Joint trajectory, offloading and resource allocation optimization method for multi-UAV assisted MEC system

Abstract

The present invention discloses a joint trajectory, offloading and resource allocation optimization method in a multi-UAV assisted MEC system, comprising the following steps: 1. The MEC server on the UAV collects information such as computing resources, communication resources and task load of all ground user nodes in the flight area, as well as other UAV computing resource information; 2. Introducing relevant auxiliary variables to establish the corresponding optimization problem; 3. The matching relationship between the user and the UAV is initialized by the offloading matching game algorithm, and the corresponding matrix is ​​the offloading matching matrix; 4. Introducing auxiliary variables, the optimization problem is decomposed into several sub-problems for solution. The method is based on the block coordinate descent method and the continuous convex approximation technology, and is called a joint multi-UAV trajectory, offloading and resource optimization scheme; 5. Iterate until the absolute value of the difference between the two objective function values ​​is less than the threshold. This method reduces the average service delay of users.

Description

Technical Field

[0001] The present invention belongs to the field of mobile edge computing, and in particular to a joint trajectory, offloading and resource allocation optimization method in a multi-UAV assisted MEC system. Background Art

[0002] At present, with the increasing popularity of IoT devices and smart mobile terminals, many novel applications have appeared in our lives, such as augmented reality (AR), smart home, etc. However, the computing power of IoT devices is usually limited, and these devices find it difficult to process tasks and respond in a timely manner. Mobile Edge Computing (MEC) provides a possibility to bring computing resources closer to users. In some application scenarios, IoT devices are too far away from the edge cloud, and unmanned aerial vehicles (UAVs) are often used to assist MEC. UAVs are highly flexible, low-cost and easy to deploy, so they are widely used in the civilian field. UAVs will carry MEC servers to provide computing offload services for IoT devices on the ground. On the one hand, drone-assisted MEC reduces the installation cost of infrastructure, and on the other hand, it can provide computing resources for ground users as needed. In addition, cooperation between multiple drones can improve resource utilization and reduce the computing pressure of ground users. However, in recent years, research on drone-assisted MEC has mainly focused on drone path planning and resource allocation, and few studies have jointly optimized drone trajectories, offloading and resource allocation. Currently, most of the research focuses on the resource allocation problem of single UAV-assisted MEC systems, and there is little work on joint track, offloading and resource optimization of multi-UAV-assisted MEC systems. Summary of the invention

[0003] The purpose of the present invention is to provide a joint trajectory, offloading and resource allocation optimization method in a UAV-assisted MEC system, which solves the technical problem of high average user service delay by jointly optimizing the UAV trajectory in a limited period, the offloading ratio of ground user tasks on the UAV, the UAV computing and communication resources, and the UAV-user matching variables.

[0004] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0005] A joint trajectory, offloading and resource optimization method in a multi-UAV assisted MEC system comprises the following steps:

[0006] Step 1: Build a multi-UAV-assisted MEC system, which includes multiple UAVs equipped with MEC servers and multiple ground user nodes in the flight area of ​​the multiple UAVs; the MEC server on the UAV collects the computing resources, bandwidth resources and task load information of all ground user nodes in the flight area, as well as the computing resource information of the UAV;

[0007] Step 2: Establish the optimization problem of joint UAV trajectory, user offloading ratio on UAV, computing resources and bandwidth resources in order to minimize the objective function user average service delay;

[0008] Step 3: The matching relationship between the user and the drone is initialized through the offloading matching game algorithm, and the corresponding matrix is ​​the offloading matching matrix;

[0009] Step 4: Introduce auxiliary variables and decompose the optimization problem in step 2 into several sub-problems for solution; this method is based on the block coordinate descent BCD method and the continuous convex approximation SCA technology, and is called the joint multi-UAV trajectory, offloading and resource optimization JMORO scheme;

[0010] Step 5, iterate step 4. If the absolute value of the difference between the two objective function values ​​exceeds the threshold, return to step 4. If the absolute value of the difference between the two objective function values ​​is not greater than the threshold, the process ends and the optimization results of the drone and user matching variables, drone trajectory, communication resources, computing resources and offloading ratio are obtained.

[0011] Furthermore, in step 1, the specific scenario of the multi-UAV-assisted MEC system is described as follows:

[0012] N UAVs fly at the same altitude H, providing computing offloading services to K users on the ground; represents the set of UAVs, Represents the set of ground user nodes; Used to represent the nth UAV, Used to represent the kth user node; the total time is divided into M time slots, each time slot length δ t , represents a collection of time slots, It is used to represent the mth time slot. When the time slot is small, the position of the UAV in the time slot is considered fixed. Set W k represents the two-dimensional coordinates of the kth user node, and the two-dimensional projection coordinates of UAVn in time slot m are Q n [m]=(x n [m],y n [m]), where x n [m] is the horizontal axis coordinate of the ground projection of the nth UAV in time slot m, y n[m] is the ordinate coordinate of the ground projection of the nth UAV at time slot m;

[0013] There are mobility constraints on drones: first, drones need to fly back to their original positions in the last time slot; second, the distance between drones must be kept at a minimum safe distance d min Because the speed of the drone is limited, the flight distance between two time slots is limited and cannot exceed the maximum flight distance d max ;

[0014] A ground user will match a drone to offload tasks, and a drone will calculate the tasks of multiple ground users; the drone-user matching variable is defined as A k，n [m], when drone n provides service to user k in time slot m, A k，n [m]=1, otherwise A k,n [m] = 0;

[0015] In the mth time slot, set the computing task D of ground user k in this time slot k [m], the task is based on θ k,n [m] ratio is unloaded to UAV n for calculation, while D k The remaining 1-θ in [m] k,n [m] scale tasks are calculated locally;

[0016] The communication link between the UAV and the ground user is set to line-of-sight transmission, that is, the channel gain is expressed as:

[0017]

[0018] Where β represents the channel power gain at a reference distance of 1m, d k,n represents the distance between UAV n and ground user k;

[0019] According to Shannon's formula, the transmission rate of user k to drone n in time slot m is:

[0020]

[0021] Where P is the user's upload power, which is set constant; B k.n [m] is the bandwidth allocated to user k by drone n in time slot m, σ 2 is the additive white Gaussian noise power;

[0022] Indicates that in the mth time slot, when user k and drone n match, the ground user will θ k,n [m] ratio of the task load to be offloaded to the transmission time of UAV n;

[0023]

[0024] Considering the bandwidth situation on the drone, a drone offloads computing services to multiple ground users, so the bandwidth resources of the drone must be allocated; the bandwidth resources occupied by ground users cannot exceed the upper limit of the bandwidth resources of the drone; define B n is the bandwidth resource of the nth drone;

[0025] Considering the local computing situation, Indicates the local execution time of ground user k when user k and drone n match in the mth time slot:

[0026]

[0027] Among them, f k represents the computing power of user k, C k represents the number of CPU cycles required for user k to execute each bit of task;

[0028] The computing situation on the drone is the same as the bandwidth situation on the drone, and should not exceed the resource limit; define f n is the computing resource of the nth drone, f k,n [m] is the computing resources allocated by the nth drone to the kth user in the mth time slot;

[0029] use Indicates that in the mth time slot, when user k and drone n match, the ground user will θ k,n After the task volume of proportion [m] is offloaded to UAV n, the computing time required for UAV n to process this part of data is:

[0030]

[0031] Furthermore, the optimization problem of the joint UAV trajectory, the user offloading ratio on the UAV, the computing resources and the bandwidth resources established in step S2 specifically includes the following steps:

[0032] Define the average service delay for users

[0033]

[0034] Let A k,n [m],θ k,n [m],f k,n [m],B k,n [m],Q n The matrices corresponding to [m] are A, Θ, F, B, Q, and the average service delay is T(A, Θ, F, B, Q), and we have:

[0035]

[0036] Taking minimizing the average service delay as the optimization goal, while considering the matching of drones and users, the user offloading ratio, drone flight coordinates, computing resource allocation, bandwidth resource allocation, and satisfying the constraints of offloading matching, drone mobility, and limited resources, the following optimization problem is obtained:

[0037]

[0038]

[0039] Among them, A1 and A2 are offloading matching constraints, A3 represents the offloading ratio constraint, A4, A5 and A6 represent UAV mobility constraints, A7 and A8 represent bandwidth resource constraints, and A9 and A10 represent computing resource constraints.

[0040] Furthermore, in step 3, the unloading matching game algorithm is used, and the solution process is as follows:

[0041] Step 3.1, initialize all users to unmatched state;

[0042] Step 3.2, the drone and the user obtain a preference list according to their respective preferences, and match them based on the matching scheme of the Gale-Shapley algorithm in the matching game;

[0043] Step 3.3: Output the offloading matching matrix A * .

[0044] Furthermore, in step 4, a solution based on the BCD method is used, and the process is as follows:

[0045] Step 4.1: Obtain A from step 3 * , and initialize to obtain θ 0 ,F 0 ,B 0 ,Q 0 ,T 0 ; where θ 0 , is the initialization matrix of the unloading ratio of ground users on the UAV, F 0 is the initialization matrix of the computing resources allocated to the user of the drone, B 0 The initialization matrix Q is used to allocate bandwidth resources to users of drones. 0 Initialize the trajectory matrix of the drone, T 0 Initialize the matrix for the introduced auxiliary variable, which is the inverse of the transmission rate;

[0046] Step 4.2: According to F 0 ,B 0 ,Q 0 ,T0 Optimize the unloading ratio optimization sub-problem and obtain the optimized unloading ratio matrix θ in this round of iteration * ;

[0047] Step 4.3: According to Θ * ,B 0 ,Θ * ,T 0 Optimize the computing resource optimization sub-problem and obtain the optimized computing resource matrix F in this round of iteration * ;

[0048] Step 4.4: According to Θ * ,F * ,Q 0 ,T 0 Optimize the bandwidth resource optimization sub-problem and obtain the optimized bandwidth resource matrix B in this round of iteration * ;

[0049] Step 4.5: According to Θ * ,F * ,B * Optimize the drone trajectory optimization sub-problem and obtain the optimized drone trajectory matrix Q in this round of iteration * ;

[0050] Iterate until the difference between the two values ​​of the average service delay of the user before and after is less than the threshold of 10 -4 , obtain the suboptimal solution θ of the offloading ratio, the suboptimal solution F of the computing resources, the suboptimal solution B of the bandwidth resources and the suboptimal solution Q of the UAV trajectory;

[0051] In step 5, an auxiliary variable x is introduced. k,n [m] and auxiliary variable λ k',n [m], by relaxing the constraints, the non-convex problem is transformed into a convex problem and the suboptimal solution is iteratively obtained.

[0052] The joint trajectory, unloading and resource allocation optimization scheme in a multi-UAV assisted MEC system of the present invention has the following advantages:

[0053] The present invention is aimed at the partial unloading scenario assisted by multiple drones. Based on convex optimization theory, the original optimization problem is transformed by introducing auxiliary variables with the goal of minimizing the average service delay of users. Then the matching game algorithm is used to solve the drone-user matching matrix. The problem is then divided into multiple sub-problems, which are solved by alternate optimization using the BCD method. For the part of solving the drone trajectory in the sub-problem, the present invention uses SCA technology for approximate solution. Compared with traditional algorithms, this algorithm not only takes into account the trajectory constraints of multiple drones, but also takes into account the resource allocation in the partial unloading scenario, taking into account both computational complexity and system delay.

[0054] The present invention solves the drone-user matching matrix through a matching game algorithm, so that the matching relationship between drones and users can be clearly obtained. In addition, the BCD method and SCA technology effectively deal with the non-convexity of the original problem. Ultimately, multiple drones and local computing resources can better handle the task load, thereby reducing the average service delay of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the joint trajectory, offloading and resource allocation optimization scenario in the multi-UAV assisted MEC system proposed in the present invention.

[0056] Figure 2 It is a flow chart of the joint trajectory, offloading and resource allocation optimization algorithm in the multi-UAV assisted MEC system proposed in the present invention. DETAILED DESCRIPTION

[0057] In order to better understand the purpose, structure and function of the present invention, the following is a further detailed description of the joint trajectory, unloading and resource allocation optimization scheme in a multi-UAV assisted MEC system of the present invention in conjunction with the accompanying drawings.

[0058] See also Figure 1 This embodiment provides a joint trajectory, offloading and resource allocation optimization method in a multi-UAV assisted MEC system. The method improves the common single UAV scenario and the traditional block coordinate descent method in the current UAV assisted MEC system, and proposes a JMORO method based on matching game algorithm, BCD method and SCA technology in a multi-UAV assisted MEC system to achieve a high-reliability, low-latency UAV task scheduling solution. This method enables multiple UAVs to handle the task load of ground user nodes while ensuring that the upper limit of local nodes and UAV resources is not exceeded. This problem can be modeled as a non-convex optimization problem, which can be approximated as a convex optimization problem by introducing auxiliary variables. Specifically, the specific process of this method is as follows: Figure 2 As shown, the method specifically comprises the following steps:

[0059] Step S1: construct a multi-UAV-assisted MEC system, which includes multiple UAVs equipped with MEC servers and multiple user nodes in the flight area of ​​the multiple UAVs; the MEC server on the UAV collects computing resources, bandwidth resources and task load information of all ground user nodes in the flight area, as well as computing resource information of the UAV;

[0060] Specifically, in this embodiment, the following definition is made for the multi-drone assisted MEC system, including:

[0061] N UAVs fly at the same altitude H and provide computing offloading services to K users on the ground. represents the set of UAVs, Represents the set of ground user nodes. Used to represent the nth UAV, is used to represent the kth user node. For ease of analysis, the total time is divided into M time slots, each time slot length δ t , represents a collection of time slots, is used to represent the mth time slot. When the time slot is small enough, the position of the UAV in the time slot is considered fixed. Assume that W k represents the two-dimensional coordinates of the kth user node, and the two-dimensional projection coordinates of UAVn in time slot m are Q n [m]=(x n [m],y n [m]).

[0062] There are some mobility constraints on drones. First, the drones need to fly back to their original positions in the last time slot; second, the distance between drones needs to be kept within d min Because the speed of the drone is limited, the flight distance between two time slots is limited and cannot exceed d max .

[0063] A ground user will match a UAV to offload tasks, and a UAV can calculate the tasks of multiple ground users. Define the UAV-user matching variable as A k，n [m], when UAVn provides service to user k in time slot m, A k，n [m]=1, otherwise A k,n [m]=0.

[0064] A ground user will match a UAV to offload tasks, and a UAV can calculate the tasks of multiple ground users. Define the UAV-user matching variable as A k，n [m], when UAVn provides service to user k in time slot m, A k，n [m]=1, otherwise A k,n [m]=0.

[0065] The user does not offload all tasks to the UAV for computation, but only offloads part of them. In the mth time slot, suppose the computational task amount D of user k in this time slot is k [m], the task is based on θ k,n [m] ratio is unloaded to UAVn for calculation, while D k The remaining 1-θ in [m] k,n [m] scale tasks are computed locally.

[0066] Assuming that the communication link between the UAV and the ground user is line-of-sight transmission, the channel gain can be expressed as:

[0067]

[0068] Where β represents the channel power gain at a reference distance of 1m, d k,n represents the distance between UAV n and ground user k;

[0069] According to Shannon's formula, the transmission rate of user k to UAVn in time slot m is:

[0070]

[0071] Where P is the user's upload power, which is set constant; B k.n [m] is the bandwidth allocated to user k by drone n in time slot m, σ 2 is the additive white Gaussian noise power.

[0072] Indicates that in the mth time slot, when user k and UAVn match, the ground user will θ k,n [m] ratio of the task load to be offloaded to the transmission time of UAVn.

[0073]

[0074] Consider the bandwidth situation on the UAV. A UAV offloads computing services to multiple ground users, so the bandwidth resources of the UAV must be allocated. The bandwidth resources occupied by ground users cannot exceed the upper limit of the bandwidth resources of the UAV. Definition B n is the bandwidth resource of the nth UAV.

[0075] Consider the local computing situation. It represents the local execution time of ground user k when user k and UAVn match in the mth time slot.

[0076]

[0077] Among them, f k represents the computing power of user k, C k It represents the number of CPU cycles required for user k to execute each bit of task.

[0078] The computation on the UAV is the same as the bandwidth on the UAV and should not exceed the resource limit. Define f n is the computing resource of the nth UAV, f k，n [m] is the computing resource allocated to the kth user by the nth UAV in the mth time slot.

[0079] use Indicates that in the mth time slot, when user k and UAVn match, the ground user will θ k,n After the task volume [m] is offloaded to UAVn, the computing time required for UAVn to process this part of data.

[0080]

[0081] Step S2, establishing a corresponding optimization problem;

[0082] Define the average service delay for users

[0083]

[0084] Let A k,n [m],θ k,n [m],f k,n [m],B k,n [m],Q n The matrices corresponding to [m] are A, Θ, F, B, Q, and the average service delay is T(A, Θ, F, B, Q), and we have:

[0085]

[0086] Taking minimizing the average service delay as the optimization goal, taking into account the drone-user matching, user offloading ratio, drone flight coordinates, computing resource allocation, bandwidth resource allocation, and satisfying the constraints of offloading matching, drone mobility, and limited resources, the following optimization problem can be obtained:

[0087]

[0088] Among them, A1 and A2 are offloading matching constraints, A3 represents offloading ratio constraint, A4, A5 and A6 represent UAV mobility constraints, A7 and A8 represent bandwidth resource constraints, and A9 and A10 represent computing resource constraints.

[0089] Step S3, initializing the offloading matching matrix through an offloading matching game algorithm;

[0090] Problem P0 is a non-convex problem with strong coupling between optimization parameters, so it is difficult to solve.

[0091] Therefore, we can first use the offloading matching game algorithm to obtain a stable matching relationship between the UAV and the ground user, and then obtain the initial allocation results of offloading matching, offloading ratio, computing resources, bandwidth resources and UAV position, which can be expressed as * ,Θ 0 ,F 0 ,B 0 ,Q​0 >. The UAV trajectory is initialized as a circular trajectory. The initial value of the offloading ratio of users to UAV is 1. For the initial allocation of computing resources, UAV only allocates resources to users matching the UAV. If multiple users' tasks are offloaded to the same UAV, the UAV allocates computing resources to these users in equal parts. The same applies to bandwidth resources.

[0092] After obtaining the uninstall matching result A * Finally, the BCD method is used to optimize the offloading ratio, computing resources, bandwidth resources and UAV trajectory respectively, and finally the suboptimal solution of the original problem P0 can be obtained.

[0093] The unloading matching game algorithm is improved based on the Gale-Shapley algorithm. The specific process is described as follows:

[0094]

[0095]

[0096] Step S4, introducing auxiliary variables and decomposing the optimization problem into several sub-problems for solution;

[0097] In order to obtain the matching result A * Under the condition of , problem P0 can be transformed into the following sub-problems:

[0098]

[0099] Using the BCD method, we can continue to decompose problem P1 into multiple sub-problems. In Relative UAV trajectory Q n [m] is non-convex, so the constraints need to be relaxed. At the same time use To replace In We can get T2(Θ,Q,F,B), where A k,n [m] is a fixed value after optimization by the uninstall matching algorithm:

[0100]

[0101] make The corresponding matrix is ​​T. Therefore, P1 can be transformed into:

[0102]

[0103] make The initial value is Therefore, Q 0 and B0 Get T 0 . Q 0 , F 0 , B 0 and T 0 Substituting into P2, we get subproblem P3, which is only related to Θ Related:

[0104]

[0105] stA3 (12)

[0106] It can be seen that when UAV trajectory, computing resources, communication resources and auxiliary variables When fixed, P3 is convex, and by solving this convex problem with CVX, we can obtain the optimal solution to subproblem P3.

[0107] After the above processing, the fixed unloading ratio, UAV trajectory, bandwidth resources and auxiliary variables Transform problem P2 into problem P4:

[0108] P4:

[0109]

[0110] stA9,A10(13)

[0111] It can be seen that when the above variables are fixed, P4 is convex. By solving this convex problem with CVX, the optimal solution of subproblem P4 can be obtained.

[0112] Next, fix the offloading ratio, UAV trajectory, computing resources and auxiliary variables Transform problem P2 into problem P5:

[0113]

[0114] It can be seen that when the above variables are fixed, P5 is convex. By solving this convex problem with CVX, the optimal solution of subproblem P5 can be obtained.

[0115] After decomposing the original problem P1 multiple times, the problem is greatly simplified. Fix the offloading ratio, computing resources and bandwidth resources, and transform the problem P2 into problem P6:

[0116]

[0117] According to formulas (6) and (7), we can get:

[0118]

[0119] Introduce two new auxiliary variables χ k,n[m] and λ k',n [m], using continuous convex optimization techniques to solve the problem. The two auxiliary variables satisfy:

[0120]

[0121] χ k,n [m]≥0 (18)

[0122]

[0123] because Relative to χ k,n [m] is convex, ||Q n [m]-W k ||Relative Q n [m] is also convex, and for any given local point and have:

[0124]

[0125]

[0126] Among them, the local point and is the solution obtained in the l-1th iteration.

[0127] The right sides of the two inequalities are about χ k,n [m] and Q n [m] is an affine function, so we can get the following inequality:

[0128]

[0129]

[0130] Also define variables

[0131]

[0132] This constraint is convex.

[0133] In addition, constraint A5 is also non-convex and needs to be handled. and Applying the first-order Taylor expansion at , we can obtain the convex constraint:

[0134]

[0135] Therefore, problem P6 can be transformed into a convex problem P7, where A,Λ is χ k,n [m] and λ k',n[m] corresponds to the matrix:

[0136]

[0137]

[0138] P7 is a convex optimization problem that can be solved using CVX.

[0139] The JMORO method in step 4 is summarized as follows:

[0140]

[0141] Step S5, iterate step S4. If the absolute value of the difference between the two objective function values ​​exceeds the threshold, return to step S4. If it is less than the threshold, the process ends and the optimization results of drone-user matching variables, drone tracks, communication resources, computing resources and offloading ratios are obtained.

[0142] The results of each iteration are obtained according to the JMORO method, and the iteration is performed until the difference between the average service delays of the two users is less than 10 -4 .

[0143] The above technical solution relates to a joint trajectory, offloading and resource allocation optimization method in a multi-UAV assisted MEC system. The computing tasks of ground users in the system are divided into two parts, one part is offloaded to the relevant UAV for calculation, and the rest is calculated locally. Each UAV flies at the same altitude above the ground user to provide computing services for the ground user. This method jointly optimizes the offloading task ratio, UAV trajectory, bandwidth resources, computing resources and UAV-user matching variables to minimize the average service delay of users.

[0144] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A joint trajectory, offloading and resource optimization method in a multi-UAV assisted MEC system, characterized in that: The steps include: Step 1: Build a multi-UAV-assisted MEC system, which includes multiple UAVs equipped with MEC servers and multiple ground user nodes in the flight area of ​​the multiple UAVs; the MEC server on the UAV collects the computing resources, bandwidth resources and task load information of all ground user nodes in the flight area, as well as the computing resource information of the UAV; Step 2: Establish the optimization problem of joint UAV trajectory, user offloading ratio on UAV, computing resources and bandwidth resources in order to minimize the objective function user average service delay; Step 3: The matching relationship between the user and the drone is initialized through the offloading matching game algorithm, and the corresponding matrix is ​​the offloading matching matrix; Step 4: Introduce auxiliary variables and decompose the optimization problem in step 2 into several sub-problems for solution; this method is based on the block coordinate descent BCD method and the continuous convex approximation SCA technology, and is called the joint multi-UAV trajectory, offloading and resource optimization JMORO scheme; Step 5, iterate step 4. If the absolute value of the difference between the two objective function values ​​exceeds the threshold, return to step 4. If the absolute value of the difference between the two objective function values ​​is not greater than the threshold, the process ends and the optimization results of the drone and user matching variables, drone track, communication resources, computing resources and offloading ratio are obtained. In step 1, the specific scenario of the multi-drone assisted MEC system is described as follows: N UAVs fly at the same altitude H, providing computing offloading services to K users on the ground; represents the set of UAVs, Represents the set of ground user nodes; Used to represent the nth UAV, Used to represent the kth user node; the total time is divided into M time slots, each time slot length δ t , represents a collection of time slots, It is used to represent the mth time slot. When the time slot is small, the position of the UAV in the time slot is considered fixed. Set W k represents the two-dimensional coordinates of the kth user node, and the two-dimensional projection coordinates of UAVn in time slot m are Q n [m]=(x n [m],y n [m]), where x n [m] is the horizontal axis coordinate of the ground projection of the nth UAV in time slot m, y n [m] is the ordinate coordinate of the ground projection of the nth UAV at time slot m; There are mobility constraints on drones: first, drones need to fly back to their original positions in the last time slot; second, the distance between drones must be kept at a minimum safe distance d min Because the speed of the drone is limited, the flight distance between two time slots is limited and cannot exceed the maximum flight distance d max ; A ground user will match a drone to offload tasks, and a drone will calculate the tasks of multiple ground users; the drone-user matching variable is defined as A k，n [m], when drone n provides service to user k in time slot m, A k，n [m]=1, otherwise A k,n [m] = 0; In the mth time slot, set the computing task D of ground user k in this time slot k [m], the task is based on θ k,n [m] ratio is unloaded to UAV n for calculation, while D k The remaining 1-θ in [m] k,n [m] scale tasks are calculated locally; The communication link between the UAV and the ground user is set to line-of-sight transmission, that is, the channel gain is expressed as: Where β represents the channel power gain at a reference distance of 1m, d k,n represents the distance between UAV n and ground user k; According to Shannon's formula, the transmission rate of user k to drone n in time slot m is: Where P is the user's upload power, which is set constant; B k.n [m] is the bandwidth allocated to user k by drone n in time slot m, σ 2 is the additive white Gaussian noise power; Indicates that in the mth time slot, when user k and drone n match, the ground user will θ k,n [m] ratio of the task load to be offloaded to the transmission time of UAV n; Considering the bandwidth situation on the drone, a drone offloads computing services to multiple ground users, so the bandwidth resources of the drone must be allocated; the bandwidth resources occupied by ground users cannot exceed the upper limit of the bandwidth resources of the drone; define B n is the bandwidth resource of the nth drone; Considering the local computing situation, Indicates the local execution time of ground user k when user k and drone n match in the mth time slot: Among them, f k represents the computing power of user k, C k represents the number of CPU cycles required for user k to execute each bit of task; The computing situation on the drone is the same as the bandwidth situation on the drone, and should not exceed the resource limit; define f n is the computing resource of the nth drone, f k,n [m] is the computing resources allocated by the nth drone to the kth user in the mth time slot; use Indicates that in the mth time slot, when user k and drone n match, the ground user will θ k,n After the task volume of proportion [m] is offloaded to UAV n, the computing time required for UAV n to process this part of data is: The optimization problem of the joint UAV trajectory, the user offloading ratio on the UAV, the computing resources and the bandwidth resources established in step S2 specifically includes the following steps: Define the average service delay for users Let A k,n [m],θ k,n [m],f k,n [m],B k,n [m],Q n The matrices corresponding to [m] are A, Θ, F, B, Q, and the average service delay is T(A, Θ, F, B, Q), and we have: Taking minimizing the average service delay as the optimization goal, while considering the matching of drones and users, the user offloading ratio, drone flight coordinates, computing resource allocation, bandwidth resource allocation, and satisfying the constraints of offloading matching, drone mobility, and limited resources, the following optimization problem is obtained: Among them, A1 and A2 are offloading matching constraints, A3 represents offloading ratio constraints, A4, A5 and A6 represent UAV mobility constraints, A7 and A8 represent bandwidth resource constraints, and A9 and A10 represent computing resource constraints; In step 3, the unloading matching game algorithm is used, and the solution process is as follows: Step 3.1, initialize all users to unmatched state; Step 3.2, the drone and the user obtain a preference list according to their respective preferences, and match them based on the matching scheme of the Gale-Shapley algorithm in the matching game; Step 3.3: Output the offloading matching matrix A * ; In step 4, a solution based on the BCD method is used, and the process is as follows: Step 4.1: Obtain A from step 3 * , and initialize to obtain θ 0 ,F 0 ,B 0 ,Q 0 ,T 0 ; where θ 0 , is the initialization matrix of the unloading ratio of ground users on the UAV, F 0 is the initialization matrix of the computing resources allocated to the user of the drone, B 0 The initialization matrix Q is used to allocate bandwidth resources to users of drones. 0 Initialize the trajectory matrix of the drone, T 0 Initialize the matrix for the introduced auxiliary variable, which is the inverse of the transmission rate; Step 4.2: According to F 0 ,B 0 ,Q 0 ,T 0 Optimize the unloading ratio optimization sub-problem and obtain the optimized unloading ratio matrix θ in this round of iteration * ; Step 4.3: According to Θ * ,B 0 ,Θ * ,T 0 Optimize the computing resource optimization sub-problem and obtain the optimized computing resource matrix F in this round of iteration * ; Step 4.4: According to Θ * ,F * ,Q 0 ,T 0 Optimize the bandwidth resource optimization sub-problem and obtain the optimized bandwidth resource matrix B in this round of iteration * ; Step 4.5: According to Θ * ,F * ,B * Optimize the drone trajectory optimization sub-problem and obtain the optimized drone trajectory matrix Q in this round of iteration * ; Iterate until the difference between the two values ​​of the average service delay of the user before and after is less than the threshold of 10 -4 , obtain the suboptimal solution θ of the offloading ratio, the suboptimal solution F of the computing resources, the suboptimal solution B of the bandwidth resources and the suboptimal solution Q of the UAV trajectory; In step 5, an auxiliary variable x is introduced. k,n [m] and auxiliary variable λ k',n [m], by relaxing the constraints, the non-convex problem is transformed into a convex problem and the suboptimal solution is iteratively obtained.

Images