k =1 indicates that the task is in the latter half of task chain S;

[0146] Let any task be s k Tasks k The latency limit is: t k ≤τ k Tasks k The computing power requirement is Tasks k Data upload time is Where com represents compute and tran represents transmit;

[0147] Establish a three-dimensional Cartesian coordinate system, and let the coordinate position of the ground user server be (x... l ,y l ,0), the position of the fixed-wing UAV is set to The position of the quadcopter drone is set to Tasks k The previous task k-1 The position is set to (x k-1 ,y k-1 ,z k-1 According to task s k The coordinates of the task k-1 The coordinates are used to calculate the distance D between the destinations of the two tasks executed consecutively. k The distance between the destinations of the two tasks is the distance between the destination of the previous task and the destination of the current task, and the process satisfies:

[0148] If task s k If executed on the ground user server, then the distance to D k =D k，local , If task s k When executed by a fixed-wing UAV, the distance is D k =D k，u0 , If task s k The quadcopter drones operating in the new attention group B are at a distance from D. k =D k，ui , Wherein, the distance D k D represents the distance D between the destinations of the two tasks executed consecutively. k ;

[0149] Among them, s k =(d k ,w k ,τ k ,f k ); d represents the size of the task input data, d k Indicates task s kThe size of the input data; w represents the number of CPU cycles required to complete the task, w k Indicates completion of task s k The number of CPU cycles required, i.e., the computational complexity; τ k f represents the latency tolerance, i.e., the maximum processing latency. k Indicates task s k The corresponding service; w′ k Indicates task s k The number of CPU cycles executed per second by the server, w′ k =M k That is, the number of CPU cycles executed per second by the quadcopter drone is The number of CPU cycles executed per second by a fixed-wing UAV is The number of CPU cycles executed per second by the ground user server is r represents the communication transmission rate of the task, r k Indicates task s k The communication transmission rate is given by: B represents the bandwidth of the communication channel, N represents the noise power of the channel, S represents the signal power of the signal, P represents the transmission power of the device, and G represents the transmission gain of the signal.

[0150] According to task s k Computing power requirements Distance D k And the position of the task in the task chain S L k Collaborative service caching and task unloading determine the tasks s in the task chain S. k The process is performed on the quadcopter or fixed-wing drone cloud platform or ground user server in the new attention group B, and the process satisfies:

[0151] S411: Determine the computing power requirements of the ground user server like This indicates that the ground user server satisfies task s. k Computing power requirements The option to retain ground user servers is retained; if This indicates that the ground user server cannot fulfill task s. k Computing power requirements Remove the option to select ground user servers;

[0152] S412: Determine task s k Position L in task chain S k If L k =0 indicates task s k The first half of task chain S indicates that the quadcopter drone in the new attention group B has not yet downloaded the cached task s from the fixed-wing drone cloud platform.k Required services f k Remove the quadcopter drone selection from the new attention group B; if L k =1, indicating task s k The latter half of task chain S indicates that the quadcopter drone in the new attention group B has downloaded and cached task s from the fixed-wing drone cloud platform. k Required services f k The option of quadcopter drones will be retained in the new attention group B;

[0153] S413: Combine the remaining ground user server (after the judgments in steps S411 and S412) with the quadcopter drone in the new attention group B, and the fixed-wing drone cloud platform to integrate task s. k Computing power requirements and distance D k Select one of the following to perform the task: a ground user server, a quadcopter drone, or a fixed-wing drone cloud platform from the new attention group B. k ;

[0154] Let the task unloading decision factor be: E k E k =w1M k +w2D k , where w1 and w2 are weighting coefficients used to adjust the contribution of different factors to the decision.

[0155] The comprehensive representation computational capability requirement described in step S413 and distance D k Both factors should be considered together. Because of the different tasks, some tasks may require very little computing power but have very high latency requirements; others may require a lot of computing power but are not sensitive to latency. That's why it's necessary to consider both factors together.

[0156] S414: Set the service caching and task offloading strategy to α. k α k =0 indicates task s k Processed on a fixed-wing UAV cloud platform; α k =i, where i = {1, 2, ..., I} represents task s k Processed on quadcopter drones in the new attention group B; α k =I+1 represents task s k Processed on the ground user server; where α k ∈{0}∪N; that is

[0157]

[0158] And when u′ k When >1, that is, task s k The task was broken down into multiple tasks and processed in parallel on multiple new attention groups B quadcopter drones. Service caching and task offloading were implemented. Where, u′ k Indicates task s k The number of parallel tasks that it is split into.

[0159] Let the correlation factor be a, Indicates task s k Whether the processing is performed on a fixed-wing UAV cloud platform, 1 for yes, 0 for no; Indicates task s k Whether to process on the quadcopter drone in the new attention group B, 1 for yes, 0 for no; a k,local Indicates task s k Whether to execute on the ground user server, 1 for yes, 0 for no;

[0160]

[0161] Performing the same task k At the same time, the expression is satisfied:

[0162] For example, firstly, based on the types of services cached by the quadcopter drones in the new attention group A, tasks are assigned to the corresponding task positions in the task chain for processing. After the tasks for the quadcopter drones in the new attention group A are assigned, the remaining tasks in the task chain need to choose one of three destinations: the quadcopter drones in the new attention group B, the fixed-wing drone cloud platform, or the ground user server. The quadcopter drones in the new attention group B are characterized by sufficient computing resources and proximity to the user end, but they do not cache the services required for the tasks (the solution is to download the required services from the fixed-wing drone cloud platform, but caching services from the cloud platform takes a long time). The fixed-wing drone cloud platform is characterized by caching all services required for the service chain and having sufficient resources, but it is far from the user end and has a large transmission latency. The ground user server is characterized by no transmission latency and caching all services required for the service chain, but its computing power is limited.

[0163] Therefore, the evaluation criterion for selecting the destination of the task execution is the computational capability requirement of the task. The distance D between the destinations of the two tasks performed consecutively k And the position of the task in the task chain S L k Computing power requirements The computational complexity w of the task k (i.e., the number of CPU cycles required to complete the computational task) and the server's computing power M k(i.e., when task s) k Executed on the ground user server, with computing power of... When task s k Executed on a fixed-wing UAV cloud platform, with a computing power of [missing information]. When task s k The quadcopter drone performs the task, with computing power of... To determine the distance D between the destinations of the two tasks executed consecutively. k The distance between the location where the previous task was executed and the destination of the current task is executed is the distance between the location where the task is executed and the destination of the current task. The position of a task in the task chain S is divided into the first half and the second half of the task chain S based on the middle task.

[0164] The distance D between the destinations of the two tasks performed consecutively. k To simplify the description of the problem, a three-dimensional Cartesian coordinate system is established. The coordinate position of the ground user server is set as (x... l ,y l The fixed-wing UAV hovers at a fixed altitude h, and its coordinate position is set to 0). The coordinates of the quadcopter drone are set to Tasks k The previous task k-1 The position is set to (x k-1 ,y k-1 ,z k-1 If task s k When executed on a ground-based user server, the transmission distance is... If task s k When executed by a fixed-wing drone, the transmission distance is If task s k When executed by a quadcopter drone, the transmission distance is

[0165] Let task s k The position in task chain S is L. k When L k =0 indicates task s k In the first half of task chain S, when L k =1, indicating task s k In the latter half of task chain S.

[0166] Based on the decision-making method described above, first determine task s k Computing power requirements Does it exceed the computing power of the user's local server? That is, the user side cannot fulfill task s k Computing power requirements If the ground user server selection is not selected, then the selection of the ground user server is retained; then, the task s is determined. k At position S in task chain, L k =0, meaning task s k In the first half of task chain S, remove the quadcopter drone selection from the new attention group B; otherwise, retain the quadcopter drone selection from the new attention group B. Finally, based on task s... k Computing power requirements The distance D between the destinations of the two tasks performed consecutively. k They jointly decided to execute the mission on one of the following: a fixed-wing UAV cloud platform, a quadcopter UAV, or a ground user server. k The task unloading decision factor is E. k =w1M k +w2D k , where w1 and w2 are weighting coefficients used to adjust the contribution of different factors to the decision.

[0167] Let the service caching and task offloading strategy be α. k ∈{0}∪N. α k =0 indicates task s k Processing on a fixed-wing UAV cloud platform; α k =i, where i = {1, 2, ..., I} represents task s k Processed on a quadcopter drone in the new attention group B; α k =I+1 represents task s k Processed on the ground user server.

[0168]

[0169] And when u′ k When >1, that is, task s k The task was broken down into multiple tasks and processed in parallel on multiple new attention groups B quadcopter drones. Service caching and task offloading were implemented. Where u′ k For s k The number of parallel tasks that it is split into.

[0170] Set the correlation factor a: Indicates task s k Whether to process on a fixed-wing UAV cloud platform, 1 for yes, 0 for no. Indicates task s k Whether to process on the quadcopter drone in the new attention group B, 1 for yes, 0 for no. k,local Indicates task s k Whether it is processed on the ground user server, 1 for yes, 0 for no.

[0171]

[0172] Tasks k At any given time, a task can only be executed from one of these three types of execution destinations:

[0173] Furthermore, the specific process of constructing a two-layer UAV aerial edge network model based on the processing latency of the computation task chain S described in step S42, which is processed in the quadcopter UAV cluster, the fixed-wing UAV cloud platform, and the ground user server, is as follows:

[0174] Computational tasks s k The latency processed on a quadcopter drone swarm, i.e., α k When = i, the process satisfies:

[0175] When u′ k =1, task s k Processed on a quadcopter drone, task s k The processing latency consists of task data upload time, task calculation time, and waiting delay time;

[0176] Data upload time:

[0177] Task calculation time:

[0178] Waiting delay time: σ k ;

[0179] Tasks k The delay is:

[0180] When u′ k When >1, task s k The task is divided into multiple parts and processed in parallel on multiple quadcopter drones. k The processing latency is the slowest of all parallel programs, and the process satisfies:

[0181]

[0182]

[0183] Tasks k The delay is:

[0184] Quadrone drone swarm processing tasks k The delay is:

[0185]

[0186] in, For the communication transmission rate between ground users and quadcopter drones, This represents the bandwidth of the communication channel between the ground user and the quadcopter drone, where P represents the transmit power of the ground user. w represents the signal transmission gain between the ground user and the quadcopter drone swarm. k This indicates the completion of the computation task s. k The number of CPU cycles required u′ represents the number of CPU cycles executed per second by a quadcopter drone swarm. k For task s k The number of parallel tasks that it is split into.

[0187] Computational tasks s k The latency processed on the fixed-wing UAV cloud platform, i.e., α k When = 0, the process satisfies:

[0188] When task s k Unloaded to the fixed-wing UAV cloud platform, task s k The processing latency is the task data upload time and the task calculation time;

[0189] Data upload time:

[0190] Task calculation time:

[0191] Task processing on fixed-wing UAVs k Delay:

[0192] in, This indicates the communication transmission rate between ground users and fixed-wing UAVs. This represents the bandwidth of the communication channel between the ground user and the fixed-wing UAV, where P represents the transmit power of the ground user. w represents the signal transmission gain between ground users and fixed-wing UAVs. k This indicates the completion of the computation task s. k The number of CPU cycles required This indicates the number of CPU cycles executed per second by a fixed-wing drone.

[0193] Computational tasks s k The latency processed on the ground user server, i.e., α k When = I+1, the process satisfies:

[0194] Execute tasks on the ground user server k Tasks kThe processing latency consists of task computation time and waiting delay time;

[0195] Task calculation time:

[0196] Waiting delay time:

[0197] Ground user server processing tasks k Delay:

[0198] Among them, w k This indicates the number of CPU cycles required to complete the computation task. This represents the number of CPU cycles executed per second by the ground user server.

[0199] Based on task s k The processing latency of the quadcopter drone swarm, the fixed-wing drone cloud platform, and the ground user server is used to obtain the total processing latency of task chain S as follows:

[0200] Based on the total processing latency of task chain S, the two-layer UAV airborne edge network model is derived as follows:

[0201]

[0202] Considering the constraints of correlation factors and time delay limitations, the particle swarm optimization algorithm is used to solve the two-layer UAV airborne edge network model.

[0203] Considering the correlation factor constraint: at any given time, only one of the following three devices can be selected to execute the same task: the ground user server, the quadcopter UAV, or the fixed-wing UAV.

[0204]

[0205]

[0206] Considering latency constraints: the completion latency of each subtask cannot exceed the latency required by the task itself.

[0207]

[0208] The minimum latency for task unloading and computation in the task chain of the two-layer UAV airborne edge network system is obtained:

[0209] minT.

[0210] To solve an edge network system model using the Particle Swarm Optimization (PSO) algorithm, we first define the objective function and constraints to be optimized, and then initialize the particle swarm. The objective function minimizes the system's latency; therefore, the particle fitness is evaluated by minimizing the fitness value. Then, for each example, the optimal position of the particle is updated based on its current fitness value and the historical best fitness value; this is the optimal solution. The position of the particle with the best fitness value in the entire swarm is selected as the global optimal position, i.e., the global optimal solution. Next, the particle's velocity and position are updated using an appropriate update formula based on its current velocity, position, and optimal position. This process iterates until a termination condition is met. After iteration, the solution vector corresponding to the global optimal position is output as the optimal solution.

[0211] In this embodiment, after the quadcopter drones in the new attention group A process and cache the tasks corresponding to the services required in task chain S, the collaborative service caching and task offloading strategy allows any task in task chain S to be processed on the quadcopter drones, fixed-wing drone cloud platforms, or ground user servers in the new attention group B. Simultaneously, the processing latency of task chain S on the quadcopter drones, fixed-wing drone cloud platforms, and ground user servers in the new attention group B is calculated. Based on the processing latency of task chain S, a two-layer UAV aerial edge network model is constructed and solved using the particle swarm optimization algorithm. The minimum latency for the two-layer UAV aerial edge network system to offload and compute tasks in the task chain is obtained: minT. Through the cooperation between the quadcopter drone cluster, fixed-wing drones, and ground users, the task chain S generated by ground users is effectively offloaded and computed, reducing the burden of task computation and storage in the edge network system and improving the overall performance of the edge network system.

[0212] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping, characterized in that, Includes the following steps: S1: Construct a two-layer UAV airborne edge network system, which includes: a fixed-wing UAV, I quadcopter UAVs and J ground users. The I quadcopter UAVs form a quadcopter UAV cluster, and the fixed-wing UAVs, the quadcopter UAV cluster and the ground users establish communication connections with each other. S2: The fixed-wing UAV and the quadcopter UAV cluster work together to unload and process the task chain generated by the ground user. The task in the task chain is a computing task. The unloading process means unloading the computing task to the quadcopter UAV cluster or the fixed-wing UAV for computing. S3: With the goal of minimizing the latency of the two-layer UAV aerial edge network system in processing tasks, attention mechanism is used to group the quadcopter drone swarm and further subdivide the tasks based on the grouping. S4: Based on grouping and task subdivision, coordinate service caching and task offloading, construct and solve a two-layer UAV air edge network model, and obtain the minimum latency of the two-layer UAV air edge network system for task offloading and computation processing of the task chain. Wherein, UAV refers to unmanned aerial vehicle (UAV), and dual-layer UAV refers to a fixed-wing UAV layer and a quadcopter UAV swarm layer; The specific process of grouping the quadcopter drone swarm using an attention mechanism and further subdividing the tasks based on the grouping in step S3 is as follows: S31: Obtain the service cache information of the service chain and the quadcopter drone required by the task chain, and obtain the service cache matching degree between the task chain and the quadcopter drone based on the service chain and the service cache information. S32: Based on the service cache matching degree between the quadcopter drones and the task chain, the quadcopter drone cluster is grouped using an attention mechanism; S33: Calculate the attention score of each group of quadcopter drones, update the quadcopter drone members in the attention group based on the attention score, and re-subdivide the initial task chain according to the new quadcopter drone members in the attention group to obtain a new serial-parallel hybrid task execution order chain. The specific process of constructing and solving the two-layer UAV airborne edge network model based on grouping and task subdivision, and coordinating service caching and task offloading, as described in step S4, is as follows: S41: Utilizing the quadcopter drones in the new attention group A to process cached task chains. S The tasks corresponding to the required services, the collaborative service caching and task unloading strategies will link the task chain. S Any task in the process is processed on a quadcopter drone or a fixed-wing drone cloud platform or a ground user server in the new attention group B; wherein the fixed-wing drone serves as the computing cloud platform, i.e., the fixed-wing drone cloud platform. S42: Computational Task Chain S Latency processing in the new Attention Group B, including quadcopter drones, fixed-wing drone cloud platforms, and ground user servers, is based on task chains. S To address processing latency, a two-layer UAV aerial edge network model was constructed and solved.

2. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 1, characterized in that, The specific process of obtaining the service cache information of the service chain and the quadcopter drone required for the task chain in step S31, and obtaining the service cache matching degree between the task chain and the quadcopter drone based on the service chain and the service cache information, is as follows: Set up a task chain initiated by a ground user S The task chain S The required service chain is F Let the set of quadcopter drone swarms be . UAV The service cache information for quadcopter drones is U Let the service cache matching degree between the quadcopter UAV and the task chain be . ; ； in, , This indicates that the current task chain contains One task, Indicates the first in the task chain One task, , Indicates task Required services; , The number of quadcopter drones, , Indicates the first All services cached by the quadcopter drone , express The number of cached services Indicates the first A quadcopter drone.

3. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 2, characterized in that, The specific process of grouping the quadcopter drone cluster using an attention mechanism based on the service cache matching degree between the quadcopter drone and the task chain, as described in step S32, is as follows: The service is cached. The set of the number of quadcopter drones is The service chain is cached. F The set of the number of quadcopter drones corresponding to the services in the middle is Determine the set The size of the variables in the set depends on the set. The variable size in the data divides the quadcopter drone swarm into attention group A and attention group B, the set The size of the variable in the set represents the size of the set. u Does the service chain have a quadcopter drone cache? F The corresponding service in the process satisfies: If set If the variable in the cache is greater than or equal to 1, it indicates that one or more quadcopter drones are caching the service simultaneously. Let's call this attention group A; If set If the variable in the value is less than 1, it indicates that there is no quadcopter drone cache service chain. F The service in this group is denoted as Attention Group B; in, , This indicates that the service is cached. The number of quadcopter drones, .

4. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 3, characterized in that, Step S33, which involves calculating the attention score of each quadcopter drone group and updating the quadcopter drone members in the attention group based on the attention score, is as follows: Set a task Communication quality between attention group A and attention group B and computing power According to communication quality and computing power The specific process for calculating the attention scores of attention group A and attention group B is as follows: Communication quality and computing power Transform into features and characteristics This indicates that features are used. and characteristics The attention weights are calculated, and the process satisfies the following: Let the query vector be Additive attention is used to calculate attention weights, and a softmax function is added to normalize the obtained attention weights to obtain the communication quality. The attention weights are computing power The attention weights are ; Task is obtained by calculating attention weights Attention scores of the middle attention group A Attention scores of attention group B ; in, , , , For feature mapping function, , For feature mapping function, This indicates the level of concern regarding communication quality. This indicates the level of attention paid to computing power. and These are learning parameters. This represents the dot product. and These are learning parameters. , ; Based on attention score Quadrone drones with an attention score below 0.5 are eliminated, and the remaining quadcopters are arranged according to the service order of the service chain to form a new attention group A. Based on attention score Quadrone drones with an attention score below 0.5 will be eliminated, and the remaining quadcopter drones will be used as service providers, forming a new attention group B. The attention score ranges from 0 to 1, and the service provider refers to a quadcopter drone that is capable of providing services and is waiting to provide services.

5. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 4, characterized in that, The specific process described in step S33, which involves re-subdividing the initial task chain based on the quadcopter drone members in the new attention group to obtain a new serial-parallel hybrid task execution sequence chain, is as follows: Based on the new attention group A and attention group B, assume that the new attention group A caches the service chain. F The set of the number of quadcopter drones corresponding to the services in the middle is ,in, , This indicates that the service is cached in the new attention group A. The number of drones; Utilize ground users to initiate the mission chain S Based on the task subdivision of the quadcopter drone members in the new attention group A, a new serial-parallel hybrid task execution sequence chain is obtained.

6. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 5, characterized in that, The collaborative service caching and task offloading strategy described in step S41 will link the task chain. S The specific process by which any task in the new attention group B is processed in the quadcopter drone swarm, fixed-wing drone cloud platform, or ground user server is as follows: Assume the task is in the task chain S The position in the middle is ;when This indicates that the task is in the task chain. S The first half, when This indicates that the task is in the task chain. S The latter half; Let any task be ,Task The latency limit is: ,Task The computing power requirement is ,Task Data upload time is Where com represents compute and tran represents transmit; Establish a three-dimensional Cartesian coordinate system, and let the coordinate position of the ground user server be... The position of the fixed-wing UAV is set as The position of the quadcopter drone is set as ,Task Previous task Set the position According to the task Coordinates and task The coordinates are used to calculate the distance between the destinations of the two tasks executed consecutively. The distance between the destinations of the two tasks is the distance between the destination of the previous task and the destination of the current task, and the process satisfies: If task Executed on the ground user server, then the distance If the task When executed by a fixed-wing drone, the distance... If the task The quadcopter drones deployed in the new attention group B are at a distance Wherein, the distance This indicates the distance between the destinations of two consecutive tasks. ; According to the task Computing power requirements ,distance and tasks In the task chain S The position in the middle Collaborative service caching and task unloading determine the task chain. S Tasks in Processing is performed on the cloud platform or ground user server of the quadcopter or fixed-wing drone in the new attention group B; in, ; Indicates the size of the task input data. Indicates task The size of the input data; This indicates the number of CPU cycles required to complete the task. Indicates completion of task The number of CPU cycles required, i.e., the computational complexity; This indicates the latency tolerance, i.e., the maximum processing latency. Indicates task The corresponding services; Indicates task The number of CPU cycles executed per second by the server. = The number of CPU cycles executed per second by the quadcopter drone is The number of CPU cycles executed per second by a fixed-wing UAV is The number of CPU cycles executed per second by the ground user server is ; , Indicates the communication transmission rate of the task. Indicates task The communication transmission rate is given by: B represents the bandwidth of the communication channel, N represents the noise power of the channel, S represents the signal power of the signal, P represents the transmission power of the device, and G represents the transmission gain of the signal.

7. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 6, characterized in that, According to the task Computing power requirements ,distance And tasks in the task chain S The position in the middle Collaborative service caching and task unloading determine the task chain. S Tasks in The specific process for processing on the quadcopter or fixed-wing drone cloud platform or ground user server in the new attention group B is as follows: S411: Determine the computing power requirements of the ground user server ,like This indicates that the ground user server has fulfilled the task. Computing power requirements Retain the option of ground user servers; if This indicates that the ground user server cannot fulfill the task. Computing power requirements Remove the option to select a ground user server; S412: Determine the task In the task chain S The position in the middle ,like , indicating task In the task chain S The first part indicates that the quadcopter drones in the new attention group B have not yet downloaded the cached tasks from the fixed-wing drone cloud platform. Required services Remove the quadcopter drone option from the new attention group B; if , indicating task In the task chain S The latter part indicates that the quadcopter drone in the new attention group B has downloaded and cached the task from the fixed-wing drone cloud platform. Required services The option of quadcopter drones will be retained in the new attention group B; S413: Combine the remaining ground user server (after the judgments in steps S411 and S412) with the quadcopter drone in the new attention group B, and integrate them with the fixed-wing drone cloud platform to conduct a comprehensive mission. Computing power requirements and distance Choose one of the following to perform the task: a ground user server, a quadcopter drone, or a fixed-wing drone cloud platform from the new attention group B. ; S414: Set the collaborative service caching and task offloading policy as follows: , Indicates task Processed on a fixed-wing UAV cloud platform; ,in Indicates task Processed on a quadcopter drone; Indicates task Processed on the ground user server; among which, ; Let the correlation factor be , Indicates task Whether to process on a fixed-wing UAV, 1 for yes, 0 for no; Indicates task Whether to process on a quadcopter drone, 1 for yes, 0 for no; Indicates task Whether to execute on the ground user server, 1 for yes, 0 for no; ； Performing the same task At the same time, the expression is satisfied: .

8. The method for caching and offloading unmanned aerial vehicle (UAV) assisted services based on attention grouping according to claim 7, characterized in that, The computational task chain described in step S42 S Latency processing in quadcopter drone swarms, fixed-wing drone cloud platforms, and ground user servers is based on task chains. S To address processing latency, the specific process for constructing a two-layer UAV aerial edge network model is as follows: Computational tasks The latency processed on a quadcopter drone swarm satisfies the following conditions: when ,Task Processed on a quadcopter drone, the task The processing latency consists of task data upload time, task calculation time, and waiting delay time; Data upload time: ; Task calculation time: ; Waiting delay time: ; Task The delay is: ; when At that time, the task The task is divided into multiple parts and processed in parallel on multiple quadcopter drones. The processing latency is the slowest of all parallel programs, and the process satisfies: ； ； Task The delay is: ; Quadrone swarm processing task The delay is: ； in, For the communication transmission rate between ground users and quadcopter drones, , This represents the bandwidth of the communication channel between the ground user and the quadcopter drone, where P represents the transmit power of the ground user. This indicates the signal transmission gain between the ground user and the quadcopter drone swarm. Indicates completion of the calculation task The number of CPU cycles required This indicates the number of CPU cycles executed per second by the quadcopter drone swarm. For the task The number of parallel tasks it is split into; Computational tasks The latency processed on the fixed-wing UAV cloud platform satisfies the following conditions: When the task Unload to the fixed-wing UAV cloud platform, mission The processing latency is the task data upload time and the task calculation time; Data upload time: ; Task calculation time: ; Task processing on fixed-wing drones Delay: ; in, This indicates the communication transmission rate between ground users and fixed-wing UAVs. , This represents the bandwidth of the communication channel between the ground user and the fixed-wing UAV, where P represents the transmit power of the ground user. This indicates the signal transmission gain between ground users and fixed-wing UAVs. Indicates completion of the calculation task The number of CPU cycles required This indicates the number of CPU cycles executed per second by a fixed-wing UAV. Computational tasks The latency processed on the ground user server satisfies the following process: Execute tasks on the ground user server ,Task The processing latency consists of task computation time and waiting delay time; Task calculation time: ; Waiting delay time: ; Ground user server processing tasks Delay: ; in, This indicates the number of CPU cycles required to complete the computation task. This represents the number of CPU cycles executed per second by the ground user server. Based on task The processing latency of the quadcopter drone swarm, the fixed-wing drone cloud platform, and the ground user server is used to obtain the total processing latency of task chain S as follows: ; Based on the total processing latency of task chain S, the two-layer UAV airborne edge network model is derived as follows: ； The particle swarm optimization algorithm is used to solve the two-layer UAV airborne edge network model, and the minimum latency for task offloading and computation of the two-layer UAV airborne edge network system is obtained: 。

Images