Method and system for jointly optimizing track of unmanned aerial vehicle and unloading dependent tasks

Abstract

The invention belongs to the technical field of unmanned aerial vehicles, and discloses a method and a system for jointly optimizing the trajectory of an unmanned aerial vehicle and unloading a dependent task, and the algorithm optimizes the flight trajectory of the unmanned aerial vehicle and the unloading decision of the dependent task through a problem decomposition method, so as to reduce the calculation complexity of direct solution. On the premise that an unloading strategy is given, a multi-target particle swarm optimization algorithm is used for optimizing the flight path of the unmanned aerial vehicle, so that the unmanned aerial vehicle covers more ground users as much as possible, and the maximum value of end-to-end time delay is reduced. On the premise that the trajectory of the unmanned aerial vehicle is determined, the PPO algorithm is used to optimize the task-dependent unloading decision, so that the task completion time and the energy consumption of the user equipment are effectively reduced. And converging the flight path of the unmanned aerial vehicle and the unloading strategy depending on the task to be globally optimal in an alternate iteration mode. And finally, verifying the effectiveness of the UTDTOA algorithm through a simulation experiment.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles (UAVs), and in particular relates to a method and system for jointly optimizing UAV trajectories and offloading dependent tasks. Background Art

[0002] To increase the number of users served and reduce maximum end-to-end latency, drones should maintain an appropriate distance between each other. However, to improve mission transmission rates, light drones need to be as close as possible to enhanced drones. This inevitably leads to overlapping service areas. Therefore, it is necessary to find a balance between high coverage and low latency.

[0003] The key difference between MOPSO and traditional particle swarm optimization lies in its strategy for selecting global and individual optimal solutions. In traditional particle swarm optimization, each particle updates its velocity and direction based on its current optimal solution and the historical optimal solution found by the entire swarm. In single-objective optimization problems, the quality of solutions can be distinguished by comparing the objective function values, effectively guiding particles to converge toward the global optimal solution. Therefore, traditional particle swarm optimization performs particularly well in solving single-objective optimization problems.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows:

[0005] Light drones need to be as close as possible to enhanced drones, which inevitably leads to overlapping service areas. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a method and system for jointly optimizing UAV trajectories and dependent task offloading.

[0007] The present invention is implemented as follows: a method for jointly optimizing UAV trajectories and offloading dependent tasks includes:

[0008] Step 1: Randomly generate initial drone flight trajectories and task offloading strategies, and calculate global performance benchmark indicators based on these. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization.

[0009] Step 2: Under the premise of a fixed task offloading strategy, a multi-objective particle swarm optimization algorithm is used to optimize the UAV flight trajectory, with the goal of maximizing user coverage and reducing network latency while avoiding overlapping coverage areas.

[0010] Step 3: Based on the fixed UAV flight trajectory, the proximal strategy optimization algorithm is used to optimize the offloading strategy of the dependent tasks;

[0011] Step 4: Compare the change in the current global performance index with the previous round of indicators; if the change is less than the preset convergence threshold, stop the iteration and output the optimal result; otherwise, return to the "trajectory optimization stage" and continue the optimization cycle.

[0012] Furthermore, the drone:

[0013] It includes two categories: light drones and enhanced drones;

[0014] The present invention assumes that there are M light UAVs and one enhanced UAV in the scene, and the set UAV = {uav0, uav1, uav2, .... uav m ,...uav M} indicates that uav0 is an enhanced UAV;

[0015] In this area, N ground users are randomly distributed, and the user set is represented as U = {u1,u2,...u n ,...u N The present invention considers a discrete time system model, in which user u n Generate task set V according to Bernoulli distribution n ={v s ,v1,...v i ,...v I ,v e}, the arrival rate of the task is p task ; Among them, v s and v e are virtual tasks, which represent the start and end of a series of dependent tasks generated by the user; the dependency relationship between tasks is modeled using a directed acyclic graph; i Represents task v i To task v i+1 The amount of intermediate result data that needs to be transmitted; d i Represents task v i The input data size, c i Represents the computing task v i The CPU resources required for each bit of data; i =d i ×c i Represents the computing task v i The CPU cycles required for virtual task l s =l e =0;

[0016] Consider a three-dimensional Cartesian coordinate system model, user u n The position is defined as w n =(x n ,y n,0), and UAV m The position at time slot t is q m [t]=(x m [t],y m [t],H); the coordinates of the drone in the next time slot can be expressed as:

[0017] q m [t+1]=(x m [t]+t fly v m [t]cosθ m [t],y m [t]+t fly v m [t]sinθ m [t]) (1)

[0018] in, It's a UAV m The flight speed at time slot t, θ m [t]∈[0,2π] is the flight angle, t fly is the flight time of each time slot; in each time slot, the UAV flies t fly seconds, and it hovers for the remaining time to provide MEC services to ground users. In each time slot, the maximum moving distance of the UAV can be expressed as To ensure safety, the minimum distance between drones cannot be less than D min ; The starting position of each drone is known and is recorded as

[0019] Furthermore, the global performance benchmark indicators are calculated as follows:

[0020] (1) Communication model;

[0021] (2) Completion time model;

[0022] (3) Energy consumption model.

[0023] Furthermore, the communication model:

[0024] The performance of task offloading depends on the quality of the communication links between the user, the UAV, and the satellite, so it is necessary to analyze the communication links between these three.

[0025] Between the user and the drone

[0026] Assume that the communication link between the UAV and the ground user is always within the line of sight, and in each time slot, the channel state between the UAV and the ground user remains constant; user u n and UAVs m The distance between According to Shannon's formula, the transmission rate from user to drone and the transmission rate from drone to user are expressed as:

[0027]

[0028]

[0029] Among them, B e represents the communication bandwidth between the user and the UAV, p UE and p UAV denote the transmit power of the user and the UAV respectively, ρ0 is the channel gain when the transmission distance is 1 meter, and N0 denotes the Gaussian noise power under the channel;

[0030] Between drones

[0031] UAVs use line-of-sight links to communicate with each other, and do not consider non-line-of-sight communication; light UAVs m The distance between it and the enhanced UAV u0 is denoted as d m,0 [t]=||q m [t]-q0[t]||, then the data transmission rate between the light UAV and the enhanced UAV is:

[0032]

[0033] Among them, B u represents the communication bandwidth between UAVs;

[0034] Between drones and satellites

[0035] The UAV communicates with LEO via a line-of-sight link, so the transmission rate between the UAV and the satellite can be expressed as:

[0036]

[0037] in, and Denote the transmission rate from UAV to satellite and the transmission rate from satellite to UAV, respectively. s represents the communication bandwidth between the UAV and the LEO satellite, p LEO represents the transmission power of the LEO satellite, ρ up and ρ down They represent the channel gain between the UAV and the LEO satellite, and N represents the Gaussian white noise power in the channel.

[0038] Furthermore, the completion time model:

[0039] In each time slot, ground users can choose different offloading decisions according to specific needs, offloading tasks with dependencies to other nodes for processing, or completing them all locally; if the task v i The calculation and processing are performed at the ground user terminal. Since there is no need to transmit data, the task v i The computational latency of local processing can be expressed as:

[0040]

[0041] Among them, f local Indicates the CPU frequency of the ground user equipment; if the user sets the task v i To offload the task to the UAV for processing, it needs to go through two stages: task transmission and task calculation. In the task transmission stage, the user will i The transmission delay caused by offloading to the UAV can be expressed as:

[0042]

[0043] In the task calculation phase, the user sets the task v i The computational delay of the UAV can be expressed as:

[0044]

[0045] Among them, f m UAV m CPU frequency; if m≠0, it is a light drone, it can choose to set the task v i Further offload to the enhanced UAV or LEO satellite processing; if you choose to offload the calculation on the enhanced UAV, the task v i The resulting transmission delay can be expressed as:

[0046]

[0047] If you choose to set the task v i If the task is offloaded to the LEO satellite for processing, the transmission delay and computation delay of the task are expressed as:

[0048]

[0049] Among them, f s Indicates the CPU frequency of the LEO satellite;

[0050] Unlike independent tasks, the execution of dependent tasks depends on the computation results of their predecessor tasks. If dependent tasks are offloaded to the same processing node, no intermediate result data needs to be transmitted. However, if they are offloaded to different processing nodes, intermediate result data needs to be transmitted. Therefore, the transmission latency of intermediate result data can be expressed as:

[0051]

[0052] where R k′,k represents the transmission rate from processing node k′ to k, task v i There may be multiple predecessor tasks, so the time it takes to receive the execution results of all predecessor tasks can be expressed as:

[0053]

[0054] where pred(v i ) represents task v i The set of all predecessor tasks, Represents task v j Computational latency at the processing node;

[0055] The earliest starting time of a task depends not only on the completion time of its predecessor task, but also on the delay of task transmission to the processing node; i The earliest start time can be expressed as:

[0056]

[0057] In summary, ground user u n The completion time can be expressed as:

[0058]

[0059] The total delay of all users' tasks is the maximum delay of all users completing all tasks, that is,

[0060]

[0061] Furthermore, the energy consumption model:

[0062] When a task is processed on a local device, no data transmission is required, and the transmission energy consumption is 0. Therefore, the energy consumption is mainly generated by the energy consumption generated by the user device during the calculation process, that is, the calculation energy consumption:

[0063]

[0064] Where κ is the effective switching capacitance, which depends on the hardware chip;

[0065] When the task is offloaded to the edge server for processing, there is no computing process on the local user device. Therefore, the energy consumption is mainly the energy consumption generated by the user device sending to the drone, which is defined as:

[0066]

[0067] Therefore, user u n The total energy consumption can be expressed as:

[0068]

[0069] where α i =0 means the task is processed locally, α i =1 means that the task is processed by the MEC server; and the total energy consumption of completing all tasks is the sum of the energy consumption of all user devices to complete the tasks, that is,

[0070]

[0071] Another object of the present invention is to provide a system for jointly optimizing UAV trajectories and offloading dependent tasks, comprising:

[0072] The calculation module is used to randomly generate initial drone flight trajectories and task offloading strategies, and based on these, calculate global performance benchmark indicators. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization.

[0073] The flight trajectory optimization module is used to optimize the UAV flight trajectory using a multi-objective particle swarm optimization algorithm under the premise of a fixed task offloading strategy. The goal is to maximize user coverage and reduce network latency while avoiding overlapping coverage areas.

[0074] The offloading strategy optimization module is used to optimize the offloading strategy of dependent tasks based on the fixed UAV flight trajectory using the proximal strategy optimization algorithm;

[0075] The comparison module is used to compare the change in the current global performance index with that of the previous round of indicators. If the change is less than the preset convergence threshold, the iteration is stopped and the optimal result is output; otherwise, the algorithm returns to the "trajectory optimization stage" and continues the optimization cycle.

[0076] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method for jointly optimizing drone trajectories and dependent task offloading.

[0077] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the method for jointly optimizing drone trajectories and dependent task offloading.

[0078] Another object of the present invention is to provide an information data processing terminal, which is used to implement the system for jointly optimizing drone trajectories and dependent task offloading.

[0079] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0080] First, for emergency communication scenarios such as disaster areas where ground infrastructure is damaged or ineffective, an algorithm for jointly optimizing drone trajectories and dependent task offloading is proposed. This algorithm uses a problem decomposition approach to optimize the drone's flight trajectory and dependent task offloading strategy separately. In the drone trajectory optimization phase, a multi-objective particle swarm optimization algorithm is used to optimize the drone's flight trajectory, while maintaining a fixed task offloading strategy, to achieve the optimization goals of high coverage and low latency. In the dependent task offloading strategy optimization phase, a PPO algorithm is used to optimize task offloading decisions, while maintaining a fixed drone flight trajectory, effectively reducing task completion time and user device energy consumption. Ultimately, through alternating iterations, the drone's flight trajectory and dependent task offloading strategy converge to a global optimum.

[0081] For emergency communication scenarios such as disaster zones and uninhabited areas where ground infrastructure is damaged or inoperable, an algorithm is proposed to jointly optimize UAV trajectories and dependent task offloading. This algorithm uses a problem decomposition approach to optimize the UAV's flight trajectory and the offloading decisions for dependent tasks, reducing the computational complexity of a direct solution. Given an offloading strategy, a multi-objective particle swarm optimization (MOPSO) algorithm is used to optimize the UAV's flight trajectory, ensuring maximum coverage for ground users and minimizing the maximum end-to-end delay. Given a fixed UAV trajectory, a particle swarm optimization (PPO) algorithm is used to optimize the offloading decisions for dependent tasks, effectively reducing task completion time and user device energy consumption. Through alternating iterations, the UAV's flight trajectory and the dependent task offloading strategy converge to a global optimum. Finally, simulation experiments verify the effectiveness of the UTDTOA algorithm.

[0082] Second, the Space-Air-Ground Integrated Network (SAGIN) integrates the computing and communication capabilities of satellites, unmanned aerial vehicles (UAVs) and ground base stations to provide flexible and scalable computing services for resource-constrained user devices. The present invention proposes an algorithm for jointly optimizing drone trajectories and dependent task offloading in emergency communication scenarios such as disaster areas and uninhabited areas where ground infrastructure is damaged or fails. The algorithm can provide more optimized drone trajectories and more efficient offloading decisions, significantly reduce the system's latency and energy consumption, and fully meet users' requirements for service quality. The present invention aims to provide available low-altitude network connections in emergency communication situations, with lower costs and better services.

[0083] The present invention can process tasks at edge nodes close to users, reduce data transmission delays and reduce the load on the core network, by constructing a task dependency model based on a directed acyclic graph (DAG). The algorithm designed by the present invention optimizes the drone flight trajectory and task offloading decisions respectively by problem decomposition to reduce the complexity of the overall solution. First, under the premise of a given task offloading strategy, the flight trajectory of the drone is optimized to serve as many users as possible, while reducing the overlap of coverage areas and minimizing the maximum end-to-end delay. Secondly, when the drone trajectory is determined, the offloading scheme of dependent tasks is optimized to effectively reduce the task completion time and equipment energy consumption. Finally, by alternately iteratively optimizing the two sub-problems, the UTDTOA gradually converges to the global optimum.

[0084] With the prevalence of aerial devices such as airplanes and drones, users' demand for stable and efficient communications is also growing. Although fifth-generation mobile communications have brought significant network improvements to many regions, existing networks mainly rely on ground infrastructure, and construction costs are high in complex environments such as mountains, oceans, and remote uninhabited areas. Integrated air-space-ground networks, by breaking through the limitations of the ground to provide users with reliable wide-area coverage services, are gradually becoming one of the key technologies for the future development of communications. Mobile devices such as satellites and drones face serious energy consumption issues during operation. These devices not only consume a large amount of energy to maintain network connections, but also consume energy to process computing tasks. Excessive data transmission and computing will lead to excessive energy consumption, seriously affecting the operating time and service life of the equipment. The present invention effectively reduces the load on terminal devices and optimizes the allocation of computing resources by offloading computationally intensive or delay-sensitive tasks from terminal devices to edge nodes or cloud data platforms. It can significantly reduce the distance of data transmission, reduce network latency, and improve the efficiency and performance of the entire network. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 This is a flow chart of a method for jointly optimizing drone trajectories and offloading dependent tasks provided by an embodiment of the present invention.

[0086] Figure 2 This is a block diagram of the system structure for jointly optimizing drone trajectories and offloading dependent tasks provided by an embodiment of the present invention.

[0087] Figure 3 This is a system model diagram provided by an embodiment of the present invention.

[0088] Figure 4 This is a flow chart of the UTDTOA algorithm provided by an embodiment of the present invention.

[0089] Figure 5 This is a framework diagram of the PPO algorithm provided by an embodiment of the present invention.

[0090] Figure 6 This is a comparison chart of the convergence performance of algorithms with different learning rates provided by an embodiment of the present invention.

[0091] Figure 7 This is a drone trajectory map provided by an embodiment of the present invention.

[0092] Figure 8 This is a diagram showing the impact of the number of ground users on the system Z value provided by an embodiment of the present invention.

[0093] Figure 9 This is a diagram showing the impact of the amount of task data on the system Z value provided by an embodiment of the present invention.

[0094] Figure 10 3 is a diagram showing the influence of the computing capability of the light UAV provided by an embodiment of the present invention on the system Z value.

[0095] Figure 11 3 is a diagram showing the impact of the computing capability of the enhanced UAV provided by an embodiment of the present invention on the system Z value. DETAILED DESCRIPTION

[0096] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0097] like Figure 1 As shown, a method for jointly optimizing drone trajectories and offloading dependent tasks provided by an embodiment of the present invention includes the following steps:

[0098] S101: Randomly generate initial drone flight trajectories and task offloading strategies, and calculate global performance benchmark indicators based on these. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization.

[0099] S102: Under the premise of a fixed task offloading strategy, a multi-objective particle swarm optimization algorithm is used to optimize the UAV flight trajectory, with the goal of maximizing user coverage and reducing network latency while avoiding overlapping coverage areas;

[0100] S103: Based on the fixed UAV flight trajectory, the proximal strategy optimization algorithm is used to optimize the offloading strategy of the dependent tasks;

[0101] S104: Compare the change range of the current global performance index with that of the previous round of indexes; if the change range is less than the preset convergence threshold, stop the iteration and output the optimal result; otherwise, return to the "trajectory optimization stage" and continue the optimization cycle.

[0102] like Figure 2 As shown, an embodiment of the present invention provides a system for jointly optimizing drone trajectories and offloading dependent tasks, including:

[0103] The calculation module is used to randomly generate initial drone flight trajectories and task offloading strategies, and based on these, calculate global performance benchmark indicators. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization.

[0104] The flight trajectory optimization module is used to optimize the UAV flight trajectory using a multi-objective particle swarm optimization algorithm under the premise of a fixed task offloading strategy. The goal is to maximize user coverage and reduce network latency while avoiding overlapping coverage areas.

[0105] The offloading strategy optimization module is used to optimize the offloading strategy of dependent tasks based on the fixed UAV flight trajectory using the proximal strategy optimization algorithm;

[0106] The comparison module is used to compare the change in the current global performance index with that of the previous round of indicators. If the change is less than the preset convergence threshold, the iteration is stopped and the optimal result is output; otherwise, the algorithm returns to the "trajectory optimization stage" and continues the optimization cycle.

[0107] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method for jointly optimizing drone trajectories and dependent task offloading.

[0108] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the method for jointly optimizing drone trajectories and dependent task offloading.

[0109] Another object of the present invention is to provide an information data processing terminal, which is used to implement the system for jointly optimizing drone trajectories and dependent task offloading.

[0110] The present invention is specifically implemented:

[0111] 1.1 System Model and Problem Modeling

[0112] The present invention focuses on the application of integrated air-ground-space network in disaster areas or field environments. Figure 1 It shows the position changes caused by drone flight at adjacent moments. In this scenario, the LEO satellite equipped with MEC server, as an important part of the space-based network, provides a wider range of computing services to ground users in its coverage area. The LEO coverage model is consistent with the low-orbit satellite coverage model. As an aerial edge computing node, the drone flies at an altitude of H meters above the ground. Each drone is equipped with a MEC server. According to the computing power, drones are divided into two categories: light drones and enhanced drones. Since light drones have limited computing resources and cannot complete tasks with large computational loads in a short time, some tasks can be offloaded to enhanced drones with stronger computing power to improve the overall task processing efficiency. The present invention assumes that there are M light drones and one enhanced drone in the scene, and the set UAV = {uav0, uav1, uav2,....uav m ,...uav M}, where uav0 represents an enhanced UAV.

[0113] In this area, N ground users are randomly distributed, and the user set is represented as U = {u1,u2,...u n ,...u N The present invention considers a discrete time system model where in each time slot, user u n Generate task set V according to Bernoulli distribution n ={v s ,v1,...v i ,...v I ,v e}, the arrival rate of the task is p task Among them, v s and v eare virtual tasks, representing the start and end of a series of dependent tasks generated by the user. The dependency relationship between tasks is modeled using a directed acyclic graph. i Represents task v i To task v i+1 The amount of intermediate result data that needs to be transmitted. i Represents task v i The input data size, c i Represents the computing task v i The CPU resources required for each bit of data. i =d i ×c i Represents the computing task v i The CPU cycles required for virtual task l s =l e =0.

[0114] Consider a three-dimensional Cartesian coordinate system model, user u n The position is defined as w n =(x n ,y n ,0), and UAV m The position at time slot t is q m [t]=(x m [t],y m [t],H). The coordinates of the drone in the next time slot can be expressed as:

[0115] q m [t+1]=(x m [t]+t fly v m [t]cosθ m [t],y m [t]+t fly v m [t]sinθ m [t]) (1)

[0116] in, It's a UAV m The flight speed at time slot t, θ m [t]∈[0,2π] is the flight angle, t fly is the flight time of each time slot. In each time slot, the UAV flies t fly After 1 second, it hovers for the remaining time to provide MEC services to ground users. The maximum moving distance of the UAV in each time slot can be expressed as To ensure safety, the minimum distance between drones cannot be less than D min The starting position of each UAV is known and is recorded as

[0117] 1.1.1 Communication model

[0118] Due to the limited computing power of terminal devices, users tend to offload tasks to drones or satellites with stronger computing power to improve task execution efficiency and reduce local energy consumption. The performance of task offloading depends on the quality of the communication link between the user, drone, and satellite. Therefore, it is necessary to analyze the communication link between these three.

[0119] Between the user and the drone

[0120] It is assumed that the communication link between the UAV and the ground user is always within the line of sight, and the channel state between the UAV and the ground user remains constant in each time slot. n and UAVs m The distance between According to Shannon's formula, the transmission rate from user to drone and the transmission rate from drone to user are expressed as:

[0121]

[0122] Among them, B e represents the communication bandwidth between the user and the UAV, p UE and p UAV They represent the transmission power of the user and the UAV respectively, ρ0 is the channel gain when the transmission distance is 1 meter, and N0 represents the Gaussian noise power under the channel.

[0123] Between drones

[0124] UAVs use line-of-sight links to communicate with each other, and do not consider non-line-of-sight communication. m The distance between it and the enhanced UAV u0 is denoted as d m,0 [t]=||q m [t]-q0[t]||, then the data transmission rate between the light UAV and the enhanced UAV is:

[0125]

[0126] Among them, B u Indicates the communication bandwidth between drones.

[0127] Between drones and satellites

[0128] The UAV communicates with LEO via a line-of-sight link, so the transmission rate between the UAV and the satellite can be expressed as:

[0129]

[0130] in, and Denote the transmission rate from UAV to satellite and the transmission rate from satellite to UAV, respectively. s represents the communication bandwidth between the UAV and the LEO satellite, p LEO represents the transmission power of the LEO satellite, ρ up and ρ down They represent the channel gain between the UAV and the LEO satellite, and N represents the Gaussian white noise power in the channel.

[0131] 1.1.2 Completion Time Model

[0132] In each time slot, ground users can choose different offloading decisions according to specific needs, offloading tasks with dependencies to other nodes for processing, or completing them all locally. i The calculation and processing are performed at the ground user terminal. Since there is no need to transmit data, the task v i The computational latency of local processing can be expressed as:

[0133]

[0134] Among them, f local Indicates the CPU frequency of the ground user equipment. If the user sets the task v i To offload the task to the UAV for processing, it needs to go through two stages: task transmission and task calculation. In the task transmission stage, the user will i The transmission delay caused by offloading to the UAV can be expressed as:

[0135]

[0136] In the task calculation phase, the user sets the task v i The computational delay of the UAV can be expressed as:

[0137]

[0138] Among them, f m UAV m If m≠0, it is a light drone, and it can choose to set the task v i Further offload to the enhanced UAV or LEO satellite processing. If you choose to offload the calculation on the enhanced UAV, the task v i The resulting transmission delay can be expressed as:

[0139]

[0140] If you choose to set the task v iIf the task is offloaded to the LEO satellite for processing, the transmission delay and computation delay of the task are expressed as:

[0141]

[0142] Among them, f s Indicates the CPU frequency of the LEO satellite.

[0143] Unlike independent tasks, the execution of dependent tasks depends on the computational results of their predecessor tasks. If dependent tasks are offloaded to the same processing node, no intermediate result data needs to be transferred; however, if they are offloaded to different processing nodes, intermediate result data needs to be transferred. Therefore, the transmission delay of intermediate result data can be expressed as:

[0144]

[0145] where R k′,k represents the transmission rate from processing node k′ to k. In addition, task v i There may be multiple predecessor tasks, so the time it takes to receive the execution results of all predecessor tasks can be expressed as:

[0146]

[0147] where pred(v i ) represents task v i The set of all predecessor tasks, Represents task v j Computational latency at the processing node.

[0148] The earliest start time of a task depends not only on the completion time of its predecessor task, but also on the delay of task transmission to the processing node. i The earliest start time can be expressed as:

[0149]

[0150] In summary, ground user u n The completion time can be expressed as:

[0151]

[0152] The total delay of all users' tasks is the maximum delay of all users completing all tasks, that is,

[0153]

[0154] 1.1.3 Energy Consumption Model

[0155] The energy consumption of computing tasks includes energy consumed by the local device and other edge nodes. This paper focuses only on the energy consumption of the local device. Generally speaking, user device energy consumption mainly includes computing energy and transmission energy. Different offloading decisions selected by the user will result in different energy consumption.

[0156] When a task is processed on a local device, no data transmission is required, and the transmission energy consumption is 0. Therefore, the energy consumption is mainly generated by the energy consumption generated by the user device during the calculation process, that is, the calculation energy consumption:

[0157]

[0158] Where κ is the effective switching capacitance, which depends on the hardware chip.

[0159] When the task is offloaded to the edge server for processing, there is no computing process on the local user device. Therefore, the energy consumption is mainly the energy consumption generated by the user device sending to the drone, which is defined as:

[0160]

[0161] Therefore, user u n The total energy consumption can be expressed as:

[0162]

[0163] where α i =0 means the task is processed locally, α i =1 means the task is processed by the MEC server. The total energy consumption of completing all tasks is the sum of the energy consumption of all user devices completing tasks, that is,

[0164]

[0165] 1.1.4 Problem Description

[0166] The present invention defines the optimization target as the product of latency and energy consumption, i.e., Z = T × E. The larger the Z value, the higher the overall overhead and the worse the service quality; conversely, the smaller the Z value, the better the service quality. In order to better describe the task offloading decision, two variables α are defined i and β i The processing location of the task can be described as:

[0167]

[0168] The variable α∈{0,1} represents the task processing method, and its value determines the specific execution location of the task. When α=0, the task is processed locally on the user device; when α=1, the task is offloaded to an edge node for processing. The variable β={0,1,2} represents the specific location of the edge node to which the task is offloaded, where β=0 indicates that the task is processed on the current drone, β=1 indicates that the task is transferred from the current drone to the enhanced drone for processing, and β=2 indicates that the task is transferred from the current drone to the LEO satellite for processing.

[0169] Based on the above analysis, the research problem of this invention can be expressed as:

[0170]

[0171] Among them: C1 and C2 represent the task offloading decision rules. C3 indicates that the first and last tasks in the task set must be calculated and processed locally. C4 stipulates that the initial position of the drone has been pre-set, and C5 limits the movement distance of the drone in a single time slot to not exceed its maximum movement distance. For safety reasons, C6 requires that the distance between drones must not be less than the safety distance D min In addition, the flight angle and speed of the drone need to meet the restrictions of C7 and C8, and C9 limits the flight area of ​​the drone. C10 constraints depend on the execution order between tasks, j < i means task v j It is task v i C11 indicates that the completion time of a task generated by a single user must not exceed its tolerable delay.

[0172] 1.2 Research on UAV trajectory planning and task offloading strategies

[0173] In the air-ground integrated network scenario, for the joint optimization problem of drone flight trajectory and dependent task computing offloading, the present invention uses the problem decomposition method to split the complex problem into two sub-problems to be solved separately: drone trajectory planning and dependent task offloading. First, optimize the drone's flight trajectory to reduce coverage overlap while serving multiple users and minimize the maximum end-to-end delay. Secondly, based on the given drone trajectory, optimize the offloading decision of dependent tasks to minimize the task completion time and user equipment energy consumption. Finally, the global optimum is achieved through iterative optimization. The UTDTOA algorithm flow chart is as follows: Figure 4 shown.

[0174] The UTDTOA algorithm can be divided into the following steps:

[0175] Step 1: Randomly generate initial drone flight trajectories and task offloading strategies, and calculate global performance benchmarks based on these. These metrics include drone coverage, task execution latency, and device energy consumption. They are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization.

[0176] Step 2: Under the premise of a fixed task offloading strategy, a multi-objective particle swarm optimization algorithm is used to optimize the UAV flight trajectory, with the goal of maximizing user coverage and reducing network latency while avoiding overlapping coverage areas.

[0177] Step 3: Based on the fixed UAV flight trajectory, the proximal strategy optimization algorithm is used to optimize the offloading strategy of dependent tasks to reduce task latency and equipment energy consumption.

[0178] Step 4: Compare the change in the current global performance indicator with the change in the previous round of indicators. If the change is less than the preset convergence threshold, the iteration stops and the optimal result is output; otherwise, return to the "trajectory optimization stage" and continue the optimization cycle.

[0179] 1.2.1 UAV trajectory planning based on multi-objective particle swarm

[0180] To maximize the number of users served and minimize maximum end-to-end latency, drones should maintain an appropriate distance between each other. However, to improve mission transmission rates, light drones need to be as close as possible to enhanced drones. This inevitably leads to overlapping service areas. Therefore, it is necessary to find a balance between high coverage and low latency. This section uses a multi-objective particle swarm optimization algorithm to optimize the drone's flight trajectory to achieve both high coverage and low latency.

[0181] The key difference between MOPSO and traditional particle swarm optimization lies in its strategy for selecting global and individual optimal solutions. In traditional particle swarm optimization, each particle updates its velocity and direction based on its current optimal solution and the historical optimal solution found by the entire swarm. In single-objective optimization problems, the quality of solutions can be distinguished by comparing the objective function values, effectively guiding particles to converge toward the global optimal solution. Therefore, traditional particle swarm optimization performs particularly well in solving single-objective optimization problems.

[0182] In multi-objective problems, the solution set consists of multiple Pareto optimal solutions rather than a single global optimal solution. Therefore, MOPSO makes special adjustments in the selection of individual optimal solutions and global optimal solutions to adapt to the characteristics of multi-objective problems. In MOPSO, the individual optimal solution is a set consisting of multiple solutions, and the elements in the set represent the optimal solutions found by the particles during the search process. In addition, the global optimal solution is selected from an external archive that saves the Pareto frontier solution set currently searched. In this way, MOPSO can effectively guide the particle swarm to perform multi-objective optimization and find the optimal solution that can balance multiple objectives. In the MOPSO optimization process, the constructed multi-objective function mainly includes the following three parts:

[0183] 1) Maximize the number of user devices served: Ensure that more users can access the network. This optimization goal can be expressed as:

[0184]

[0185] Where N is the total number of ground users, χ n represents the ground user u n Is it covered by the drone? If so, n =1, otherwise χ n =0.

[0186] 2) Minimize the coverage ratio between drones: reduce the waste of coverage area. This optimization goal can be expressed as:

[0187]

[0188] Among them, M represents the total number of drones, A m and A k Respectively represent uav m and UAV k The coverage area, A m,k Indicates the overlapping area of ​​the two coverage areas.

[0189] 3) Minimize the maximum end-to-end delay: By rationally designing the UAV’s flight path, we can improve the mission’s transmission rate and reduce the delay. This optimization objective can be expressed as:

[0190]

[0191] In summary, the multi-objective optimization problem can be expressed as:

[0192] minZ1={-h1,h2,h3} (27)

[0193] Among them, -h1 means that the goal of maximizing the number of service users is transformed into a minimization problem.

[0194] In the MOPSO algorithm, each particle Represents the trajectory of a group of drones, and updates the speed and position of the drones using the following formula:

[0195]

[0196] in, and They represent the velocity and position of particle m at the tth iteration, is the inertia weight, c1 and c2 are learning factors, r1 and r2 are uniformly distributed random numbers, and p m,best represents the optimal position of the particle itself in history, g best represents the global optimal position. The external archive is updated in each iteration according to the non-dominated sorting rule to store the current set of optimal solutions. When the maximum number of iterations is reached or the particle swarm converges, the non-dominated solution set in the archive is output as the final optimization result.

[0197] Through the above analysis, a UAV trajectory planning algorithm based on multi-objective particle swarm optimization is proposed, and its pseudo code is shown in Table 1.

[0198] Table 1 Pseudocode of UAV trajectory planning algorithm

[0199]

[0200]

[0201] The entire algorithm is divided into the following steps:

[0202] Step 1: Randomly generate a particle swarm. Randomly generate a particle swarm and initialize the velocity and position of each particle. At the same time, create an external archive to store non-dominated solutions.

[0203] Step 2: Calculate the multi-objective function values. For each particle, calculate the corresponding multi-objective function value (e.g., maximize coverage, minimize delay). Store the non-dominated solutions of the current particle in an external archive to ensure that the potential optimal solution is preserved.

[0204] Step 3: Update the particle's velocity and position. Based on the particle's individual optimal solution and the global optimal solution, the particle's velocity and position are updated. During this process, constraints are imposed on the drone's trajectory to ensure it always remains within the permitted flight range.

[0205] Step 4: Check and update the external archive. Filter the solutions stored in the external archive, remove the dominated solutions, and retain the updated non-dominated solution set to ensure the diversity and superiority of the solution set.

[0206] Step 5: Determine the termination condition. Check whether the termination condition is met: if the maximum number of iterations is reached or the particle swarm has converged, stop the algorithm; otherwise, return to step 2 to continue optimization.

[0207] Step 6: Output the optimal solution. After the algorithm is completed, the final optimal drone trajectory set is extracted from the external archive as the solution that meets the multi-objective optimization.

[0208] 1.2.2 Design of Dependent Task Offloading Algorithm Based on PPO

[0209] Given a given UAV trajectory, the offloading strategy for dependent tasks is optimized by comprehensively considering the computing power of both the satellite and the UAV to minimize total user latency and total energy consumption. To more clearly describe this problem, this section models it as a Markov decision process. In the MDP framework, the agent interacts with the environment and continuously adjusts its strategy to maximize the cumulative reward. To achieve this, specific definitions of state, action, and reward are required. These definitions cover multiple parameters of the ground user terminal, UAV, and satellite, as well as the real-time state of the network, providing a complete description and decision-making basis for optimizing the task offloading strategy.

[0210] 1) State collection

[0211] The state set in the Markov decision process is used to describe the current task characteristics and the network status of all device nodes. For time slot t = {1, 2, ..., T}, the state set is defined as:

[0212] S={s t |s t ={d i ,c i ,o i ,q m [t],f local ,f m ,f s}},t∈T (30)

[0213] Among them, d i Indicates the data size of the task, c i Indicates the CPU cycles required to process a unit bit task, o i Indicates the amount of data of the intermediate result of the task, q m [t] represents the position of the UAV at time t, f local Indicates the CPU frequency of the user device, f m UAV m CPU frequency, f s Indicates the CPU frequency of the LEO satellite.

[0214] 2) Action Set

[0215] The agent needs to select a set of actions based on the current state to complete the computational processing of the task. The action set contains the unloading position of the current task in the time slot t. Therefore, the action set can be expressed as:

[0216] A={a t |a t ={b i,t}},i∈I,t∈T (31)

[0217] Among them, b i,t Represents task v i Offloading location selection at time t. Specifically, each element in the action set can correspond to one of the following options: the task is executed locally at the ground user terminal, offloaded to a UAV for processing, or offloaded to a satellite for processing. By making decisions about the action set within each time slot, the agent can dynamically optimize the task offloading strategy to minimize total latency and total energy consumption.

[0218] 3) Reward Function

[0219] The reward function is used to measure the quality of the offloading strategy, and its definition is based on the task's latency and energy consumption. The expression of the reward function is:

[0220]

[0221] in, Represents task v i The completion deadline, Represents task v i The actual completion time, Represents the total energy consumption of the user device executing the task. C is a constant used to encourage the model to accumulate rewards and continue. By applying a logarithmic transformation to the rewards for latency and energy consumption, reward fluctuations can be smoothed to prevent model instability caused by excessive feedback. If the task is completed on time, the system receives a positive reward; if it exceeds the deadline, a penalty is incurred.

[0222] Based on the aforementioned MDP framework, a proximal policy optimization algorithm is used for task offloading. PPO effectively avoids policy collapse during the optimization process by restricting the range of variation between the old and new policies by clipping the objective function. Figure 5 The proposed PPO-based task offloading algorithm framework is demonstrated.

[0223] The PPO algorithm framework consists of three networks: the new action network, the old action network, and the evaluation network. The parameters θ and θ′ represent the new and old action networks, respectively, while the parameters of the evaluation network are represented by ζ. The evaluation network primarily assesses the value of the current state and provides guidance for policy updates. The new action network generates the next action based on the state of the current time slot. During algorithm training, the new action network undergoes multiple iterative updates. The introduction of the old action network ensures that sampled data remains suitable for policy updates during policy updates, thereby enhancing algorithm stability and network convergence. In addition to the three core networks mentioned above, the PPO algorithm architecture also includes an environment module and an experience pool module. The environment module simulates the real physical world and generates the next state by observing the current environment state and receiving action feedback. The experience pool module acts as a memory component, storing information such as historical states, actions, rewards, and policy probabilities, providing rich and reliable training samples for subsequent policy optimization.

[0224] During the training process, the new action network first receives state information from the environment and generates corresponding actions. The received state information needs to be initially normalized before entering the strategy update phase. Subsequently, by comparing the ratio of the new and old strategies and adjusting the clipping parameter, the current evaluation network is updated. Next, the evaluation network is evaluated to minimize the loss function L c (ζ) is the target, and the model parameters θ are updated. Finally, the state information, actions, and rewards are stored in the cache to prepare for subsequent cyclic training. Specifically, the loss function of the action network is defined as:

[0225]

[0226] Among them, P t (θ)=π θ (a t |s t ) / π θ′ (a t |s t ) is expressed as the ratio of the new and old strategies, ε is the clipping parameter, is the advantage function, which evaluates the performance of the current action compared to the average level. The introduction of the pruning operation limits the range of policy changes and avoids instability caused by too fast policy updates.

[0227] In order to improve the robustness of policy updates, the PPO algorithm introduces General Advantage Estimation (GAE). GAE calculates the advantage function by balancing the deviation and variance. Its calculation formula is as follows:

[0228]

[0229] Among them, δ t =r t (s t ,a t )+γV(s t+1 )-V(s t ) represents the time-dependent differential error, γ is the discount factor, and υ is the coefficient used to weigh the deviation and variance, which is controlled between 0 and 1. The generalized advantage estimation method can smooth the policy update process and improve the update effect. In addition, the state value function V(s t ) is defined as:

[0230]

[0231] It represents the expected value of the future cumulative reward starting from the current state. In order to optimize the state value, the PPO algorithm also introduces an evaluation network, which is trained by minimizing the following objective function:

[0232] L c (ζ)=E[V ζ (s t+1 )-V ζ (s t )] 2 (36)

[0233] Through the above analysis, in order to obtain the optimal offloading decision, a dependent task offloading algorithm based on PPO is proposed, and its pseudo code is shown in Table 2.

[0234] Table 2 Pseudocode of the PPO-based dependent task offloading algorithm

[0235]

[0236]

[0237] 2. Application Examples

[0238] Based on the above analysis, the present invention proposes an algorithm for jointly optimizing UAV trajectories and dependent task offloading, and achieves the improvement of overall network performance by alternately optimizing the flight trajectory of the UAV and the task offloading strategy. In the trajectory optimization stage, under the premise of a fixed task offloading strategy, MOPSO is used to optimize the flight trajectory of the UAV. The optimization goal is to provide low-latency services to as many users as possible while reducing the overlap of service areas between UAVs. In the offloading strategy optimization stage, under the premise of a fixed UAV flight trajectory, the dependencies between tasks are comprehensively considered and the offloading strategy of dependent tasks is dynamically adjusted. The optimization goal is to minimize task delay and equipment energy consumption. Finally, in the alternating optimization process, the global performance indicators need to be re-evaluated each time the trajectory optimization and offloading optimization are completed to ensure that the optimization direction is consistent with the goal. If the change in the global performance indicator is lower than the preset convergence threshold, the optimization process is terminated, and the optimal UAV trajectory set and task offloading strategy set are output.

[0239] In summary, the design and application embodiments of the present invention are as follows: the pseudo code of the UAV trajectory and task offloading joint optimization algorithm is shown in Table 3.

[0240] Table 3 Pseudocode of the joint optimization algorithm for trajectory planning and task offloading of UAV collaboration

[0241]

[0242] The UTDTOA algorithm improves task execution efficiency by alternately optimizing the drone's trajectory and task offloading strategy. First, with the task offloading strategy fixed, the drone's trajectory is updated using a multi-objective particle swarm optimization algorithm. Then, with the drone's trajectory fixed, the PPO algorithm optimizes the task offloading decision. After each iteration, a global performance metric is calculated and an update is made if the solution outperforms the current optimal solution. The algorithm terminates when the convergence criteria are met or the maximum number of iterations is reached, returning the optimal trajectory and offloading strategy.

[0243] The application examples of the present invention were constructed using Python 3.9.0, and all experiments were run on a Windows 10 operating system. The computer hardware used in the experiments included a 64-bit Intel(R) Celeron(R) N4000 CPU with a main frequency of 1.10 GHz and 8 GB of RAM. The main parameters used in the application examples are shown in Table 4.

[0244] Table 4 Parameter settings

[0245]

[0246] The application embodiment of the present invention is deployed in a 300m x 300m square area, where 1 to 30 ground users, 3 light UAVs, 1 enhanced UAV, and 1 LEO satellite are randomly deployed. The UAVs fly at an altitude of 100m, and the LEO satellite orbits at an altitude of 784km. The UAVs' initial positions are fixed, with coordinates of [300,300], [300,0], [0,300], and [0,0]. The UAVs have a maximum movement speed of 30m / s, and they must maintain a safe distance of at least 1m between each other. The ground user randomly generates 3 to 10 tasks with dependencies, with the task data size ranging from 20KB to 50KB, and the computation cycle required to calculate each bit of the task is 1000 cycles / bit. The resulting data size after the task is completed is randomly distributed between 10KB and 20KB, and the maximum tolerable latency of the ground user equipment is set to 1000ms. To account for the differences in computing power for task processing, the system's computing performance is as follows: 200M cycles / s for ground user equipment, 1.2G cycles / s for light UAVs, 1.8G cycles / s for enhanced UAVs, and 3G cycles / s for LEO satellites. Furthermore, to ensure efficient task transmission, the UAV's communication channel bandwidth is set to 300MHz, while the LEO satellite's available bandwidth is 20GHz.

[0247] 3. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0248] The present invention verifies the effectiveness of the proposed algorithm by analyzing the results of the application examples. First, the influence of different learning rates on the convergence performance of the algorithm is explored. Figure 6 As shown in the figure, we can observe that when the learning rate is 0.0001, the algorithm converges fastest and stabilizes after approximately 400 iterations. When the learning rate is 0.001, the algorithm converges more slowly, reaching a stable state after approximately 500 iterations. When the learning rate is 0.01, the algorithm converges the slowest, and the initial loss value fluctuates greatly. Therefore, the learning rate is set to 0.0001 in subsequent experiments.

[0249] Figure 7The optimized drone flight paths are shown. The orange triangles, green squares, and red diamonds represent the trajectories of the three light drones, respectively, while the blue circle represents the flight path of the enhanced drone. Five-pointed stars indicate the locations of ground users, with darker colors indicating greater data volumes. It can be seen that all drones flew from their initial positions toward high-density areas. The flight path of UAV0 is further to the right than that of UAV1 horizontally and further upward than that of UAV2 vertically. This is because the enhanced drone's greater computing power requires collaboration with the light drone during flight to achieve more efficient task allocation and collaborative computing. This collaborative mechanism not only optimizes resource utilization but also provides users with better service.

[0250] In addition, in order to comprehensively evaluate the effectiveness of the algorithm of the present invention, it is compared with the following algorithms:

[0251] Individual task offloading optimization (fixed drone trajectory): Under the premise that the drone trajectory is fixed, the PPO algorithm is used to optimize the task offloading strategy.

[0252] Individual trajectory optimization (fixed task offloading strategy): Optimize the flight trajectory of the drone under the premise of a fixed task offloading strategy.

[0253] Fixed trajectory + random offloading: The predetermined drone flight trajectory is used without trajectory optimization, and the target server (terminal device, drone or LEO satellite) is randomly selected for task offloading.

[0254] Figure 8 The system Z value performance of different algorithms under different numbers of users is shown. When the number of users is 20, the system Z values ​​of the task offloading algorithm alone and the trajectory optimization algorithm alone are 13803 and 14723, respectively, while the system Z value of the fixed trajectory + random offloading algorithm is as high as 17850. The system Z value of the UTDTOA algorithm is 12053, which is significantly lower than the system Z values ​​of the other three algorithms. As the number of ground users increases, the system Z value of the algorithm also shows a clear growth trend. This is because the increase in the number of ground users will lead to an increase in the number of task requests in the system, resulting in higher latency and energy consumption, and ultimately causing a cumulative increase in the system Z value. The system Z value of the UTDTOA algorithm is always lower than the other three algorithms, which shows the excellent performance of the UTDTOA algorithm in handling complex task scenarios.

[0255] Figure 9The system Z-value performance of different algorithms under different task data volumes is demonstrated when there are 10 ground users and each user randomly generates 3 to 10 tasks. Taking a task data volume of 40KB as an example, the system Z-value of the task offloading algorithm alone is 10094, while the system Z-values ​​of the trajectory optimization algorithm alone and the fixed trajectory + random offloading algorithm are 11521 and 12301, respectively. The system Z-value of the UTDTOA algorithm is 9531.7, significantly lower than that of the other three algorithms. As the task data volume increases, the system Z-value of the algorithm also shows an upward trend. This is because an increase in the task data volume not only significantly prolongs the task processing time but also leads to increased energy consumption, which in turn increases the overall system Z-value. The system Z-value of the UTDTOA algorithm is consistently lower than that of the other three algorithms, demonstrating excellent performance. This demonstrates the excellent adaptability of the UTDTOA algorithm in high-data-volume task scenarios.

[0256] Figure 10 The performance of the system Z values ​​of different algorithms under different light UAV computing power is shown. When the computing power of the light UAV is 0.4G cycles / s, the system Z value of the UDTOA algorithm is 4166.3, and the system Z values ​​of the other three algorithms are all greater than 4500. The system Z value of the UDTOA algorithm is significantly lower than the other three algorithms, and the optimization effect is the best. When the computing power of the light UAV is 1.6G cycles / s, the difference in the system Z values ​​of the four algorithms is small, but the UDTOA algorithm still shows a slight advantage and is better than the other three algorithms. As the computing power of the light UAV increases, the system Z values ​​of the algorithms all show a downward trend. This is because the increase in the computing power of the light UAV will shorten the completion time of the task. The system Z value of the UDTOA algorithm is always lower than the other three algorithms, which shows the effectiveness of the UDTOA algorithm.

[0257] Figure 11The performance of different algorithms under different enhanced UAV computing capabilities is shown. When the enhanced UAV's computing power reaches 1.2G cycles / s, the UTDTOA algorithm achieves a system Z value of 1289.3, while the other three algorithms all achieve system Z values ​​greater than 1600. The UTDTOA algorithm achieves a significantly lower system Z value than the other three algorithms, demonstrating the best optimization results. When the enhanced UAV's computing power reaches 2.0G cycles / s, the differences in the system Z values ​​among the four algorithms are relatively small, but the UTDTOA algorithm still exhibits a certain performance advantage, with a system Z value slightly lower than the other three algorithms. As the computing power of the enhanced UAV increases, the system Z values ​​of the algorithms all show a downward trend. Similar to light UAVs, the increased computing power of the enhanced UAV can also reduce task processing latency. The increased computing power of the enhanced UAV has a more significant impact on system performance. The UTDTOA algorithm consistently maintains a lower system Z value than the other three algorithms, further demonstrating its effectiveness.

[0258] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It will be understood by those skilled in the art that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such code is provided on a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier.

[0259] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for jointly optimizing UAV trajectories and offloading dependent tasks, characterized in that: The following steps are involved: Step 1: Randomly generate initial drone flight trajectories and task offloading strategies, and calculate global performance benchmark indicators based on these. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization. Step 2: Under the premise of a fixed task offloading strategy, a multi-objective particle swarm optimization algorithm is used to optimize the UAV flight trajectory, with the goal of maximizing user coverage and reducing network latency while avoiding overlapping coverage areas. Step 3: Based on the fixed UAV flight trajectory, the proximal strategy optimization algorithm is used to optimize the offloading strategy of the dependent tasks; Step 4: Compare the change in the current global performance indicator with the change in the previous round of indicators. If the change is less than the preset convergence threshold, stop the iteration and output the optimal result. Otherwise, return to the "trajectory optimization stage" and continue the optimization cycle.

2. The method for jointly optimizing UAV trajectories and offloading dependent tasks according to claim 1, wherein: The drone: It includes two categories: light drones and enhanced drones; The present invention assumes that there are M light UAVs and one enhanced UAV in the scene, and the set UAV = {uav0, uav1, uav2, .... uav m ,...uav M } indicates that uav0 is an enhanced UAV; In this area, N ground users are randomly distributed, and the user set is represented as U = {u1,u2,...u n ,...u N The present invention considers a discrete time system model, in which user u n Generate task set V according to Bernoulli distribution n ={v s ,v1,...v i ,...v I ,v e }, the arrival rate of the task is p task [66]; Among them, v s and v e are virtual tasks, which represent the start and end of a series of dependent tasks generated by the user; the dependency relationship between tasks is modeled using a directed acyclic graph; i Represents task v i To task v i+1 The amount of intermediate result data that needs to be transmitted; d i Represents task v i The input data size, c i Represents the computing task v i The CPU resources required for each bit of data; i =d i ×c i Represents the computing task v i The CPU cycles required for virtual task l s =l e =0; Consider a three-dimensional Cartesian coordinate system model, user u n The position is defined as w n =(x n ,y n ,0), and UAV m The position at time slot t is q m [t]=(x m [t],y m [t],H); the coordinates of the drone in the next time slot can be expressed as: q m [t+1]=(x m [t]+t fly v m [t]cosθ m [t],y m [t]+t fly v m [t]sinθ m [t]) (1) in, It's a UAV m The flight speed at time slot t, θ m [t]∈[0,2π] is the flight angle, t fly is the flight time of each time slot; in each time slot, the UAV flies t fly seconds, and it hovers for the remaining time to provide MEC services to ground users. In each time slot, the maximum moving distance of the UAV can be expressed as To ensure safety, the minimum distance between drones cannot be less than D min ; The starting position of each drone is known and is recorded as 3. The method for jointly optimizing UAV trajectories and offloading dependent tasks according to claim 1, wherein: The calculation of global performance benchmark indicators: (1) Communication model; (2) Completion time model; (3) Energy consumption model.

4. The method for jointly optimizing UAV trajectories and offloading dependent tasks as claimed in claim 3, wherein: The communication model: The performance of task offloading depends on the quality of the communication links between the user, the UAV, and the satellite, so it is necessary to analyze the communication links between these three. Between the user and the drone Assume that the communication link between the UAV and the ground user is always within the line of sight, and in each time slot, the channel state between the UAV and the ground user remains constant; user u n and UAVs m The distance between According to Shannon's formula, the transmission rate from user to drone and the transmission rate from drone to user are expressed as: Among them, B e represents the communication bandwidth between the user and the UAV, p UE and p UAV denote the transmit power of the user and the UAV respectively, ρ0 is the channel gain when the transmission distance is 1 meter, and N0 denotes the Gaussian noise power under the channel; Between drones UAVs use line-of-sight links to communicate with each other, and do not consider non-line-of-sight communication; light UAVs m The distance between the enhanced UAV u0 is denoted as d m,0 [t]=||q m [t]-q0[t]||, then the data transmission rate between the light UAV and the enhanced UAV is: Among them, B u represents the communication bandwidth between UAVs; Between drones and satellites The UAV communicates with LEO via a line-of-sight link, so the transmission rate between the UAV and the satellite can be expressed as: in, and Denote the transmission rate from UAV to satellite and the transmission rate from satellite to UAV, respectively. s represents the communication bandwidth between the UAV and the LEO satellite, p LEO represents the transmission power of the LEO satellite, ρ up and ρ down They represent the channel gain between the UAV and the LEO satellite, and N represents the Gaussian white noise power in the channel.

5. The method for jointly optimizing UAV trajectories and offloading dependent tasks as claimed in claim 3, characterized in that: The completion time model: In each time slot, ground users can choose different offloading decisions according to specific needs, offloading tasks with dependencies to other nodes for processing, or completing them all locally; if the task v i The calculation and processing are performed at the ground user terminal. Since there is no need to transmit data, the task v i The computational latency of local processing can be expressed as: Among them, f local Indicates the CPU frequency of the ground user equipment; if the user sets the task v i To offload the task to the UAV for processing, it needs to go through two stages: task transmission and task calculation. In the task transmission stage, the user will i The transmission delay caused by offloading to the UAV can be expressed as: In the task calculation phase, the user sets the task v i The computational delay of the UAV can be expressed as: Among them, f m UAV m CPU frequency; if m≠0, it is a light drone, it can choose to set the task v i Further offload to the enhanced UAV or LEO satellite processing; if you choose to offload the calculation on the enhanced UAV, the task v i The resulting transmission delay can be expressed as: If you choose to set the task v i If the task is offloaded to the LEO satellite for processing, the transmission delay and computation delay of the task are expressed as: Among them, f s Indicates the CPU frequency of the LEO satellite; Unlike independent tasks, the execution of dependent tasks depends on the computation results of their predecessor tasks. If dependent tasks are offloaded to the same processing node, no intermediate result data needs to be transmitted. However, if they are offloaded to different processing nodes, intermediate result data needs to be transmitted. Therefore, the transmission latency of intermediate result data can be expressed as: where R k′,k represents the transmission rate from processing node k′ to k, task v i There may be multiple predecessor tasks, so the time it takes to receive the execution results of all predecessor tasks can be expressed as: where pred(v i ) represents task v i The set of all predecessor tasks, Represents task v j Computational latency at the processing node; The earliest starting time of a task depends not only on the completion time of its predecessor task, but also on the delay of task transmission to the processing node; i The earliest start time can be expressed as: In summary, ground user u n The completion time can be expressed as: The total delay of all users' tasks is the maximum delay of all users completing all tasks, that is, 6. The method for jointly optimizing UAV trajectories and offloading dependent tasks as claimed in claim 3, wherein: The energy consumption model: When a task is processed on a local device, no data transmission is required, and the transmission energy consumption is 0. Therefore, the energy consumption is mainly generated by the energy consumption generated by the user device during the calculation process, that is, the calculation energy consumption: Where κ is the effective switching capacitance, which depends on the hardware chip; When the task is offloaded to the edge server for processing, there is no computing process on the local user device. Therefore, the energy consumption is mainly the energy consumption generated by the user device sending to the drone, which is defined as: Therefore, user u n The total energy consumption can be expressed as: where α i =0 means the task is processed locally, α i =1 means that the task is processed by the MEC server; and the total energy consumption of completing all tasks is the sum of the energy consumption of all user devices to complete the tasks, that is, 7. A system for jointly optimizing UAV trajectories and offloading dependent tasks, which implements the method for jointly optimizing UAV trajectories and offloading dependent tasks as described in any one of claims 1 to 6, characterized in that: The system for jointly optimizing UAV trajectories and offloading dependent tasks includes: The calculation module is used to randomly generate initial drone flight trajectories and task offloading strategies, and based on these, calculate global performance benchmark indicators. These indicators include drone coverage, task execution latency, and device energy consumption. These indicators are used to evaluate the initial effectiveness of the current strategy and provide a reference benchmark for subsequent optimization. The flight trajectory optimization module is used to optimize the UAV flight trajectory using a multi-objective particle swarm optimization algorithm under the premise of a fixed task offloading strategy. The goal is to maximize user coverage and reduce network latency while avoiding overlapping coverage areas. The offloading strategy optimization module is used to optimize the offloading strategy of dependent tasks based on the fixed UAV flight trajectory using the proximal strategy optimization algorithm; The comparison module is used to compare the change in the current global performance indicator with that of the previous round. If the change is less than the preset convergence threshold, the iteration is stopped and the optimal result is output. Otherwise, the algorithm returns to the "trajectory optimization stage" and continues the optimization cycle.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method for jointly optimizing drone trajectories and dependent task offloading as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for jointly optimizing drone trajectories and dependent task offloading as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the system for jointly optimizing drone trajectories and dependent task offloading as described in claim 7.

Images