A multi-UAV optimal edge service method and system for mobile vehicles

Abstract

The present invention discloses a multi-UAV optimal edge service method and system for a mobile vehicle. The method includes that within the research scope, there are multiple UAVs equipped with mobile edge computing MEC to provide edge computing services for the mobile vehicle, and there is also a base station to provide additional computing services for the UAVs. Each mobile vehicle carries only one task, and the task type is determined by priority: routine or emergency. A routine task completes task unloading and calculation within the coverage of its initial UAV. If the task calculation is not completed in the initial UAV, the task is secondary unloaded between UAVs as the vehicle's driving area changes. When a vehicle carrying an emergency task appears, all routine tasks in the initial UAV are secondary unloaded to the base station BS to complete subsequent calculations. The initial UAV follows the vehicle carrying the emergency task and returns to its original position after completing the task calculation.

Description

Technical Field

[0001] The present invention relates to the field of drone technology, and in particular to a multi-drone optimal edge service method and system for a mobile vehicle. Background Art

[0002] Due to the excellent deployment flexibility of drones, drone edge computing aims to use drones as edge computing nodes to provide transmission, computing, and data caching services to ground equipment. This can significantly improve computing capabilities in areas with weak infrastructure and has applications in scenarios such as disaster relief and military communications. Ground mobile devices can offload computing tasks to drone edge nodes on demand. After completing the computing tasks, the drone edge nodes return the results to ground equipment, thereby saving energy consumption on ground equipment and improving the timeliness of computing services. To improve service coverage and service quality, multiple drone edge nodes jointly providing services for specific mission areas has become a viable new computing paradigm. In this scenario, how to efficiently utilize the limited computing resources and battery supply of drone platforms is a key issue to consider in the field of drone edge computing.

[0003] At present, the industry has carried out some work on optimization scheduling in the field of drone edge computing. For example, the invention patent application with application number 202211168983.0, "A task offloading scheduling method for energy efficiency optimization in drone edge computing networks", provides a task offloading scheduling method for energy efficiency optimization in drone edge computing networks, which minimizes the energy consumption of IoT mobile devices by jointly optimizing offloading decisions, task scheduling sequence, transmission bit allocation, and drone trajectory; the Chinese invention patent application with application number 202210268185.9, "A linearly dependent task offloading method for drone edge computing networks", can effectively obtain the optimal value of energy consumption by jointly optimizing offloading decisions, resource allocation, and drone trajectory, thereby reducing equipment energy consumption; the Chinese invention patent application with application number 20221106470.7, "A drone deployment and task offloading method in a multi-drone edge computing network", maximizes the number of tasks carried by drones by jointly optimizing task offloading decisions, resource allocation, and drone position and elevation angle;

[0004] However, current work still has the following shortcomings: First, it fails to jointly consider multi-UAV trajectory planning, MEC resource scheduling, and mobility task offloading. Second, most approaches fail to consider the fact that traditional optimization algorithms cannot adapt to dynamically changing environments and cannot achieve long-term optimization throughout the service lifecycle. Third, when faced with multi-UAV resource scheduling systems, they still rely on single-agent deep learning algorithms, which suffer from low efficiency and poor performance. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention proposes an optimal edge service solution for multiple drones of mobile vehicles, which takes into account the impact of task requests and mobile tasks on dynamic edge computing services. Taking the amount of repeated task offloading and the energy consumption cost of repeated offloading as indicators, the drone edge service is modeled as a multi-agent Markov dynamic decision-making process by using multi-agent deep Q learning to achieve the optimal edge server resource allocation strategy.

[0006] In a first aspect, the present invention provides a multi-UAV optimal edge service method for a mobile vehicle. Within the research scope, there are multiple UAVs equipped with mobile edge computing (MEC) to provide edge computing services for the mobile vehicles, and there is also a base station to provide additional computing services for the UAVs; each of the mobile vehicles carries only one task, and the tasks are divided into two types according to priority: routine and emergency;

[0007] The method comprises:

[0008] For any mobile vehicle, first determine the type of mission it carries:

[0009] If it is a routine task, the mobile vehicle will offload the routine task it carries to the corresponding drone within the current drone service range for calculation. When the mobile vehicle moves to the next drone service range, the current drone will offload the unfinished routine task to the corresponding drone within the next drone service range to continue calculation until the mobile vehicle completes the calculation of all the routine tasks it carries.

[0010] If it is an emergency task, the corresponding drone within the drone service range of the mobile vehicle will unload all the routine tasks it undertakes to the base station, and the mobile vehicle will unload the emergency task it carries to the current drone for calculation. At the same time, the current drone will follow the mobile vehicle until the calculation of the emergency task is completed and then return to its original position.

[0011] Furthermore, if there is more than one mobile vehicle carrying an emergency task within the service range of a certain drone, after the drone undertakes one emergency task, it dispatches the remaining drones that are not currently undertaking an emergency task to provide edge computing services for the remaining emergency tasks.

[0012] Furthermore, the tasks are divided into two types according to priority: routine and urgent, including:

[0013] When the priority of task a is Y a ≥Y th When , task a is an urgent task, otherwise task a is a regular task; where a is the task number, k is the elasticity function coefficient, x a is the normalized importance index of task a, τ ais the completion time of task a.

[0014] Furthermore, the method further comprises:

[0015] Taking all mobile vehicles within the service range of each drone as a cluster and minimizing the global secondary offloading task volume of each drone as the objective function, we construct an intra-cluster resource allocation optimization problem for each cluster. Secondary offloading includes the drone offloading its tasks to other drones, and the drone offloading its tasks to the base station.

[0016] Using drones as intelligent agents, a multi-agent deep Q-learning algorithm is used to find the optimal solution to the intra-cluster resource allocation optimization problem of each cluster. The obtained optimal solution is the optimal resource allocation strategy for each drone, achieving optimal edge service for multiple drones.

[0017] Furthermore, the expression of the intra-cluster resource allocation optimization problem of each cluster is:

[0018]

[0019] stC1:d im ≤R

[0020] C2:ξ≤α

[0021]

[0022] in, is the global secondary offloading task volume of UAV m, m is the UAV number, R m s1 is the amount of unfinished routine tasks that UAV m offloads to other UAVs, R m s2 is the amount of routine tasks that UAV m unloads to the base station to complete the emergency task; in the constraint condition C1, d im is the straight-line distance between the vertical projection point of UAV m on the ground and the mobile vehicle carrying routine task i, i is the routine task number, R is the service range of UAV m; in constraint C2, ξ is the minimum task completion rate, α is the task completion rate of UAV m; in constraint C3, R i,m is the amount of data transmitted from routine mission i to drone m, D m,max is the maximum mission data transmission volume that UAV m can support; in constraint C4, is the computing capability of drone m for routine task i, is the maximum computing capacity of the MEC carried by UAV m, and U represents the set of UAVs; in constraint C5, f i bs is the computing capability of the base station BS for routine task i, is the maximum computing capability of the base station BS for routine task i, and A1 represents the set of routine tasks.

[0023] Furthermore, the amount of unfinished tasks R that drone m offloads to other drones m s1 The expression is as follows:

[0024]

[0025] (x) + =max{0,x}

[0026] Where A1 is the set of regular tasks, T mov = R / V is the driving time of a mobile vehicle carrying routine mission i within the service range of UAV m, V is the speed of a mobile vehicle carrying routine mission i, B is the bandwidth of the transmission channel between the mobile vehicle and the UAV, is the time taken by the mobile vehicle to unload the routine task i to the drone m, is the power fading of the transmission channel between a mobile vehicle carrying a conventional mission i and a UAV m, L0 is the reference distance, Γ dB is the additional loss caused by the environment, is the distance between the mobile vehicle carrying routine mission i and the UAV m, H is the flight height of the UAV, d im is the lateral distance between the mobile vehicle carrying conventional mission i and the UAV m, p is the signal power, σ 2 is the noise power.

[0027] Furthermore, the amount of routine tasks R that the drone m unloads to the base station to complete the emergency task is m s2 The expression is as follows:

[0028]

[0029] (x) + =max{0,x}

[0030] Where A1 is the set of regular tasks, j is the regular task number, R j,m is the amount of data transmitted from routine task j to UAV m, T j,m c f is the time that UAV m has provided computing for routine task j when the emergency task occurs, j,m uav is the computing capability of UAV m for routine task j.

[0031] Furthermore, using drones as intelligent agents, a multi-agent deep Q-learning algorithm is used to find the optimal solution to the intra-cluster resource allocation optimization problem for each cluster, including:

[0032] Taking drones as intelligent agents, the optimal resource allocation strategy for each intelligent agent is found through the multi-intelligence deep Q learning algorithm, where the state parameter s of the deep Q learning algorithm is {D i ,R i,m ,R m s}, the action parameter a is the resource V allocated by drone m to regular task i i ={ω i 、l i 、e i}, reward function ε is the proportional coefficient, D i ={x i ,τ i}, x i is the normalized importance index of regular task i, τ i is the completion time limit of regular task i, ω i is the number of CPU cycles per bit allocated by the MEC onboard UAV m for routine task i, l i is the capacitance coefficient assigned by the MEC onboard UAV m for routine task i, e i is the local processing energy consumption allocated by the MEC onboard UAV m for routine task i.

[0033] In a second aspect, the present invention provides a multi-UAV optimal edge service system for mobile vehicles, the system comprising a number of mobile vehicles carrying tasks, a number of UAVs equipped with MEC, and a base station;

[0034] Based on the multi-UAV optimal edge service method for mobile vehicles as described above, the UAV provides edge computing services for the mobile vehicle, and the base station provides additional computing services for the UAV.

[0035] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions that, when executed by a computing device, enable the computing device to perform the multi-UAV optimal edge service method for a mobile vehicle as described above.

[0036] In a fourth aspect, the present invention provides an electronic device comprising one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for executing the multi-UAV optimal edge service method for a mobile vehicle as described above.

[0037] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0038] (1) This solution models the resource scheduling of the MEC edge server when processing tasks for mobile vehicle users as an optimization problem of the secondary offloading task transmission volume.

[0039] (2) This plan takes into account the priorities of different tasks and gives priority to emergency vehicle tasks.

[0040] (3) This scheme seeks to minimize the amount of secondary offloading tasks while ensuring the completion of tasks under each cluster, and solves the minimum secondary offloading energy consumption.

[0041] (4) This solution aims to solve the optimization problem by developing a multi-agent deep Q-learning algorithm based on deep learning, where each cluster of drones is regarded as an agent. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the overall layout diagram of the system where the drone and user vehicles are located.

[0043] Figure 2 Diagram of the secondary offloading model for tasks between UAVs.

[0044] Figure 3 Model diagram for secondary unloading of drone missions to base stations.

[0045] Figure 4 Flowchart of the multi-agent deep learning algorithm. DETAILED DESCRIPTION

[0046] In order to further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0047] The present invention provides a multi-drone optimal edge service system for mobile vehicles, which includes several mobile vehicles carrying tasks, several drones equipped with MEC, and a base station. Multiple drones equipped with MEC provide edge computing for the tasks carried by the mobile vehicles. Each vehicle carries only one task, each task has a different priority, and there may be multiple emergency tasks among them. Each drone is modeled as a problem of minimizing the amount of tasks in the secondary unloading process for multiple tasks within its coverage area, and deep Q learning is designed to find the best resource allocation scheme in different MECs. When an emergency task occurs, all tasks of the drone are unloaded to the base station BS to provide the best MEC resources for the emergency task to complete the task. It is modeled as a global energy consumption minimization problem, and the global minimum energy consumption is found through the gradient fast descent algorithm.

[0048] The present invention provides a multi-UAV optimal edge service method for a mobile vehicle. By seeking the optimal system edge server resource allocation scheme, the method can ensure that the secondary offloading task transmission volume is minimized under the mobile edge services of different task types, and obtain the minimum user task secondary offloading energy consumption.

[0049] like Figure 1 As shown in Figure 1, within a range D*D, there are M unmanned aerial vehicles (UAVs) equipped with MECs, providing task edge services to ground user vehicles. A base station (BS) at the center of the range provides additional computing services to the UAVs. The M UAVs are represented by the set U = {1, 2, ..., M}. The mobile vehicles have a total of tasks A = {A1, A2} to perform: routine tasks A1 = {1, 2, ..., a1} and emergency tasks A2 = {1, 2, ..., a2}.

[0050] Define task attributes D a ={x a ,τ a}, x a is the normalized importance index of task a, τ a is the completion time limit of task a. When a task appears, calculate the priority Y of each task a :

[0051]

[0052] Where k is the elasticity function coefficient, which is a fixed constant. Define the priority threshold Y th , Y a ≥Y th When , task a is judged to be an urgent task.

[0053] like Figure 2 As shown, a mobile vehicle A carrying a regular task i 1i After entering the service range of UAV m, it starts to transmit mission data with UAV m. 1i When the drone reaches the service range boundary of drone n at normal speed, the routine task i has not been fully processed by drone m. At this time, drone m unloads the remaining amount of routine task i to drone n for the second time and continues the calculation and processing.

[0054] Within the service range of drone m, perform task offloading and task calculation for routine task i, specifically including:

[0055] Step a) Define the signal power p and noise power σ 2 , the sub-channel bandwidth is B, both are fixed values. The amount of data transmitted between routine task i and drone m is R i,m :

[0056]

[0057] Define the UAV flight altitude as a fixed value H, and the mission vehicle A 1i The horizontal distance is d im , then the distance between the two is You can get A 1i with U m Transmission channel power fading h i,m (t) are as follows:

[0058]

[0059] Where L0 is the reference distance, which is set to 1 meter here. dB It is the additional loss caused by the environment.

[0060] Step b) Define the computing power of drone m for routine task i as Therefore, the computation time of conventional task i on UAV m is:

[0061]

[0062] And the drone coverage is R, so the mission vehicle A 1i The driving time within the service range of drone m is: T mov =R / V.

[0063] like Mission Vehicle A 1i The conventional task i carried by the drone m completes all computing tasks.

[0064] like Mission Vehicle A 1i The conventional task i carried cannot complete all the computing tasks on the drone m, and the mission vehicle A 1i Driving at a constant speed from the service range of UAV m to the service range of UAV n, UAV m communicates with UAV n and completes the unloading of unfinished tasks, and the unfinished tasks are completed by UAV n until the task vehicle A 1i All the regular tasks carried out have been completed.

[0065] Mission Vehicle A 1i The calculation task volume of the conventional task i carried within the service range of UAV m is The remaining unfinished data amount is

[0066]

[0067] The unfinished tasks are unloaded to UAV m for subsequent calculations. Since this part of the task volume undergoes more than one unloading, the unloading process is called a secondary unloading process.

[0068] In summary, in the secondary unloading of “UAV to UAV”, the total remaining task volume of the global UAV is

[0069]

[0070] (x) + =max{0,x}.

[0071] like Figure 3 As shown, a mobile vehicle A carrying an emergency task j 2j , after entering the service range of UAV m, UAV m will interrupt the routine tasks it is currently processing and transfer all of them to the base station BS for subsequent task processing. This process is called secondary unloading of tasks from UAV to base station. 2j Unload the emergency mission j to UAV m, which follows A. 2j Fly until all calculations of the emergency task j are completed and then fly back to the original position.

[0072] When an emergency task occurs within the service range of UAV m, UAV m offloads all the original tasks to the base station BS, including:

[0073] Assume that when emergency task k in the emergency task set A2 appears in the service range of drone m, drone m is unloading and calculating routine task j, and its calculated time is Then the remaining task volume of conventional task j in UAV m is

[0074]

[0075] In summary, in the secondary unloading from drone to base station, the total remaining task volume of the global drone is

[0076]

[0077] It should be noted here that if more than one emergency task in the emergency task set A2 appears within the service range of drone m at the same time, the additional emergency tasks are communicated and calculated by dispatching other drones around drone m to fly to the mobile vehicle carrying the additional emergency tasks. The flight energy consumption of the drones in this scheduling process is negligible.

[0078] In the secondary unloading of "UAV to UAV" and "UAV to BS" defined in this invention, the secondary unloading energy consumption ρ is defined for UAV m:

[0079]

[0080] in is the secondary offloading energy consumption required for UAV m to offload routine tasks to UAV n, is the secondary offloading energy consumption required for UAV m to offload routine tasks to base station BS.

[0081] Formulas for each part of secondary unloading energy consumption are explained:

[0082]

[0083] where p m,n is the information transmission power from UAV m to UAV n, p m,bs is the information transmission power from UAV m to base station BS, is the time required for UAV m to offload routine tasks to UAV n, is the time required for UAV m to offload routine tasks to base station BS.

[0084] It should be noted here that the secondary unloading energy consumption defined in the present invention only considers the energy consumption generated by the drone secondary unloading unfinished tasks to other drones and base station BS. The rest of the energy consumption in this process is not within the scope of consideration of the present invention.

[0085] In this paper, all mobile vehicles within the drone service range are referred to as a cluster, and local resource allocation modeling of the cluster is implemented. Intra-cluster resource allocation is modeled as a global secondary offloading task minimization problem while ensuring a certain task completion rate. The specific modeling is as follows:

[0086]

[0087] stC1:d im ≤R

[0088] C2:ξ≤α

[0089]

[0090] in, is the global secondary offloading task volume of UAV m, m is the UAV number, R m s1 is the amount of unfinished routine tasks that UAV m offloads to other UAVs, R m s2 is the amount of routine tasks that UAV m unloads to the base station to complete the emergency task. In the constraint condition C1, d imis the straight-line distance between the vertical projection point of UAV m on the ground and the mobile vehicle carrying routine task i, i is the routine task number, R is the service range of UAV m; in constraint condition C2, ξ is the minimum task completion rate, α is the task completion rate of UAV m; in constraint condition C3, R i,m is the amount of data transmitted from routine mission i to drone m, D m,max is the maximum mission data transmission volume that UAV m can support; in constraint C4, is the computing capability of drone m for routine task i, is the maximum computing capacity of the MEC carried by UAV m, and U represents the set of UAVs; in constraint C5, f i bs is the computing capability of the base station BS for routine task i, is the maximum computing capability of the base station BS for routine task i, and A1 represents the set of routine tasks.

[0091] For UAV m, constraint C1 ensures that the distance between the UAV and the ground mission allows normal communication, constraint C2 ensures the mission completion rate, constraint C3 ensures that the communication capacity is within the maximum communication capacity of the UAV, and constraints C4 and C5 respectively constrain the computing capabilities of the UAV and the base station.

[0092] like Figure 4 As shown in the figure, in the present invention, drones are used as intelligent agents, and a multi-agent deep Q learning algorithm is used to find the optimal solution to the intra-cluster resource allocation optimization problem of each cluster. The obtained optimal solution is the optimal resource allocation strategy for each drone, realizing the optimal edge service of multiple drones.

[0093] Furthermore, the state parameter s of the multi-agent deep Q learning algorithm is s={D i ,R i,m ,R m s}, the action parameter a is the resource V allocated by drone m to regular task i i ={ω i 、l i 、e i}, reward function ε is the proportional coefficient, D i ={x i ,τ i}, x i is the normalized importance index of regular task i, τ i is the completion time limit of regular task i, ω i is the number of CPU cycles per bit allocated by the MEC onboard UAV m for routine task i, l i is the capacitance coefficient assigned by the MEC onboard UAV m for routine task i, ei is the local processing energy consumption allocated by the MEC onboard UAV m for routine task i.

[0094] Agent m observes the environment Then make an action And through the reward and punishment value Affects the next environmental observation value And so on, we can obtain the experience trajectory of agent m:

[0095]

[0096] According to the experience trajectory of agent m, after z iterations, the resource V allocated to task i by the agent is i ={ω i 、l i 、e i}, i∈A1 will tend to be stable, and the resource will be the optimal resource allocation result for task i under agent m.

[0097] The present invention also designs a multi-UAV optimal edge service system for mobile vehicles, comprising:

[0098] Several mobile vehicles carrying tasks, several drones equipped with MEC, and a base station;

[0099] The drone provides edge computing services for the mobile vehicle, and the base station provides additional computing services for the drone.

[0100] The technical solution of the above-mentioned edge service system is similar to the above-mentioned method and will not be repeated here.

[0101] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute the above-mentioned multi-UAV optimal edge service method for mobile vehicles.

[0102] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned multi-UAV optimal edge service method for mobile vehicles.

[0103] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0105] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0106] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0107] The above embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A multi-UAV optimal edge service method for mobile vehicles, characterized in that: Within the research scope, there are multiple drones equipped with mobile edge computing (MEC) to provide edge computing services for mobile vehicles, and there is also a base station that provides additional computing services for drones. Each of the mobile vehicles carries only one mission, and the missions are divided into two types: routine and emergency according to priority; The method comprises: For any mobile vehicle, first determine the type of mission it carries: If it is a routine task, the mobile vehicle will offload the routine task it carries to the corresponding drone within the current drone service range for calculation. When the mobile vehicle moves to the next drone service range, the current drone will offload the unfinished routine task to the corresponding drone within the next drone service range to continue calculation until the mobile vehicle completes the calculation of all the routine tasks it carries. If it is an emergency task, the corresponding drone within the drone service range of the mobile vehicle will unload all the routine tasks it undertakes to the base station, and the mobile vehicle will unload the emergency task it carries to the current drone for calculation. At the same time, the current drone will follow the mobile vehicle until the calculation of the emergency task is completed and then return to its original position.

2. The multi-UAV optimal edge service method for a mobile vehicle according to claim 1, characterized in that: If there is more than one mobile vehicle carrying an emergency mission within the service range of a drone, after the drone undertakes one emergency mission, it dispatches the remaining drones that are not currently undertaking emergency missions to provide edge computing services for the remaining emergency missions.

3. The multi-UAV optimal edge service method for a mobile vehicle according to claim 1, characterized in that: The tasks are divided into two types according to priority: routine and urgent, including: When the priority of task a is Y a ≥Y th When , task a is an urgent task, otherwise task a is a regular task; where a is the task number, x a ∈[0,1], k>0, k is the elasticity function coefficient, x a is the normalized importance index of task a, τ a is the completion time of task a.

4. The multi-UAV optimal edge service method for a mobile vehicle according to claim 1, characterized in that: The method further comprises: Taking all mobile vehicles within the service range of each drone as a cluster and minimizing the global secondary offloading task volume of each drone as the objective function, we construct an intra-cluster resource allocation optimization problem for each cluster. Secondary offloading includes the drone offloading its tasks to other drones, and the drone offloading its tasks to the base station. Using drones as intelligent agents, a multi-agent deep Q-learning algorithm is used to find the optimal solution to the intra-cluster resource allocation optimization problem of each cluster. The obtained optimal solution is the optimal resource allocation strategy for each drone, achieving optimal edge service for multiple drones.

5. The multi-UAV optimal edge service method for a mobile vehicle according to claim 4, characterized in that: The expression of the intra-cluster resource allocation optimization problem of each cluster is: s.t.C1:d im ≤R C2:ξ≤α in, is the global secondary offloading task volume of UAV m, m is the UAV number, R m s1 is the amount of unfinished routine tasks that UAV m offloads to other UAVs, R m s2 is the amount of routine tasks that UAV m unloads to the base station to complete the emergency task; in the constraint condition C1, d im is the straight-line distance between the vertical projection point of UAV m on the ground and the mobile vehicle carrying routine task i, i is the routine task number, R is the service range of UAV m; in constraint condition C2, ξ is the minimum task completion rate, α is the task completion rate of UAV m; in constraint condition C3, R i,m is the amount of data transmitted from routine task i to drone m, D m,max is the maximum mission data transmission volume that UAV m can support; in constraint C4, is the computing capability of drone m for routine task i, is the maximum computing capacity of the MEC carried by UAV m, and U represents the set of UAVs; in constraint C5, f i bs is the computing capability of the base station BS for routine task i, is the maximum computing capability of the base station BS for routine task i, and A1 represents the set of routine tasks.

6. The multi-UAV optimal edge service method for a mobile vehicle according to claim 5, characterized in that: The amount of unfinished tasks R that drone m offloads to other drones m s1 The expression is as follows: (x) + =max{0,x} Where A1 is the set of regular tasks, T mov = R / V is the driving time of a mobile vehicle carrying routine mission i within the service range of UAV m, V is the speed of a mobile vehicle carrying routine mission i, B is the bandwidth of the transmission channel between the mobile vehicle and the UAV, is the time taken by the mobile vehicle to unload the routine task i to the drone m, is the power fading of the transmission channel between a mobile vehicle carrying a conventional mission i and a UAV m, L0 is the reference distance, Γ dB is the additional loss caused by the environment, is the distance between the mobile vehicle carrying routine mission i and the UAV m, H is the flight height of the UAV, d im is the lateral distance between the mobile vehicle carrying conventional mission i and the UAV m, p is the signal power, σ 2 is the noise power.

7. The multi-UAV optimal edge service method for a mobile vehicle according to claim 5, characterized in that: The amount of routine tasks R that drone m unloads to the base station to complete the emergency task m s2 The expression is as follows: (x) + =max{0,x} Where A1 is the set of regular tasks, j is the regular task number, R j,m is the amount of data transmitted from routine task j to UAV m, T j,m c f is the time that UAV m has provided computing for routine task j when the emergency task occurs, j,m uav is the computing capability of UAV m for routine task j.

8. The multi-UAV optimal edge service method for a mobile vehicle according to claim 5, characterized in that: Using drones as intelligent agents, a multi-agent deep Q-learning algorithm is used to find the optimal solution to the intra-cluster resource allocation optimization problem for each cluster, including: Taking drones as intelligent agents, the optimal resource allocation strategy for each intelligent agent is found through the multi-intelligence deep Q learning algorithm, where the state parameter s of the deep Q learning algorithm is {D i ,R i,m ,R m s }, the action parameter a is the resource V allocated by drone m to regular task i i ={ω i 、l i 、e i }, reward function ε>0, ε is the proportional coefficient, D i ={x i ,τ i }, x i is the normalized importance index of regular task i, τ i is the completion time limit of regular task i, ω i is the number of CPU cycles per bit allocated by the MEC onboard UAV m for routine task i, l i is the capacitance coefficient assigned by the MEC onboard UAV m for routine task i, e i is the local processing energy consumption allocated by the MEC onboard UAV m for routine task i.

9. A multi-UAV optimal edge service system for mobile vehicles, characterized in that: The system includes several mobile vehicles carrying tasks, several drones equipped with MEC, and a base station; Based on the method according to any one of claims 1 to 8, the drone provides edge computing services for the mobile vehicle, and the base station provides additional computing services for the drone.

10. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 8.

Images