A multi-uav assisted edge computing trajectory and resource joint optimization method

Abstract

The present application belongs to the technical field of unmanned aerial vehicle communication, and discloses a kind of trajectory and resource joint optimization method of multi-unmanned aerial vehicle assisted edge computing, aims at improving system energy efficiency and reducing user task backlog.The method first establishes multi-unmanned aerial vehicle operation system, constructs the joint optimization problem with the maximum energy efficiency and the minimum task backlog as the target, and decomposes it into two sub-problems of outer trajectory planning and inner resource scheduling.The outer layer generates multiple groups of candidate flight trajectories, combines the resource scheduling results fed back by the inner layer for comprehensive evaluation, selects the optimal trajectory to determine the flight speed and angle of the current time slot;The inner layer generates user access decision according to the flight state, and determines the corresponding power and computing resource allocation.Through inner and outer layer collaborative iterative optimization, the strong coupling problem of trajectory and resource allocation is effectively solved, and the overall performance and resource utilization of multi-unmanned aerial vehicle operation system are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) communication technology, and in particular to a method for joint optimization of trajectory and resources using multi-UAV assisted edge computing. Background Technology

[0002] Multi-UAV-assisted edge computing systems provide near-field computing services to ground users via UAVs equipped with edge servers, effectively compensating for insufficient ground infrastructure coverage. However, in actual deployments, there is a strong coupling relationship between the UAV's flight trajectory and the allocation of communication and computing resources: on the one hand, the flight trajectory determines channel quality and user access range; on the other hand, the resource allocation result, in turn, affects task processing efficiency and UAV energy consumption. Existing technologies typically handle trajectory planning and resource allocation separately, ignoring their interactive effects, resulting in low system energy efficiency, severe task backlog, and difficulty in adapting to dynamically changing user needs and environmental conditions.

[0003] Furthermore, multi-UAV collaborative scenarios also face the challenge of joint optimization of user access selection, power allocation, and computing resource allocation. These three types of decision variables are discrete and continuous, respectively, making it difficult for traditional optimization methods to find the global optimum in the mixed action space. Simultaneously, UAVs must satisfy multiple constraints such as flight speed, safe distance, and energy consumption, further increasing the complexity of the optimization problem. Existing methods often suffer from high computational complexity, slow convergence speed, and difficulty in real-time decision-making when dealing with such multi-variable, multi-constraint joint optimization problems, limiting the practical application effectiveness of multi-UAV operating systems. Summary of the Invention

[0004] Based on this, it is necessary to propose a joint optimization method for trajectory and resources using multi-UAV assisted edge computing to address the above problems.

[0005] A method for joint optimization of trajectory and resources using multi-UAV assisted edge computing, the method comprising: A multi-UAV operation system is established based on the flight status, communication quality, mission transmission and reception status, mission processing status, energy consumption status, and user movement status of each UAV. Based on the energy consumption status and the task processing status, a joint optimization problem is determined with the objective of maximizing the energy efficiency of the multi-UAV operating system and minimizing the total backlog of tasks for each user. The joint optimization problem is decomposed into an outer trajectory planning problem and an inner resource scheduling problem; In the outer trajectory planning problem, the current UAV position, user position, task queue status, and remaining energy status are obtained to generate multiple sets of UAV candidate flight trajectory sequences; Each group of candidate UAV flight trajectory sequences is input into the inner-layer resource scheduling problem to obtain the corresponding resource scheduling result; based on the resource scheduling result, the energy efficiency of the multi-UAV operation system and the total backlog of each user task corresponding to each group of candidate UAV flight trajectory sequences are calculated to determine the optimal trajectory sequence, and the flight speed and flight angle of the UAV in the current time slot are determined based on the optimal trajectory sequence; In the inner-layer resource scheduling problem, a user access decision list is generated based on the flight speed and flight angle of the UAV in the current time slot, and the corresponding power allocation and computing resource allocation results are determined based on the user access decision list.

[0006] In the above scheme, the multi-UAV operation system includes: UAV flight unit, user mobility unit, air-to-ground channel unit, user access unit, uplink non-orthogonal multiple access communication unit, UAV computing resource unit, task queue unit, and UAV power consumption unit; The UAV flight unit is established based on the flight speed, flight angle, and position update rules of each UAV; The user movement unit is established based on the update patterns of each user's movement speed and direction; The air-to-ground channel unit is established based on the line-of-sight link probability and non-line-of-sight link probability between each user and each UAV. The user access unit is established based on the access variables and access uniqueness constraints between each user and each drone; The uplink non-orthogonal multiple access communication unit is established based on the uplink transmit power and interference cancellation decoding order of each user; Based on the maximum computing power of each UAV and the computing resources allocated to each user, the UAV computing resource unit is established. The task queue unit is established based on the task arrival process and task queue update rules of each user; Based on the flight energy consumption and computing energy consumption of each UAV, the UAV energy consumption unit is established.

[0007] In the above scheme, the number is updated according to the following formula. A drone in a time slot Location:

[0008] The drone's flight speed meets the following constraints:

[0009] The drone's flight angle meets the following constraints:

[0010] The safe distance constraint must be met between any two drones:

[0011] in, For the first A drone in a time slot Horizontal position, For the first A drone in a time slot Horizontal position, The time slot length, For the first A drone in a time slot Flight speed, , The first The cosine and sine values ​​of the flight angle of the drone. For the first The maximum flight speed of the drone For a collection of drones, its index is , This is a set of time slots, with the index being... , For the first A drone in a time slot Flight angle For the first A drone in a time slot Horizontal position, This represents the minimum safe distance between any two drones.

[0012] In the above scheme, the step of determining a joint optimization problem based on the energy consumption state and the task processing state, with the objective of maximizing the energy efficiency of the multi-UAV operating system and minimizing the total backlog of tasks for each user, specifically includes: The preset optimization variables are: ;in, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle To provide users with access to variables, For user transmit power, Allocating computing resources for drones To optimize variables; The time slot of the multi-UAV operation system is determined according to the following formula. Its energy efficiency is:

[0013] in, To avoid extremely small numbers with a denominator of zero, For time slots Energy efficiency of multi-drone operation systems For time slots The amount of task data processed by the drone. For time slots Total energy consumption of the internal drone; Obtain the multi-user in the time slot The total backlog of user tasks is:

[0014] in, For time slots Total backlog of task queues for all users. For time slots Internal users Total backlog of user tasks. The total number of users, index is ; Construct the objective function for the joint optimization problem according to the following formula:

[0015] in, and They are time slots Energy efficiency and time slots of multi-drone operation systems The weighting coefficient corresponding to the total backlog of task queues for all users. This represents the total number of time slots. For time slots Total backlog of task queues for all users. For time slots Energy efficiency of multi-drone operation systems.

[0016] In the above scheme, the decomposition of the joint optimization problem into an outer trajectory planning problem and an inner resource scheduling problem specifically includes: Using the UAV's flight speed and flight angle as optimization variables in the outer trajectory planning problem, it can be expressed as follows: ; in, For the optimization variables of the outer trajectory planning problem, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle For a collection of drones, its index is , This is a set of time slots, with the index being... ; User access variables, power allocation variables, and UAV computing resource allocation variables are used as optimization variables for the inner resource scheduling problem, expressed as: ; in, For the optimization variables of the inner resource scheduling problem, To provide users with access to variables, For user transmit power, Allocate computing resources for drones.

[0017] In the above scheme, the step of obtaining the current UAV position, user position, task queue status, and remaining energy status in the outer trajectory planning problem, and generating multiple sets of candidate UAV flight trajectory sequences, specifically includes: The current time slot is constructed based on the drone's location, the user's location, the task queue status, and the remaining energy status. Status: ; in, For the first A drone in a time slot Horizontal position, For the first Individual users in time slots Location, For time slots Internal users Total backlog of user tasks. For the first The remaining energy of the drone; In the future Using time slots as prediction windows, a trajectory control sequence is established:

[0018] in, To predict the window length, and The first A drone in the predicted time slot Flight speed and flight angle, For a collection of drones, its index is ; Determine the cross-entropy sampling distribution of the trajectory control sequence: ,in, and These are the current time slots. The mean and covariance of the downsampling distribution; Sampled from the sampling distribution Group of candidate UAV trajectory control sequences: , among which, the The group trajectory control sequence is , For the first Group of candidate UAV trajectory control sequences The total number of candidate trajectory sequences sampled. The index of the candidate sequence. , The first In the candidate sequence group, the first drones in future prediction time slots Flight speed and flight angle, For a collection of drones, its index is .

[0019] In the above scheme, the step of calculating the energy efficiency of the multi-UAV operation system and the total backlog of each user task corresponding to each group of UAV candidate flight trajectory sequences based on the resource scheduling results, and determining the optimal trajectory sequence, specifically includes: For the Group trajectory control sequence Based on the recursive prediction of the UAV's location in future time slots, user location, channel state, task generation volume, and UAV's remaining energy state, a predicted resource allocation state is constructed. The predicted resource allocation status is input into the inner resource scheduling problem to obtain the corresponding resource scheduling result, and the uplink transmission rate, task completion amount, total user task backlog, UAV energy consumption and system energy efficiency are calculated under the predicted time slot. Definition of the first The evaluation value of the group trajectory control sequence is: ,

[0020]

[0021] in, For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning, In the first In the candidate sequence group, the predicted time slot The system energy efficiency, In the first In the candidate sequence group, the predicted time slot Total backlog of user tasks in multi-UAV operating systems For insufficient safe distance, This is a penalty for insufficient energy. , These are the collision safety penalty coefficient and the remaining energy penalty coefficient, respectively. This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks. For a collection of drones, its index is , The minimum safe distance between any two drones. In time slot , No. Group sequence of drones With drones The Euclidean distance between them The preset minimum safe remaining energy threshold for drones, In time slot , No. The first group in the sequence The remaining energy of the drone; Based on the evaluation value of the candidate UAV trajectory control sequence ,from Select the highest evaluation value from the group trajectory control sequence The group serves as the optimal candidate trajectory set; The mean and covariance of the sampling distribution are updated based on the optimal candidate trajectory set until a preset number of cross-entropy iterations are reached. The candidate trajectory control sequence with the highest evaluation value is then selected as the optimal trajectory sequence.

[0022] in, In time slot The optimal candidate trajectory control sequence was selected after evaluation. For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning.

[0023] In the above scheme, the step of generating a user access decision list based on the flight speed and flight angle of the UAV in the current time slot, and determining the corresponding power allocation and computing resource allocation results based on the user access decision list, specifically includes: The inner resource allocation problem is modeled as a Markov decision process, which mainly includes a state space, an action space, and a reward function. Defined in time slot The state space at time is:

[0024] in, For time slots The state space of time, For the first A drone in a time slot Location, For the first Individual users in time slots Location, For users In the time slot Task queue length, For the first A drone in a time slot The remaining energy, and These are the current time slot UAV flight speed and flight angle determined by the outer trajectory planning; Defined in time slot Action space during time:

[0025]

[0026]

[0027]

[0028] ; in, For time slots The action space of time, For user access matrix, For time slots At that time, the user With drones Access relationship, For time slots The set of continuous resource allocation parameters, For power allocation vector, For time slots Below, the transmission power of User 1, For time slots Users below Access to drones At that time, the computing resources allocated to that user To compute the resource allocation matrix; Define the reward function as follows:

[0029] in, For the reward function, For time slots The system energy efficiency is as follows. For time slots Total backlog of user tasks. The weighting coefficient for the energy penalty term. For time slots The energy constraint violation penalty term, This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks.

[0030] In the above scheme, generating the user access decision list includes: Using a discrete diffusion Actor network, the current time slot is... status As conditional guiding information, from Gaussian noise Start Passing The reverse denoising step yields the access weight matrix:

[0031] in, , For users In the time slot Access to drones Action weighting coefficients, For the process The final noise output after denoising steps This represents the total number of reverse denoising steps; Normalize the access weight matrix to obtain the user access probability:

[0032] in, Indicates user In the time slot Access to drones The probability, For specific users The weight coefficients of all drones are summed. The elements in the access weight matrix represent the time slots. Below, user Access to drones Action weighting coefficients; Based on the user access probability, sampling is performed to obtain user access actions, and a user access matrix is ​​generated based on the user access actions.

[0033] In the above scheme, determining the corresponding power allocation and computing resource allocation results based on the user access decision list includes: By continuously diffusing the Actor network, with the current time slot state As conditional guiding information, from Gaussian noise Beginning, then Step-by-step reverse denoising to obtain continuous parameters corresponding to all access actions. ; Based on the user access action ultimately selected by the Discrete Diffusion Actor From the continuous parameters corresponding to all the access actions Select the corresponding continuous parameter ; For the continuous parameters Perform deterministic constraint mapping to obtain executable continuous resource allocation results, which constitute the resource allocation action for the current time slot: .

[0034] The embodiments of this invention have the following beneficial effects: Based on the flight status, communication quality, task transmission and reception status, task processing status, energy consumption status, and user movement status of each UAV, this invention establishes a multi-UAV operating system, providing a unified descriptive basis for subsequent optimization, encompassing multi-dimensional information such as communication, computation, energy, and movement. Furthermore, based on the energy consumption status and task processing status, a joint optimization problem is determined, aiming to maximize system energy efficiency and minimize task backlog. This incorporates the mutually coupled system energy efficiency and total user task backlog into the same optimization framework, avoiding the problems of uneven resource utilization or excessive task processing latency caused by a single optimization objective. On this basis, the joint optimization problem is decomposed into an outer-layer trajectory planning problem and an inner-layer resource scheduling problem. Layered decoupling reduces the solution complexity of the original mixed-integer nonlinear optimization problem, enabling UAV trajectories and resource allocation to be collaboratively optimized within the same framework rather than processed independently. Then, in the outer-layer trajectory planning problem, the current UAV position is obtained... The system generates multiple candidate flight trajectory sequences based on the user's location, task queue status, and remaining energy status. By exploring the trajectory control action space through multi-sequence sampling, it overcomes the limitations of single path planning in adapting to dynamic environmental changes. Each candidate trajectory sequence is then input into an inner-layer resource scheduling problem to obtain resource scheduling results. Based on these results, the system energy efficiency and task backlog corresponding to each candidate trajectory are calculated, determining the optimal trajectory sequence and the flight speed and angle for the current time slot. This achieves a two-way collaborative optimization mechanism where outer-layer trajectory planning guides inner-layer resource scheduling, and inner-layer resource scheduling provides feedback to evaluate outer-layer trajectories. This allows the UAV trajectory to adaptively adjust based on real-time resource scheduling effects. Finally, a user access decision list is generated based on the UAV's flight speed and angle in the current time slot. This list is then used to determine the corresponding power allocation and computational resource allocation results, ensuring that power allocation and computational resource allocation match user access decisions and avoiding resource waste caused by the disconnect between access decisions and resource allocation. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] in: Figure 1 This is a schematic diagram of a collaborative communication scenario in one embodiment; Figure 2 This is a schematic diagram of a trajectory and resource joint optimization method for multi-UAV assisted edge computing in one embodiment; Figure 3 A closed-loop architecture diagram for joint optimization of trajectory and resources in multi-UAV-assisted edge computing. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] like Figure 1 As shown, this is the collaborative communication scenario constructed by the present invention. Within a set task area, multiple drones equipped with edge servers fly along a planned trajectory in the air to dynamically cover mobile users on the ground. These mobile users generate data with task queues according to their own business needs and offload or transmit tasks to drones through the uplink. In addition, the continuous displacement of ground mobile users and the movement of the drones themselves together constitute a dynamic and time-varying communication environment, which fully demonstrates the complexity and flexibility of the model in handling mobility management and air-ground integrated resource scheduling.

[0039] like Figure 2 As shown, in one embodiment, a method for joint optimization of trajectory and resources in multi-UAV assisted edge computing is provided. This method includes steps S101 to S106, which are detailed below: S101. Establish a multi-UAV operation system based on the flight status, communication quality, mission transmission and reception status, mission processing status, energy consumption status, and user movement status of each UAV. By integrating the flight status, communication quality, mission transmission and reception status, mission processing status, energy consumption status, and user movement status of each UAV, a multi-UAV operation system that comprehensively reflects the system's operating mechanism was constructed. This model provides an accurate mathematical description and state awareness foundation for subsequent joint optimization, ensuring that optimization decisions are based on real and reliable system information.

[0040] Preferably, the multi-UAV operation system includes drones and For each ground mobile user, the model employs a discrete-time structure, dividing time into multiple equal-length time slots. The set of time slots is as follows: The user set is: The drones are grouped as follows: Each time slot is [length missing] .

[0041] in, It can be configured based on the service area, user density, and drone computing power, for example... Option 2 is acceptable; It can be configured according to business scenarios and access requirements, for example 6 is acceptable; It can be set to 1 second. The above values ​​are merely examples and do not constitute a limitation on the scope of protection of this invention.

[0042] In some embodiments, a multi-UAV operating system includes: a UAV flight unit, a user mobility unit, an air-to-ground channel unit, a user access unit, an uplink non-orthogonal multiple access communication unit, a UAV computing resource unit, a task queue unit, and a UAV power consumption unit. Based on the flight speed, flight angle, and position update rules of each UAV, establish UAV flight units; Based on the update patterns of each user's movement speed and direction, user movement units are established; Based on the line-of-sight link probability and non-line-of-sight link probability between each user and each UAV, an air-to-ground channel unit is established. Based on the access variables and access uniqueness constraints between each user and each drone, a user access unit is established; Based on the uplink transmit power and interference cancellation decoding order of each user, an uplink non-orthogonal multiple access communication unit is established. Based on the maximum computing power of each drone and the computing resources allocated to each user, establish drone computing resource units; Based on the task arrival process and task queue update rules of each user, establish task queue units; Based on the flight energy consumption and computing energy consumption of each UAV, an UAV energy consumption unit is established.

[0043] Specifically, this multi-UAV operation system constructs a complete system simulation and optimization closed loop through the organic connection of eight functional units. Among them, the UAV flight unit and user mobility unit respectively characterize the spatiotemporal evolution of the air platform and ground terminal, providing dynamic position input for the communication link; the air-to-ground channel unit calculates channel gain based on the relative positional relationship between the two, mapping physical space into a communication quality metric; the user access unit determines service matching relationships based on channel conditions and access constraints, while the uplink non-orthogonal multiple access communication unit simulates multi-user interference and rate allocation based on access; the UAV computing resource unit allocates computing power according to the number of accessing users, the task queue unit tracks task arrival and processing progress, jointly determining the task backlog status; and the UAV energy consumption unit summarizes flight and computing power consumption, providing feedback on energy constraints. These eight units drive and progress layer by layer, from physical motion to communication transmission and then to computational processing, ultimately converging into two core evaluation indicators: system energy efficiency and task backlog, providing unified model support for outer-layer trajectory planning and inner-layer resource scheduling.

[0044] In some embodiments, the UAV flight unit updates the following formula: A drone in a time slot Location:

[0045] The drone's flight speed meets the following constraints:

[0046] The drone's flight angle meets the following constraints:

[0047] The safe distance constraint must be met between any two drones:

[0048] in, For the first A drone in a time slot Horizontal position, For the first A drone in a time slot Horizontal position, The time slot length, For the first A drone in a time slot Flight speed, , The first The cosine and sine values ​​of the flight angle of the drone. For the first The maximum flight speed of the drone For a collection of drones, its index is , This is a set of time slots, with the index being... , For the first A drone in a time slot Flight angle For the first A drone in a time slot Horizontal position, This represents the minimum safe distance between any two drones.

[0049] Preferably, the user movement unit defines the user. In the time slot The two-dimensional position is Its moving speed and moving direction angle are updated using the Gauss-Markov model as follows:

[0050] Time slot Internal user location The location has been updated to:

[0051] in, For the first Individual users in time slots Location, , The first Individual users in time slots The x-coordinate and y-coordinate of a Cartesian coordinate system. For the first Individual users in time slots Location, For the first Individual users in time slots movement speed, For the first Individual users in time slots The direction of movement, The time slot length, For motor memory factors, For users Average moving speed For users The average direction of movement, and These are Gaussian perturbation random terms representing velocity and direction of movement, respectively.

[0052] In some embodiments, the air-to-ground channel unit defines the user With drones In the time slot The two-dimensional distance is: ; user With drones In the time slot The three-dimensional spatial distance is: ; user With drones In the time slot The elevation angle of the air-to-ground link is: ; user With drones In the time slot The line-of-sight link probability is: ; user With drones In the time slot The probability of a non-line-of-sight link is: ; The path losses corresponding to the above line-of-sight link probabilities and non-line-of-sight link probabilities are as follows: , ; In summary, users With drones In the time slot The average path loss of the air-to-ground link is: ; user With drones In the time slot The channel gain is: ; in, For users With drones In the time slot Two-dimensional distance, For users Two-dimensional position, For drones Two-dimensional position, For users With drones In the time slot The three-dimensional spatial distance The flight altitude of the drone. For users With drones In the time slot The elevation angle of the air-to-ground link, For users With drones In the time slot line-of-sight link probability, a and b These are environmentally relevant constants that can be manually set. For users With drones In the time slot Non-line-of-sight link probability, This represents the path loss corresponding to the line-of-sight link probability. For carrier frequency, At the speed of light, and These are the additional losses for line-of-sight links and non-line-of-sight links, respectively. The path loss corresponding to the probability of non-line-of-sight links. For users With drones In the time slot The average path loss of air-to-ground links, For users With drones In the time slot Channel gain.

[0053] in, The range can be set to 100m, depending on the service area and communication coverage requirements. Depending on the drone model, it can be set to 20m / s; Depending on the drone model and environmental factors, it can be set to 10m. a and b These can be set to 9.61 and 0.16 respectively; It can be set to 2GHz; Set as m / s; Set to 1dB; Set to 20dB.

[0054] Preferably, the user access unit defines the user. In the time slot With drones The access variables between them are 0 represents the user In the time slot Unconnected drones 1 represents user In the time slot Access to drones Each user accesses one drone in the same time slot, satisfying the constraints. Access to drones The user set is defined as: The user access matrix is ​​defined as follows: ;in, To provide users with access to variables, The total number of users, index is , For user collection, A collection of drones.

[0055] Preferably, the uplink non-orthogonal multiple access communication unit defines the user. In the time slot The transmission power is It satisfies the power constraint: ,in, Indicates the user's maximum transmit power. For users In the time slot The transmission power; Define user In the time slot Access to drones At that time, drones The received signal power is: ,in, To provide users with access to variables, , For users In the time slot Access to drones At that time, drones Received signal power, For users With drones In the time slot Channel gain.

[0056] drones For user set Users in the process use continuous interference cancellation for decoding, and the decoding order is determined based on the received power from high to low.

[0057] Among them, the definition of drones In the time slot The decoding order is as follows: The decoding order strictly follows the descending order of received power. Users with higher received power are decoded earlier because high-power signals are easier to extract from noise and interference. For drones In the time slot The decoding order table, where each entry represents a user's index. For user groups.

[0058] drones Decoding the first individual users When a signal is received, user signals prior to it have been removed through continuous interference cancellation, and user signals subsequent to it are considered as interference. Simultaneously, considering noise and cross-drone interference caused by users accessing other drones, the user... In drones The signal-to-interference-plus-noise ratio at this location is: ; Uplink transmission rate is: ; in, For drones Decoding the first Signal-to-interference-plus-noise ratio per user The power of the desired signal, The first in the decoding order User signals from previous users, To interfere with drones, B For bandwidth, For noise power, For user collection, This refers to the uplink transmission rate.

[0059] Preferably, the UAV computing resource unit defines the UAV. In the time slot Assigned to user The computing resources are The allocation of computing resources among all users and drones constitutes a computing resource allocation matrix: Due to users In the time slot Drones that can only be accessed by it Provide computing services, therefore satisfy ,in, To maximize the computing power of drones, To provide users with access to variables, For drones In the time slot Assigned to user The computing resources allocated to each drone in the same time slot shall not exceed its maximum computing capacity; among which, The total number of users, index is , For a collection of drones, its index is , This is a set of time slots, with the index being... .

[0060] Preferably, the task queue unit defines the user. In the time slot The initial task queue length is The newly generated computing workload is The newly generated computational workload is determined by the number of tasks arriving and the task size, defining the user. In the time slot The number of tasks reached is It follows a Poisson distribution. ,in, Indicates user Average number of tasks arriving within a time slot.

[0061] Also define user The The data size of each newly arriving task is It follows a uniform distribution ,in, and These represent the lower and upper limits of the data size for a single task, respectively.

[0062] In summary, users In the time slot The newly generated computing workload is According to the user's uplink transmission rate ,user In the time slot The number of tasks that can be uploaded is , The time slot length; based on the allocation of UAV computing resources, the user In the time slot The amount of tasks that can be processed by the drone edge server is: ,in, The number of CPU cycles required to process a unit of task. To provide users with access to variables, For drones In the time slot Assigned to user Computing resources.

[0063] Among them, users In the time slot The amount of work completed within the time limit is: The user queue has been updated to: , For users In the time slot The number of tasks that can be uploaded For users In the time slot Initial task queue length, For users In the time slot The amount of work that can be processed can be calculated through the drone edge server. For users In the time slot The amount of work completed within a given timeframe. For users In the time slot The newly generated computational workload For users In the time slot The initial length of the task queue.

[0064] Among them, in time slots The total number of tasks completed is In the time slot The total backlog of user tasks is .

[0065] Preferably, the UAV energy consumption unit describes the energy consumption generated by the UAV while serving ground mobile users, and the UAV energy consumption includes flight energy consumption and computing energy consumption. A drone in a time slot The flight speed and flight angle are as follows: and Therefore, its velocity vector can be expressed as ; Furthermore, its acceleration vector can be expressed as ; In summary, defining drones In the time slot The flight energy consumption is ; in, The time slot length, For the first A drone in a time slot Flight speed, , The first The cosine and sine values ​​of the flight angle of the drone. For the first A drone in a time slot acceleration vector, For the first A drone in a time slot Flight speed, For drones In the time slot Flight energy consumption, For drones The basic power of hovering, and This represents the correlation coefficient of drone flight energy consumption.

[0066] Furthermore, drones are also defined. In the time slot The calculated energy consumption is: ; in, The power consumption coefficient of the computing chip carried by the drone. For drones In the time slot Computational energy consumption, To provide users with access to variables, For drones In the time slot Assigned to user Computing resources The time slot length; Therefore, drones In the time slot Total energy consumption is: Define drones In the time slot The remaining energy is The remaining energy is then updated to .

[0067] S102. Based on the energy consumption status and task processing status, determine the joint optimization problem with the goal of maximizing the energy efficiency of the multi-UAV operation system and minimizing the total backlog of tasks for each user. With the optimization objectives of maximizing system energy efficiency and minimizing task backlog, the direction of joint optimization of trajectory and resources was clearly defined. This objective function enables the system to improve energy utilization efficiency while responding promptly to user computing needs, avoiding the problems of simply pursuing energy efficiency leading to a decline in service quality, or unilaterally reducing task backlog resulting in energy waste.

[0068] In some embodiments, based on the energy consumption state and task processing state, a joint optimization problem is determined with the objective of maximizing the energy efficiency of the multi-UAV operating system and minimizing the total task backlog for each user, specifically including: The preset optimization variables are: ;in, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle To provide users with access to variables, For user transmit power, Allocating computing resources for drones To optimize variables; The time slot for multi-UAV operation systems is determined using the following formula. Its energy efficiency is:

[0069] in, To avoid extremely small numbers with a denominator of zero, For time slots Energy efficiency of multi-drone operation systems For time slots The amount of task data processed by the drone. For time slots Total energy consumption of the internal drone; Acquiring multi-UAV operating systems in time slots The total backlog of user tasks is:

[0070] in, For time slots Total backlog of task queues for all users. For time slots Internal users Total backlog of user tasks. The total number of users, index is ; Construct the objective function for the joint optimization problem based on the following formula:

[0071] in, and They are time slots Energy efficiency and time slots of multi-drone operation systems The weighting coefficient corresponding to the total backlog of task queues for all users. This represents the total number of time slots. For time slots Total backlog of task queues for all users. For time slots Energy efficiency of multi-drone operation systems.

[0072] S103. Decompose the joint optimization problem into an outer trajectory planning problem and an inner resource scheduling problem; By decomposing the complex joint optimization problem into an outer-layer trajectory planning problem and an inner-layer resource scheduling problem, the decoupling of trajectory control and resource allocation is achieved. This layered architecture reduces the difficulty of solving the problem, allowing the outer layer to focus on flight path optimization and the inner layer to focus on the fine-grained allocation of communication and computing resources. The two layers cooperate and iterate together, effectively overcoming the solution difficulties caused by the coupling of dependent variables in traditional methods.

[0073] In some embodiments, the joint optimization problem is decomposed into an outer trajectory planning problem and an inner resource scheduling problem, specifically including: Using the UAV's flight speed and flight angle as optimization variables in the outer trajectory planning problem, it can be expressed as follows: ; in, For the optimization variables of the outer trajectory planning problem, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle For a collection of drones, its index is , This is a set of time slots, with the index being... ; User access variables, power allocation variables, and UAV computing resource allocation variables are used as optimization variables for the inner resource scheduling problem, expressed as: ; in, For the optimization variables of the inner resource scheduling problem, To provide users with access to variables, For user transmit power, Allocate computing resources for drones.

[0074] S104. In the outer trajectory planning problem, obtain the current UAV position, user position, task queue status and remaining energy status, and generate multiple sets of UAV candidate flight trajectory sequences; Based on the current UAV location, user location, task queue status, and remaining energy status, multiple sets of candidate flight trajectory sequences are generated. This step provides ample exploration space for trajectory optimization, enabling the system to select the optimal path from various possibilities. This avoids the problem of a single trajectory scheme potentially getting stuck in local optima and enhances the algorithm's global search capability.

[0075] In some embodiments, in the outer trajectory planning problem, the current UAV position, user position, task queue status, and remaining energy status are obtained to generate multiple sets of candidate UAV flight trajectory sequences, specifically including: The current time slot is constructed based on the drone's location, the user's location, the task queue status, and the remaining energy status. Status: ; in, For the first A drone in a time slot Horizontal position For the first Individual users in time slots Location, For time slots Internal users Total backlog of user tasks. For the first The remaining energy of the drone; In the future Using time slots as prediction windows, a trajectory control sequence is established:

[0076] in, To predict the window length, and The first A drone in the predicted time slot Flight speed and flight angle, For a collection of drones, its index is ; Determine the cross-entropy sampling distribution of the trajectory control sequence: ,in, and These are the current time slots. The mean and covariance of the downsampling distribution; Sampled from the sampling distribution Group of candidate UAV trajectory control sequences: , among which, the The group trajectory control sequence is , For the first Group of candidate UAV trajectory control sequences The total number of candidate trajectory sequences sampled. The index of the candidate sequence. , The first In the candidate sequence group, the first drones in future prediction time slots Flight speed and flight angle, For a collection of drones, its index is .

[0077] S105. Input each group of candidate flight trajectory sequences of UAVs into the inner-layer resource scheduling problem to obtain the corresponding resource scheduling results; based on the resource scheduling results, calculate the energy efficiency of the multi-UAV operation system and the total backlog of each user task corresponding to each group of candidate flight trajectory sequences of UAVs, determine the optimal trajectory sequence, and determine the flight speed and flight angle of the UAV in the current time slot based on the optimal trajectory sequence. Each candidate trajectory is input into the inner-layer resource scheduling problem for evaluation. Based on the feedback from the resource scheduling results, the corresponding system energy efficiency and task backlog are calculated, thereby selecting the optimal trajectory sequence and determining the flight speed and flight angle for the current time slot. This step achieves closed-loop linkage between outer-layer trajectory planning and inner-layer resource scheduling, ensuring that the selected trajectory is not only optimal at the flight level but also matches the resource allocation strategy, achieving a global collaborative optimization effect.

[0078] In some embodiments, based on resource scheduling results, the energy efficiency of the multi-UAV operating system and the total backlog of user tasks corresponding to each group of UAV candidate flight trajectory sequences are calculated to determine the optimal trajectory sequence, specifically including: For the Group trajectory control sequence Based on the recursive prediction of the UAV's location in future time slots, user location, channel state, task generation volume, and UAV's remaining energy state, a predicted resource allocation state is constructed. The predicted resource allocation status is input into the inner resource scheduling problem to obtain the corresponding resource scheduling results, and the uplink transmission rate, task completion amount, total backlog of user tasks, UAV energy consumption and system energy efficiency are calculated under the predicted time slot. Definition of the first The evaluation value of the group trajectory control sequence is: ,

[0079]

[0080] in, For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning, In the first In the candidate sequence group, the predicted time slot The system energy efficiency, In the first In the candidate sequence group, the predicted time slot Total backlog of user tasks in multi-UAV operating systems For insufficient safe distance, This is a penalty for insufficient energy. , These are the collision safety penalty coefficient and the remaining energy penalty coefficient, respectively. This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks. For a collection of drones, its index is , The minimum safe distance between any two drones. In time slot , No. Group sequence of drones The Euclidean distance between the drone n and the drone n The preset minimum safe remaining energy threshold for drones, In time slot , No. The first group in the sequence The remaining energy of the drone; Based on the evaluation value of the candidate UAV trajectory control sequence ,from Select the highest evaluation value from the group trajectory control sequence The group serves as the optimal candidate trajectory set; The mean and covariance of the sampling distribution are updated based on the optimal candidate trajectory set until a preset number of cross-entropy iterations are reached. The candidate trajectory control sequence with the highest evaluation value is then selected as the optimal trajectory sequence.

[0081] in, In time slot The optimal candidate trajectory control sequence was selected after evaluation. For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning.

[0082] S106. In the inner-layer resource scheduling problem, a user access decision list is generated based on the flight speed and flight angle of the UAV in the current time slot, and the corresponding power allocation and computing resource allocation results are determined based on the user access decision list.

[0083] Based on the current time slot flight speed and flight angle determined by the outer layer, a user access decision list is generated, and the corresponding power allocation and computing resource allocation results are determined accordingly. This step fully utilizes the flight status information provided by the outer trajectory planning, enabling resource allocation decisions to dynamically adapt to the UAV's real-time position and motion trends. This achieves three-dimensional collaborative optimization of user access, power control, and computing resources, maximizing the system's service capabilities and resource utilization efficiency.

[0084] In some embodiments, in the inner-layer resource scheduling problem, a user access decision list is generated based on the flight speed and flight angle of the UAV in the current time slot, and the corresponding power allocation and computing resource allocation results are determined based on the user access decision list, specifically including: The inner resource allocation problem is modeled as a Markov decision process, which mainly includes the state space, action space and reward function. Defined in time slot The state space at time is:

[0085] in, For time slots The state space of time, For the first A drone in a time slot Location, For the first Individual users in time slots Location, For users In the time slot Task queue length, For the first A drone in a time slot The remaining energy, and These are the current time slot UAV flight speed and flight angle determined by the outer trajectory planning; Defined in time slot Action space during time:

[0086]

[0087]

[0088]

[0089] ; in, For time slots The action space of time, For user access matrix, For time slots At that time, the user With drones Access relationship, For time slots The set of continuous resource allocation parameters, For power allocation vector, For time slots Below, the transmission power of User 1, For time slots Users below Access to drones At that time, the computing resources allocated to that user To compute the resource allocation matrix; Define the reward function as follows:

[0090] in, For the reward function, For time slots The system energy efficiency is as follows. For time slots Total backlog of user tasks. The weighting coefficient for the energy penalty term. For time slots The energy constraint violation penalty term, This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks.

[0091] In some embodiments, a user access decision list is generated, including: Using a discrete diffusion Actor network, the current time slot is... status As conditional guiding information, from Gaussian noise Start Passing The reverse denoising step yields the access weight matrix:

[0092] in, , For users In the time slot Access to drones Action weighting coefficients, For the process The final noise output after denoising steps This represents the total number of reverse denoising steps; Normalize the access weight matrix to obtain the user access probability:

[0093] in, Indicates user In the time slot Access to drones The probability, For specific users The weight coefficients of all drones are summed. The elements in the access weight matrix represent the time slots. Below, user Access to drones Action weighting coefficients; The user access probability is sampled to obtain the user access actions, and a user access matrix is ​​generated based on the user access actions.

[0094] Preferably, the current time slot status As conditional guidance information input to the continuous diffusion Actor network, the continuous diffusion Actor is used to generate continuous parameters corresponding to all access choices. It also defines the set of all possible user access actions as follows: ,in, Indicates the first For each user access scenario, the corresponding continuous resource allocation parameters are expressed as follows: ;in, , The first Continuous transmit power parameters and continuous resource allocation parameters under various access conditions.

[0095] In some embodiments, determining the corresponding power allocation and computing resource allocation results based on the user access decision list includes: By continuously diffusing the Actor network, with the current time slot state As conditional guiding information, from Gaussian noise Beginning, then Step-by-step reverse denoising to obtain continuous parameters corresponding to all access actions. ; Based on the user access action ultimately selected by the Discrete Diffusion Actor From the continuous parameters corresponding to all access actions Select the corresponding continuous parameter ; For continuous parameters Perform deterministic constraint mapping to obtain executable continuous resource allocation results, which constitute the resource allocation action for the current time slot: .

[0096] In some embodiments, UAV flight control, user access, power allocation, and UAV computing resource allocation for the current time slot are executed based on the resource allocation actions of the current time slot. After execution, the inner resource allocation state for the next time slot is obtained. The reward for the current time slot is obtained based on the reward function. This generates state transition samples, which are then stored in the reinforcement learning experience replay pool. .

[0097] Simultaneously, the weight matrix output by the discrete diffusion Actor will be... Continuous parameters of the continuous diffusion Actor output Store the corresponding state in the diffusion experience pool. ; when and When the number of samples in the database reaches the preset update threshold, from Collect state transition samples Constructing target action value using Actor target network and Critic target network:

[0098] in, As a discount factor, and These are two Critic target networks, For the Actor target network based on the next time slot state The generated resource allocation action, This is the reward for the current time slot.

[0099] The two Critic networks are updated based on the target action value, and the loss function is:

[0100] in, and These are two Critic networks, For the Actor target network based on the next time slot state The generated resource allocation action, Value of the target action.

[0101] From the diffusion experience pool Mid-sampling diffusion Actor output samples, and As an optimizable variable, for the user access weight matrix The user access probability distribution is obtained through softmax, and optimization is performed based on the expected action value under this probability distribution:

[0102] For continuous parameters Optimize directly in the direction of maximizing the value of the action:

[0103] in, For user access actions, These are the continuous parameters corresponding to all access actions. Select the corresponding continuous parameters for all access actions. Provide a weight matrix for user access; Obtain the optimized user access weight matrix and continuous parameters ,in, and These represent the discrete action weight matrix and the continuous parameter matrix, respectively, optimized towards higher action value.

[0104] Based on the optimized discrete action weight matrix and continuous parameter matrix, two Actor networks are updated through imitation learning to make the output of the discrete diffusion Actor approximate the target value. This makes the output of the continuously diffusing Actor approximate... Its loss function is: ,in, Represents the discrete diffusion Actor in state The generated discrete action weight matrix Represents a continuously diffusing Actor in state The continuous parameter matrix generated below.

[0105] Finally, the Actor and Critic target networks are soft-updated. Through this update process, the Critic network learns the action value of resource allocation actions, enabling the discrete and continuous diffusion Actors to generate user access, power allocation, and UAV computing resource allocation actions with higher action value.

[0106] Iteratively update the discrete diffusion Actor, continuous diffusion Actor, Critic network and their corresponding target network according to the above steps until the preset convergence condition is met, and obtain the trained inner layer resource allocation model.

[0107] like Figure 3 As shown, this invention presents a closed-loop architecture for trajectory and resource joint optimization in multi-UAV assisted edge computing. The top layer is a multi-UAV edge computing environment that outputs the current state and receives joint actions from the outer layer trajectory optimization and the inner layer resource allocation. The middle layer outputs a user access weight matrix and a continuous diffusion Actor through discrete diffusion Actor, outputs resource parameters to generate hybrid actions, and stores them in the reinforcement learning experience pool and diffusion experience pool, respectively. The bottom layer consists of a dual Critic network that evaluates the value of actions, and a dual Actor network that approximates the optimal strategy through imitation learning. Finally, the target network parameters are synchronized through a soft update mechanism to achieve joint intelligent scheduling of user access, power allocation, and UAV computing resources in dynamic air-ground collaborative scenarios.

[0108] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] The embodiments described above are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application's patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. The embodiments disclosed above are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made according to the claims of this invention are still within the scope of this invention.

Claims

1. A method for joint optimization of trajectory and resources using multi-UAV assisted edge computing, characterized in that, The method includes: A multi-UAV operation system is established based on the flight status, communication quality, mission transmission and reception status, mission processing status, energy consumption status, and user movement status of each UAV. Based on the energy consumption status and the task processing status, a joint optimization problem is determined with the objective of maximizing the energy efficiency of the multi-UAV operating system and minimizing the total backlog of tasks for each user. The joint optimization problem is decomposed into an outer trajectory planning problem and an inner resource scheduling problem; In the outer trajectory planning problem, the current UAV position, user position, task queue status, and remaining energy status are obtained to generate multiple sets of UAV candidate flight trajectory sequences; Each group of candidate UAV flight trajectory sequences is input into the inner-layer resource scheduling problem to obtain the corresponding resource scheduling result; based on the resource scheduling result, the energy efficiency of the multi-UAV operation system and the total backlog of each user task corresponding to each group of candidate UAV flight trajectory sequences are calculated to determine the optimal trajectory sequence, and the flight speed and flight angle of the UAV in the current time slot are determined based on the optimal trajectory sequence; In the inner-layer resource scheduling problem, a user access decision list is generated based on the flight speed and flight angle of the UAV in the current time slot, and the corresponding power allocation and computing resource allocation results are determined based on the user access decision list.

2. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 1, characterized in that, The multi-UAV operation system includes: UAV flight unit, user mobility unit, air-to-ground channel unit, user access unit, uplink non-orthogonal multiple access communication unit, UAV computing resource unit, task queue unit, and UAV power consumption unit; The UAV flight unit is established based on the flight speed, flight angle, and position update rules of each UAV; The user movement unit is established based on the update patterns of each user's movement speed and direction; The air-to-ground channel unit is established based on the line-of-sight link probability and non-line-of-sight link probability between each user and each UAV. The user access unit is established based on the access variables and access uniqueness constraints between each user and each drone; The uplink non-orthogonal multiple access communication unit is established based on the uplink transmit power and interference cancellation decoding order of each user; Based on the maximum computing power of each UAV and the computing resources allocated to each user, the UAV computing resource unit is established. The task queue unit is established based on the task arrival process and task queue update rules of each user; Based on the flight energy consumption and computing energy consumption of each UAV, the UAV energy consumption unit is established.

3. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 2, characterized in that, Update the following formula: A drone in a time slot Location: The drone's flight speed meets the following constraints: The drone's flight angle meets the following constraints: The safe distance constraint must be met between any two drones: in, For the first A drone in a time slot Horizontal position For the first A drone in a time slot Horizontal position The time slot length, For the first A drone in a time slot Flight speed, , The first The cosine and sine values ​​of the flight angle of the drone. For the first The maximum flight speed of the drone For a collection of drones, its index is , This is a set of time slots, with the index being... , For the first A drone in a time slot Flight angle For the first A drone in a time slot Horizontal position This represents the minimum safe distance between any two drones.

4. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 3, characterized in that, The step of determining a joint optimization problem based on the energy consumption state and the task processing state, with the objective of maximizing the energy efficiency of the multi-UAV operating system and minimizing the total backlog of tasks for each user, specifically includes: The preset optimization variables are: ;in, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle To provide users with access to variables, For user transmission power, Allocating computing resources for drones To optimize variables; The time slot of the multi-UAV operation system is determined according to the following formula. Its energy efficiency is: in, To avoid extremely small numbers with a denominator of zero, For time slots Energy efficiency of multi-drone operation systems For time slots The amount of task data processed by the drone. For time slots Total energy consumption of the internal drone; Obtain the multi-user in the time slot The total backlog of tasks in the queue is: in, For time slots Total backlog of task queues for all users. For time slots Internal users Total backlog of user tasks. The total number of users, index is ; Construct the objective function for the joint optimization problem according to the following formula: in, and They are time slots Energy efficiency and time slots of multi-drone operation systems The weighting coefficient corresponding to the total backlog of task queues for all users. This represents the total number of time slots. For time slots Total backlog of task queues for all users. For time slots Energy efficiency of multi-drone operation systems.

5. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 4, characterized in that, The decomposition of the joint optimization problem into an outer trajectory planning problem and an inner resource scheduling problem specifically includes: Using the UAV's flight speed and flight angle as optimization variables in the outer trajectory planning problem, it can be expressed as follows: ; in, For the optimization variables of the outer trajectory planning problem, For the first A drone in a time slot Flight speed, For the first A drone in a time slot Flight angle For a collection of drones, its index is , This is a set of time slots, with the index being... ; User access variables, power allocation variables, and UAV computing resource allocation variables are used as optimization variables for the inner resource scheduling problem, expressed as: ; in, For the optimization variables of the inner resource scheduling problem, To provide users with access to variables, For user transmission power, Allocate computing resources for drones.

6. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 5, characterized in that, In the outer trajectory planning problem, the current UAV position, user position, task queue status, and remaining energy status are obtained to generate multiple sets of candidate UAV flight trajectory sequences, specifically including: The current time slot is constructed based on the drone's location, the user's location, the task queue status, and the remaining energy status. Status: ; in, For the first A drone in a time slot Horizontal position For the first Individual users in time slots Location, For time slots Internal users Total backlog of user tasks. For the first The remaining energy of the drone; In the future Using time slots as prediction windows, a trajectory control sequence is established: in, To predict the window length, and The first A drone in the predicted time slot Flight speed and flight angle, For a collection of drones, its index is ; Determine the cross-entropy sampling distribution of the trajectory control sequence: ,in, and These are the current time slots. The mean and covariance of the downsampling distribution; Sampled from the sampling distribution Group of candidate UAV trajectory control sequences: , among which, the The group trajectory control sequence is , For the first Group of candidate UAV trajectory control sequences The total number of candidate trajectory sequences sampled. The index of the candidate sequence. , The first In the candidate sequence group, the first drones in future prediction time slots Flight speed and flight angle, For a collection of drones, its index is .

7. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 6, characterized in that, The step of calculating the energy efficiency of the multi-UAV operating system and the total backlog of user tasks corresponding to each group of UAV candidate flight trajectory sequences based on the resource scheduling results, and determining the optimal trajectory sequence, specifically includes: For the Group trajectory control sequence Based on the recursive prediction of the UAV's location in future time slots, user location, channel state, task generation volume, and UAV's remaining energy state, a predicted resource allocation state is constructed. The predicted resource allocation status is input into the inner resource scheduling problem to obtain the corresponding resource scheduling result, and the uplink transmission rate, task completion amount, total user task backlog, UAV energy consumption and system energy efficiency are calculated under the predicted time slot. Definition of the first The evaluation value of the group trajectory control sequence is: , in, For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning, In the first In the candidate sequence group, the predicted time slot The system energy efficiency, In the first In the candidate sequence group, the predicted time slot Total backlog of user tasks in multi-UAV operating systems For insufficient safe distance, This is a penalty for insufficient energy. , These are the collision safety penalty coefficient and the remaining energy penalty coefficient, respectively. This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks. For a collection of drones, its index is , The minimum safe distance between any two drones. In time slot , No. Group sequence of unmanned aerial vehicles With drones The Euclidean distance between them The preset minimum safe remaining energy threshold for drones, In time slot , No. The first group in the sequence The remaining energy of the drone; Based on the evaluation value of the candidate UAV trajectory control sequence ,from Select the highest evaluation value from the group trajectory control sequence The group serves as the optimal candidate trajectory set; The mean and covariance of the sampling distribution are updated based on the optimal candidate trajectory set until a preset number of cross-entropy iterations are reached. The candidate trajectory control sequence with the highest evaluation value is then selected as the optimal trajectory sequence. in, In time slot The optimal candidate trajectory control sequence was selected after evaluation. For the first Group candidate trajectory control sequence in the current time slot The overall evaluation value at the beginning.

8. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 7, characterized in that, In the inner-layer resource scheduling problem, a user access decision list is generated based on the flight speed and flight angle of the UAV in the current time slot, and the corresponding power allocation and computing resource allocation results are determined based on the user access decision list, specifically including: The inner resource allocation problem is modeled as a Markov decision process, which mainly includes a state space, an action space, and a reward function. Defined in time slot The state space at time is: in, For time slots The state space of time, For the first A drone in a time slot Location, For the first Individual users in time slots Location, For users In the time slot Task queue length, For the first A drone in a time slot The remaining energy, and These are the current time slot UAV flight speed and flight angle determined by the outer trajectory planning; Defined in time slot Action space during time: ； in, For time slots The action space of time, For user access matrix, For time slots At that time, the user With drones Access relationship, For time slots The set of continuous resource allocation parameters, For power allocation vector, For time slots Below, the transmission power of User 1, For time slots Users below Access to drones At that time, the computing resources allocated to that user To compute the resource allocation matrix; Define the reward function as follows: in, For the reward function, For time slots The system energy efficiency is as follows. For time slots Total backlog of user tasks. The weighting coefficient for the energy penalty term. For time slots The energy constraint violation penalty term, This is a weighting coefficient for system energy efficiency. The weighting coefficient for the total backlog of user tasks.

9. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 8, characterized in that, The generated user access decision list includes: Using a discrete diffusion Actor network, the current time slot is... status As conditional guiding information, from Gaussian noise Start Passing The reverse denoising step yields the access weight matrix: in, , For users In the time slot Access to drones Action weighting coefficients, For the process The final noise output after denoising steps This represents the total number of reverse denoising steps; Normalize the access weight matrix to obtain the user access probability: in, Indicates user In the time slot Access to drones The probability, For specific users The weight coefficients of all drones are summed. The elements in the access weight matrix represent the time slots. Below, user Access to drones Action weighting coefficients; Based on the user access probability, sampling is performed to obtain user access actions, and a user access matrix is ​​generated based on the user access actions.

10. The trajectory and resource joint optimization method for multi-UAV assisted edge computing according to claim 9, characterized in that, The step of determining the corresponding power allocation and computing resource allocation results based on the user access decision list includes: By continuously diffusing the Actor network, with the current time slot state As conditional guiding information, from Gaussian noise Beginning, then Step-by-step reverse denoising to obtain continuous parameters corresponding to all access actions. ; Based on the user access action ultimately selected by the Discrete Diffusion Actor From the continuous parameters corresponding to all the access actions Select the corresponding continuous parameter ; For the continuous parameters Perform deterministic constraint mapping to obtain executable continuous resource allocation results, which constitute the resource allocation action for the current time slot: .

Images