Method for maximizing task completion rate of multiple MEC servers with UAV assistance

Abstract

This invention discloses a method for maximizing the task completion rate of a UAV-assisted multi-MEC server, primarily addressing the problems of severe channel fading and limited computing resources of a single MEC server in existing technologies. The method includes: 1) constructing a UAV-assisted multi-MEC server network system model; 2) employing a time-division multiple access (TDMA) protocol, where users offload part of their task input data to the UAV, and the UAV offloads part of its unfinished task input data to multiple ground base stations for computation; 3) finding the optimal task completion rate, CPU computing frequency, transmission power, offloading decision, and UAV trajectory by maximizing user task completion rate and minimizing system energy consumption; and 4) optimizing system operating parameters based on the optimization results. This invention effectively improves network computing resources, achieves stability in environments with high computational workloads, and reduces energy consumption for both UAVs and users, making it applicable to edge computing systems.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and further relates to edge computing technology. Specifically, it is a method for maximizing the task completion rate of a drone-assisted multi-mobile edge computing MEC server, which can be used in edge computing systems to improve the task completion rate by allowing users to unload tasks in scenarios with drone-assisted multi-edge servers. Background Technology

[0002] As the number of wireless access devices continues to grow, the vast amounts of data collected from these devices need to be transmitted from one place to another for intelligent decision-making, placing a significant burden on wireless communication infrastructures with limited radio spectrum. Furthermore, numerous wireless access devices are enabling many compelling new applications, such as real-time video analytics, augmented or virtual reality, and artificial intelligence. However, these computationally intensive and latency-sensitive tasks rely on our ability to rapidly process data and extract useful information.

[0003] Mobile edge computing can offload heavy computing tasks from power-constrained users to the edge, processing data closer to the user. This reduces traffic bottlenecks in the core and backhaul networks, enabling rapid data processing and effectively solving network latency issues. Overall, this distributed cloud architecture forms a cornerstone technology of 5G systems, transforming the traditional cloud-based data processing paradigm by providing cloud computing capabilities and a service environment at the network edge.

[0004] However, in traditional mobile edge computing networks, mobile edge computing servers are typically deployed in a fixed manner on the ground. Due to ground obstacles such as blockage and shadows, effective terrestrial wireless communication between the mobile edge computing server and the user is often impossible. Given this deficiency, leveraging the advantages of drones—ease of deployment, flexible mobility, and line-of-sight link connectivity—drone-supported mobile edge computing networks have recently been proposed as a promising solution to improve reliable connectivity for ground users.

[0005] In their paper "Joint offloading and trajectory design for UAV-enabled mobile edge computing systems" (IEEE Internet of Things Journal, 2019: 1879-1892), Hu Q, Cai Y, Yu G, et al. considered deploying UAVs to provide computing services to users. By optimizing the offloading rate, user scheduling, and UAV trajectory, they proposed a problem to minimize the sum of maximum latency among users. In their paper "UAV-aided mobile edge computing systems with one byone access scheme" (IEEE Transactions on Green Communications and Networking, 2019: 664-678), Hua M, Wang Y, Li C, et al. considered using a UAV to help users offload computing tasks. Users can perform computations locally or offload the tasks to the UAV. Unlike the studies mentioned above, Hu X, Wong KK, Yang K et al., in their paper "UAV-assisted relaying and edge computing: scheduling and trajectory optimization" (IEEE transactions on wireless communications, 2019: 4738-4752), utilized UAVs as relay stations to assist users in offloading computing tasks. UAVs can not only act as mobile edge computing servers but also provide communication services to users by forwarding received computing tasks to base stations for remote computation. This improves system stability in environments with high computing workloads while reducing UAV task processing latency. The aforementioned research on UAV-assisted mobile edge computing networks primarily considers scenarios where UAVs act as mobile edge computing servers or where only one base station serves as a mobile edge computing server. However, in practical applications, UAVs have limited energy and computing resources, and multiple base stations may exist on the ground, making these methods difficult to solve real-world problems. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for maximizing the task completion rate of a drone-assisted multi-MEC server. By maximizing the user task completion rate and minimizing system energy consumption, the optimal task completion rate, CPU computing frequency, transmission power, offloading decision, and drone trajectory are identified. This solves the problems of severe channel fading and limited computing resources of a single mobile edge computing server in traditional mobile edge computing networks, effectively improving the stability of edge computing networks in environments with high computational workloads, while simultaneously reducing the energy consumption of both drones and users.

[0007] The basic idea behind this invention is as follows: In the scenario of using a drone to assist multiple mobile edge computing servers, the user uses a time-division multiple access protocol to partially offload the task input data to the drone, and the drone shares computing resources with the user to help the user compute the task; the drone also uses a time-division multiple access protocol to further offload the task input data that it cannot complete itself to multiple base stations on the ground for computation, so as to save its own energy consumption and improve computing efficiency.

[0008] To achieve the above objectives, the technical solution of the present invention includes the following steps:

[0009] (1) Building an edge computing network model:

[0010] Construct an edge computing network model consisting of one drone, K users, and M base stations equipped with mobile edge computing (MEC) servers; and Let k and m represent the sets of users and base stations, respectively, where k and m represent the k-th user and the m-th base station equipped with a mobile edge computing server, respectively.

[0011] (2) Divide the time slot structure for users and drones:

[0012] Discretize the finite task completion time T into N equal time slots, let Let N be a set of time slots; the duration of each time slot is τ = T / N. Assume the UAV's position remains constant during each time slot. The user uses a time division multiple access (TDMA) protocol, dividing each time slot into K' equal sub-time slots, K' = K, with each sub-time slot's duration being δ = T / (NK'). User k unloads its task input data in its corresponding k'-th sub-time slot. The UAV also uses a K' equal-time-division multiple access protocol, further dividing each sub-time slot into M smaller sub-time slots. The size of each smaller sub-time slot is determined by the unloading decision variable α. U,k,m [n] Determined;

[0013] (3) Obtain the optimal optimization variables in the scenario of UAV-assisted multi-mobile edge computing server:

[0014] (3.1) Using a three-dimensional Euclidean coordinate system, assuming that the positions of the base station and all users are fixed on the ground at zero elevation, the horizontal positions of the base station m and user k are obtained as b, respectively. m and w k Assume the drone flies at a fixed altitude H (H > 0) within the mission completion time T; and set the start and end points of the flight as q respectively. I and q F Based on the discrete path planning method, the horizontal position of the UAV in the nth time slot is obtained as q[n] = q[nτ], where q[0] = q I q[N]=q F Assuming that the wireless channels between the UAV and the base station, and between the UAV and the user, are dominated by line-of-sight links, the channel power gain h between the UAV and user k, and between the UAV and the base station m, is obtained within time slot n. U,k [n] and g U,m [n];

[0015] (3.2) Assume that each user has a specific computation task to be executed, and the computation task consists of triples. It means that I k C represents the size of the input data for the computation task. k T represents the computational resources required to input 1 bit of data. k Let T be the maximum tolerable delay for user k; let all users have the same maximum tolerable delay, i.e., T. k =T; Assume each user's task completion rate is μ k ;

[0016] (3.3) Each user uses a time-division multiple access (TDMA) protocol to offload a portion of the computational task to the UAV for computation, while the other portion is computed locally by the user. The UAV collects the offloaded computational tasks by optimizing its flight trajectory, and also uses a TDMA protocol to offload a portion of the collected computational tasks to multiple ground base stations for computation, while the other portion is computed locally by the UAV. The amount of task data executed locally by the k-th user within time slot n is obtained. and energy consumption The amount of task data that the kth user unloads onto the drone and transmission energy consumption The amount of task data that the drone locally calculates and executes for the kth user. and energy consumption The drone unloads the task data of the kth user to base station m. and transmission energy consumption and the flight energy consumption of drones

[0017] (3.4) Obtain the total energy consumption E of user k in each time slot n according to the following formula.k [n] and the total energy consumption E of the drone in each time slot n U [n]:

[0018]

[0019]

[0020] (3.5) Construct the optimal task completion rate μ max and the weighted energy consumption of users and drones E min The expression:

[0021]

[0022] Wherein, the optimization variable is the user uninstallation decision α = {α U,k,m [n]}、User task completion rate μ={μ k}, Local computing CPU frequency F = {f k [n],f U,k [n]}, User and drone transmission power P={p k [n],p U,k,m [n]} and the drone trajectory Q={q[n]};ω U and ω E These represent the energy consumption weight of the drone and the total energy consumption weight, respectively.

[0023] The user and the drone are set to meet the following constraints:

[0024] Indicates task completion constraints; This represents the causal constraint of information, where, in, 0≤μ k ≤1, 0≤f k [n]≤f k,max , 0≤p k [n]≤p k,max , p k [N-1]=p k [N] = 0, 0 ≤ f U,k [n]≤f U,max , f U,k [1]=f U,k [N] = 0, 0 ≤ p U,k,m [n]≤p U,max , p U,k,m [1]=p U,k,m [N] = 0, 0 ≤ α U,k,m[n]≤1, α U,k,m [1]=α U,k,m [N] = 0, q[0] = q I q[N]=q F , where f k,max f U,max p k,max and p U,max These represent the maximum available CPU frequency and maximum transmission power for the k-th user and the drone, respectively. q[n]-q[n-1]||≤τV max , This indicates the maximum speed constraint for the drone;

[0025] (3.6) Maximize the task completion rate and minimize the weighted energy consumption of the drone and the user, and obtain the task completion rate μ through an alternating optimization algorithm. max And the weighted energy consumption E of users and drones min The corresponding optimal optimization variable;

[0026] (4) The optimal task completion rate μ obtained in step (3.6) max And the weighted energy consumption E of users and drones min The corresponding optimal optimization variables are used to set the system operating parameters, and the system is made to run under these parameters to achieve optimization of system task completion rate and energy consumption.

[0027] The present invention has the following advantages compared with the prior art:

[0028] First, because this invention uses drones to establish reliable connections with users and base stations, drones can not only act as mobile edge computing servers, but also provide communication services to users by forwarding received computing tasks to base stations for remote computing, thereby improving the stability of the system in environments with high computing task intensity and reducing the task processing latency of drones.

[0029] Secondly, in real-world scenarios, network systems often face multiple ground base station scenarios, while existing UAV-assisted mobile edge computing systems mostly consider single base station scenarios, making it difficult to solve practical problems. Compared to existing methods, this invention considers multiple ground base station scenarios based on the fact that UAVs can forward computing tasks to base stations for remote computing. Under fixed latency requirements, it can distribute computing tasks to different base stations to complete more tasks and achieve minimal system energy consumption. Attached Figure Description

[0030] Figure 1 This is a schematic diagram illustrating an application scenario of the method of the present invention;

[0031] Figure 2This is a schematic diagram of the time slot structure of the present invention;

[0032] Figure 3 This is a flowchart illustrating the implementation of the method of the present invention;

[0033] Figure 4 This is a UAV trajectory diagram corresponding to different task completion times in the method of the present invention;

[0034] Figure 5 This is a speed graph of the UAV corresponding to different task completion times in the method of the present invention;

[0035] Figure 6 This is a simulation result diagram showing the impact of the workload and the number of base stations on the weighted sum of task completion rate and energy consumption in the method of this invention;

[0036] Figure 7 The following is a simulation result diagram showing the impact of task quantity and number of base stations on task completion rate and energy consumption in the method of the present invention;

[0037] Figure 8 This is a simulation result diagram showing the impact of UAV weight and system bandwidth on the weighted sum of task completion rate and energy consumption in the method of this invention;

[0038] Figure 9 The following is a simulation result diagram showing the impact of UAV weight and system bandwidth on task completion rate and energy consumption in the method of this invention;

[0039] Figure 10 This is a diagram showing the influence of UAV weight and energy consumption weight on the task completion rate and energy consumption weighted sum in the method of this invention;

[0040] Figure 11 The following is a simulation result diagram showing the impact of UAV weight and energy consumption weight on task completion rate and energy consumption in the method of this invention; Detailed Implementation

[0041] The implementation process of the technical solution of the present invention will be described in detail below with reference to the accompanying drawings:

[0042] Example 1: Refer to Figure 3 This invention provides an energy efficiency optimization method in multi-user cognitive edge computing networks, and the specific implementation steps are as follows:

[0043] Step 1: Refer to Figure 1 The edge computing network constructed by this invention consists of one drone, K users, and M base stations equipped with mobile edge computing servers; let and Let k and m represent the sets of users and base stations, respectively, where k represents the k-th user and the m-th base station equipped with a mobile edge computing server.

[0044] Step 2: Divide the time slot structure for users and drones:

[0045] Reference Figure 2 This invention discretizes the finite task completion time T into N equal time slots, letting Let N be the set of time slots. The duration of each time slot is τ = T / N, where τ is small enough that the UAV's position can be assumed to remain constant during each time slot. To avoid interference between users during the unloading process, a time division multiple access (TDMA) protocol is used, and each time slot is further divided into K equal sub-time slots. The duration of each sub-time slot is δ = T / (NK), and user k unloads its task input data in the k-th sub-time slot. To better distinguish unloading signals from different users, the UAV also uses a time division multiple access protocol with K equal time slots. Since the UAV needs to further partially offload each user's task input data to multiple ground base stations equipped with mobile edge computing servers, each sub-time slot is further divided into M smaller sub-time slots. The size of each smaller sub-time slot is determined by the unloading decision variable α. U,k,m [n] Determined.

[0046] Step 3: Obtain the optimal optimization variables in the scenario of drone-assisted multi-mobile edge computing server:

[0047] (3.1) A three-dimensional Euclidean coordinate system is adopted, and the coordinates are measured in meters in this embodiment; assuming that the positions of the base station and all users are fixed on the ground at zero altitude, the horizontal positions of the base station m and user k are obtained as b m and w k Assume the drone flies at a fixed altitude H (H > 0) within the mission completion time T; and set the start and end points of the flight as q respectively. I and q F Based on the discrete path planning method, the horizontal position of the UAV in the nth time slot is obtained as q[n] = q[nτ], where q[0] = q I q[N]=q F Assuming that the wireless channels between the UAV and the base station, and between the UAV and the user, are dominated by line-of-sight links, the channel power gain h between the UAV and user k, and between the UAV and the base station m, is obtained within time slot n. U,k [n] and g U,m [n];

[0048] (3.2) Assume that each user has a specific computation task to be executed, and the computation task consists of triples. It means that I k C represents the size of the input data for the computation task. k T represents the computational resources required to input 1 bit of data. kLet T be the maximum tolerable delay for user k; let all users have the same maximum tolerable delay, i.e., T. k =T; Assume each user's task completion rate is μ k ;

[0049] (3.3) Each user uses a time-division multiple access (TDMA) protocol to offload a portion of the computational task to the UAV for computation, while the other portion is computed locally by the user. The UAV collects the offloaded computational tasks by optimizing its flight trajectory, and also uses a TDMA protocol to offload a portion of the collected computational tasks to multiple ground base stations for computation, while the other portion is computed locally by the UAV. The amount of task data executed locally by the k-th user within time slot n is obtained. and energy consumption The amount of task data that the kth user unloads onto the drone and transmission energy consumption The amount of task data that the drone locally calculates and executes for the kth user. and energy consumption The drone unloads the task data of the kth user to base station m. and transmission energy consumption and the flight energy consumption of drones

[0050] The amount of task data executed locally by the k-th user within time slot n. and energy consumption The amount of task data that the kth user unloads onto the drone and transmission energy consumption The amount of task data that the drone locally calculates and executes for the kth user. and energy consumption The drone unloads the task data of the kth user to base station m. and transmission energy consumption and the flight energy consumption of drones The results were obtained from the following calculations:

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] Where B represents the bandwidth from user k to the drone and the bandwidth from the drone to the base station, σ 2 κ represents the noise power of any node in the system. k and κ U Let C represent the chip capacitance coefficients of user k and the drone, respectively. U This represents the computational resources required for the drone to input 1 bit of data, and P(||v[n]||) represents the power consumption of the rotorcraft drone during flight, as detailed below:

[0061]

[0062] Among them, P0 and P H These are the airfoil power and induced power in hovering state, respectively; U tip v0, d0, g, s, and A are related to aerodynamics and represent the rotor blade tip velocity, the average rotor induced velocity during hovering, the fuselage drag ratio, air density, rotor solidity, and rotor area, respectively.

[0063] (3.4) Obtain the total energy consumption E of user k in each time slot n in this scenario according to the following formula. k [n] and the total energy consumption E of the drone in each time slot n U [n]:

[0064]

[0065]

[0066] (3.5) Construct the optimal task completion rate μ for this scenario max And the weighted energy consumption E of users and drones min expression:

[0067]

[0068] Wherein, the optimization variable is the user uninstallation decision α = {α U,k,m [n]}、User task completion rate μ={μ k}, Local computing CPU frequency F = {f k [n],f U,k [n]}, User and drone transmission power P={p k [n],p U,k,m [n]} and the drone trajectory Q={q[n]};ω U and ω EThese represent the energy consumption weight of the drone and the total energy consumption weight, respectively.

[0069] In this scenario, the user and the drone must satisfy the following constraints:

[0070] Indicates task completion constraints; This represents the causal constraint of information, where, in, 0≤μ k ≤1, 0≤f k [n]≤f k,max , 0≤p k [n]≤p k,max , p k [N-1]=p k [N] = 0, 0 ≤ f U,k [n]≤f U,max , f U,k [1]=f U,k [N] = 0, 0 ≤ p U,k,m [n]≤p U,max , p U,k,m [1]=p U,k,m [N] = 0, 0 ≤ α U,k,m [n]≤1, α U,k,m [1]=α U,k,m [N] = 0, q[0] = q I q[N]=q F , where f k,max f U,max p k,max and p U,max These represent the maximum available CPU frequency and maximum transmission power for the k-th user and the drone, respectively. ||q[n]-q[n-1]||≤τV max , This indicates the maximum speed constraint for the drone;

[0071] (3.6) In this scenario, to maximize the task completion rate and minimize the weighted energy consumption of the drone and the user, an alternating optimization algorithm is used to decouple the highly complex problem into a task offloading and resource allocation problem and a drone trajectory design problem, and to obtain the task completion rate μ. max And the weighted energy consumption E of users and drones min The corresponding optimal optimization variables are determined through the following steps:

[0072] (3.6.1) Fix the drone trajectory and use variable substitution to transform the task unloading and resource allocation problem into a convex problem, that is, use variable substitution to transform the expression constructed in step (3.5) into a convex expression;

[0073] (3.6.2) The transformed convex expression is solved iteratively using the Lagrange dual decomposition algorithm and the subgradient algorithm to obtain the optimal user unloading decision α. * Optimal user task completion rate μ * Optimal local computing CPU frequency F for users and drones * And the optimal user and drone transmission power P * ;

[0074] (3.6.3) Fix α * μ * F * and P * The optimal UAV trajectory Q is obtained by solving the UAV trajectory using a continuous convex approximation algorithm. * .

[0075] Step 4: Based on the optimal task completion rate μ obtained in step (3.6) max And the weighted energy consumption E of users and drones min The corresponding optimal optimization variables are used to set the system operating parameters, and the system is made to run under these parameters to achieve optimization of system task completion rate and energy consumption.

[0076] Example 2: This example provides a method for optimizing task completion rate and energy consumption in a drone-assisted multi-mobile edge computing server network. The overall implementation steps are the same as in Example 1, but further description is provided for a drone-assisted multi-mobile edge computing server scenario:

[0077] Step a: Build an edge computing network consisting of one drone, K users, and M base stations equipped with mobile edge computing servers; and Let k and m represent the sets of users and base stations, respectively, where k represents the k-th user and the m-th base station equipped with a mobile edge computing server.

[0078] Step b: Discretize the finite task completion time T into N equal time slots, let Let N be the set of N time slots. The duration of each time slot is τ = T / N, where τ is small enough that the UAV's position can be assumed to remain constant during each time slot. To avoid interference between users during the unloading process, a time division multiple access (TDMA) protocol is used. Each time slot is further divided into K equal sub-time slots, each with a duration of δ = T / (NK), and user k unloads its task input data in the k-th sub-time slot. To better distinguish unloading signals from different users, the UAV also uses a time division multiple access protocol with K equal time slots. Since the UAV needs to further partially offload each user's task input data to multiple ground base stations equipped with mobile edge computing servers, each sub-time slot is further divided into M smaller sub-time slots, the size of which is determined by the unloading decision variable α. U,k,m [n] Determined;

[0079] Step c: For ease of explanation, a three-dimensional Euclidean coordinate system is used, with coordinates measured in meters. Assuming the base station and all users are fixed at zero altitude, the horizontal positions of base station m and user k are obtained as b... m =(x m ,y m ) and w k =(x k ,y k Assume the drone flies at a fixed altitude H (H > 0) within the mission completion time T, and the start and end points of the flight are set to q respectively. I =(x I ,y I ) and q F =(x F ,y F Based on the discrete path planning method, the horizontal position of the UAV in the nth time slot is obtained as q[n]=q[nτ]=(x[n],y[n]), where q[0]=q I q[N]=q F Assuming the wireless channels between the UAV and the base station, and between the UAV and the user, are dominated by line-of-sight links, the channel power gains between the UAV and user k, and between the UAV and the base station m, within time slot n are respectively h U,k [n] and g U,m [n]:

[0080]

[0081]

[0082] Where ρ is the channel power gain at a reference distance of 1m, and d k [n] and d m[n] represents the distance between the drone and user k, and the distance between the drone and base station m, respectively.

[0083] Step d: Assume each user has a specific computational task to execute, and the computational task consists of triples. It means that I k C represents the size of the input data for the computation task. k T represents the computational resources required to input 1 bit of data. k Let T be the maximum tolerable delay for user k, where T is the maximum tolerable delay for user k. k ≤T. Assume all users k have the same maximum tolerable delay, i.e., T. k =T. Assume each user's task completion rate is μ. k ;

[0084] Step e: Each user uses the Time Division Multiple Access (TDMA) protocol to offload part of the computing task to the UAV for computing, and performs the other part of the computing locally; the UAV collects the computing tasks offloaded by the user by optimizing its flight trajectory, and also uses the TDMA protocol to offload part of the collected computing tasks to multiple base stations on the ground for computing, and performs the other part of the computing locally on the UAV.

[0085] make and These represent the data volume and energy consumption of the task executed locally by the k-th user within time slot n, respectively, as shown in the following expressions:

[0086]

[0087]

[0088] Where, f k [n] represents the CPU clock frequency of user k within time slot n, κ k Let k be the chip capacitance coefficient for user k.

[0089] make and This represents the amount of task data and transmission energy consumed by the k-th user to unload onto the drone within time slot n, as detailed below:

[0090]

[0091]

[0092] Where, p k [n] represents the transmission power of user k in time slot n, B represents the bandwidth from user k to the drone, and σ 2Let represent the noise power at the drone. Without loss of generality, assume that the bandwidth from user k to the drone is the same as the bandwidth from the drone to the base station, and that the noise power of any node in the system is considered to be equal to σ. 2 same.

[0093] make and The expression represents the amount of data and energy consumed by the drone in locally executing the task of the kth user, as shown below:

[0094]

[0095]

[0096] Among them, f U,k [n] represents the CPU clock frequency allocated by the drone to user k within time slot n, C U κ represents the computational resources required for a drone to input 1 bit of data. U This represents the chip capacitance coefficient of the drone.

[0097] make and This represents the amount of task data and transmission energy consumed by the drone to unload the k-th user's data from the base station m, as detailed below:

[0098]

[0099]

[0100] Where, p U,k,m [n] represents the transmission power of the UAV unloading user k task from the base station m within time slot n.

[0101] Let represent the flight energy consumption of the drone, and the specific expression is as follows:

[0102]

[0103] In the formula, P(||v[n]||) represents the power consumption of the rotary-wing UAV during flight, as follows:

[0104]

[0105] Among them, P0 and P H These are the airfoil power and induced power in hovering state, U tip v0, d0, g, s, and A are related to aerodynamics and represent the rotor blade tip velocity, the average rotor induced velocity during hovering, the fuselage drag ratio, air density, rotor solidity, and rotor area, respectively.

[0106] Step f: Obtain the total energy consumption E of user k in each time slot n according to the following formula.k [n] and the total energy consumption E of the drone in each time slot n U [n]:

[0107]

[0108]

[0109] Step g: Establish the optimization problem for a drone-assisted multi-mobile edge computing server scenario, expressed as follows:

[0110]

[0111] Wherein, the optimization variable is α={α U,k,m [n]}, μ={μ k}, F={f k [n],f U,k [n]}, P={p k [n],p U,k,m [n]}, Q={q[n]};ω U and ω E Let represent the energy consumption weight of the drone and the total energy consumption weight, respectively; the task completion constraint is... and Information causal constraint is In the formula, in, The task completion rate is constrained to be 0 ≤ μ k ≤1; The user's computational task constraint is 0≤f k [n]≤f k,max , 0≤p k [n]≤p k,max , p k [N-1]=p k [N] = 0, where f k,max and p k,max These represent the maximum available CPU frequency and maximum transmission power for the k-th user, respectively; the computational task constraint for the UAV is 0 ≤ f. U,k [n]≤f U,max , f U,k [1]=f U,k [N] = 0, 0 ≤ p U,k,m [n]≤p U,max , p U,k,m [1]=p U,k,m [N] = 0, where f U,max and p U,maxThese represent the maximum available CPU frequency and maximum transmit power of the UAV, respectively; the constraint for the offloading decision variables of the UAV to different base stations is 0 ≤ α. U,k,m [n]≤1, α U,k,m [1]=α U,k,m [N] = 0; the initial and final horizontal position constraints of the UAV are q[0] = q I q[N]=q F The maximum speed constraint for the UAV is ||q[n]-q[n-1]||≤τV max ,

[0112] Step h: Solve equation <1.1> using alternating iteration, variable substitution, Lagrange dual decomposition algorithm, subgradient algorithm and continuous convex approximation algorithm to obtain the optimal parameters.

[0113] (8.1) Due to the variable α in the objective function U,k,m [n] and p U,k,m There is a nonlinear coupling between [n] and this variable is also strongly coupled with the UAV trajectory q[n]. Formula <1.1> is a complex non-convex expression. To solve it, a two-step alternating optimization algorithm is proposed. In the first step, given the UAV trajectory Q, the unloading decision variable α, the task completion rate variable μ, and the computation and communication resource scheduling variables F and P in the expression are solved.

[0114] (8.1.1) Considering the coupling between the objective function and the constraint variables, formula <1.1> is a non-convex expression. To solve it, an auxiliary variable β is introduced. U,k,m [n], such that β U,k,m [n] = α U,k,m [n]p U,k,m [n], thus eliminating variable coupling, and formula <1.1> is transformed into:

[0115]

[0116] In the formula, p = {p k [n]}, β={β U,k,m [n]}; constraints It became constraint It became Constraint 0≤p U,k,m [n]≤p U,max , It became 0≤β U,k,m [n]≤α U,k,m [n]p U,max , constraint pU,k,m [1]=p U,k,m [N] = 0 became β U,k,m [1]=β U,k,m [N] = 0, where,

[0117] (8.1.2) Let Obviously φ U,k,m [i] is ζ U,k,m The perspective function of [i]. At this time, ζ U,k,m If [i] is a concave function, then φ U,k,m [i] is also a concave function. Note that φ U,k,m [i] After approximation, constraints The left side is about β U,k,m The non-convex constraint of [i]. Therefore, the concave function of the logarithmic term on the left side of the inequality is transformed into a locally convex approximation using a continuous convex approximation algorithm. Specifically, for a given locally feasible point... Where j represents the number of iterations in the continuous convex approximation algorithm, and the logarithmic term on the left-hand side of the constraint is approximated as:

[0118]

[0119] in,

[0120]

[0121] Therefore, formula <1.2> is transformed into:

[0122]

[0123] Among them, constraints It became

[0124] (8.1.3) Formula <1.3> is a convex expression, and it is solved using the Lagrange duality method. The Lagrangian function of formula <1.3> is:

[0125]

[0126] In the formula, ο={ο k}, λ={λ k,n}, η={η k}, The dual function of formula <1.3> is:

[0127]

[0128] The constraint is 0 ≤ μ k ≤1, 0≤f k [n]≤fk,max , 0≤p k [n]≤p k,max , p k [N-1]=p k [N] = 0, 0 ≤ f U,k [n]≤f U,max , f U,k [1]=f U,k [N] = 0, 0 ≤ β U,k,m [n]≤α U,k,m [n]p U,max , β U,k,m [1]=β U,k,m [N] = 0, 0 ≤ α U,k,m [n]≤1, α U,k,m [1]=α U,k,m [N] = 0.

[0129] (8.1.4) Using the dual decomposition method, formula <1.4> is decomposed into the following sub-formulas, as follows:

[0130] For variable f k [n], the corresponding sub-formula is:

[0131]

[0132] The constraint is 0 ≤ f k [n]≤f k,max , Formula <1.4.1> is a convex expression, and its optimal solution can be obtained using the KKT conditions:

[0133]

[0134] For variable f U,k [n], the corresponding sub-formula is:

[0135]

[0136] Where the constraint is f U,k [1]=f U,k [N] = 0, 0 ≤ β U,k,m [n]≤α U,k,m [n]p U,max , Formula <1.4.2> is a convex expression, and its optimal solution can be obtained using the KKT conditions:

[0137]

[0138] For variable p k [n], the corresponding sub-formula is:

[0139]

[0140] The constraint is 0 ≤ p k [n]≤p k,max , p k [N-1]=p k [N] = 0. Formula <1.4.3> is a convex expression, and its optimal solution can be obtained using the KKT conditions:

[0141]

[0142] For variable α U,k,m [n] and β U,k,m [n], the corresponding sub-formula is:

[0143]

[0144] The constraint is 0 ≤ β U,k,m [n]≤α U,k,m [n]p U,max , β U,k,m [1]=β U,k,m [N] = 0, 0 ≤ α U,k,m [n]≤1, α U,k,m [1]=α U,k,m [N] = 0. Formula <1.4.4> is difficult to express in a closed form; it can be solved using a block-based iterative algorithm. First, initialize α. U,k,m If [n] is a feasible solution, then formula <1.4.4> becomes:

[0145]

[0146] The constraint is 0 ≤ β U,k,m [n]≤α U,k,m [n]p U,max , β U,k,m [1]=β U,k,m [N] = 0. Formula <1.4.4.1> is a convex expression, and its optimal solution can be obtained using the KKT conditions:

[0147]

[0148] make Solve for variable α U,k,m [n], Formula <1.4.4> becomes:

[0149]

[0150] The constraint is 0 ≤ α U,k,m [n]≤1, α U,k,m [1]=α U,k,m [N] = 0. Formula <1.4.4.2> still fails to provide a closed-form solution; standard convex optimization tools are used to solve for α. U,k,m The optimal solution for [n].

[0151] Get Afterwards, order Returning to the solution formula <1.4.4.1>, and repeating the above steps until convergence, the final optimal solution can be obtained. and

[0152] (8.1.5) After solving each sub-formula, the dual problem of formula <1.4> needs to be solved. The dual problem is:

[0153]

[0154] Where the constraints are Since the Salter condition is satisfied, the optimal value of equation <1.5> is the same as the optimal value of equation <1.4>. Therefore, the subgradient method can be used to solve the dual problem, optimizing the dual variable. k , and η k The subgradients are respectively:

[0155]

[0156]

[0157]

[0158]

[0159] (8.2) The second step focuses on designing the UAV trajectory Q using optimized variables α, μ, F and P.

[0160] (8.2.1) By fixing the previously optimized unloading decisions, task completion rates, and computational and communication resources, the UAV trajectory design problem can be expressed as follows:

[0161]

[0162] Where the constraints are q[0]=q I ,q[N]=qF , ||q[n]-q[n-1]||≤τV max , in,

[0163] (8.2.2) Obviously, formula <1.5> is a non-convex expression. Therefore, the continuous convex approximation algorithm is used to optimize the non-convex terms in formula <1.5> to obtain its approximate expression. First, for the non-convex term P(||v[n]||) in the objective function, variables v1[n] and v2[n] are introduced, satisfying:

[0164] v1[n]≥||v[n]||

[0165]

[0166] Therefore, we can obtain:

[0167]

[0168] Using the continuous convex approximation algorithm, given any feasible solution of v1[n] and v2[n]... and Formula <1.6> is approximated by the following convex constraint:

[0169]

[0170] in,

[0171]

[0172] Therefore, the nonconvex term P(||v[n]||) is replaced by the following convex approximation.

[0173]

[0174] (8.2.3) Under the constraints of formula <1.5>, ξ1[n] and ξ2[n] are nonconvex with respect to q[n], but nonconvex with respect to ||q[n]-b m || 2 and ||q[n]-w k || 2 The entire function is convex. Based on this, using the continuous convex approximation algorithm, given any feasible point q of q[n]... (j) [n] has

[0175]

[0176]

[0177] in,

[0178] However, constraints The left side ξ1[n] utilizes Even after approximation, it remains a non-convex constraint on q[n]. Therefore, let p U,k,m [n]g U,m [n]| / σ 2 ≥1 / Q[n], performing a first-order Taylor expansion gives:

[0179]

[0180] (8.2.4) Based on the continuous convex approximation algorithm, the suboptimal solution to the original problem can be obtained by solving the following convex approximation expression:

[0181]

[0182] Where the constraints are q[0]=q I ,q[N]=q F , ||q[n]-q[n-1]||≤τV max , v1[n]≥||v[n]||, In the formula, in,

[0183] Equation <1.7> is a standard convex expression. However, the positions of the UAV in different time slots are coupled, making it difficult to obtain a closed-form solution for q[n]. In this case, standard convex optimization tools are used to solve for an approximate expression of Equation <1.7>.

[0184] (8.3) Repeat steps (8.1)-(8.2) until the algorithm converges and the optimal optimization variable, i.e. the optimal user unloading decision α, is obtained. * Optimal user task completion rate μ * Optimal local computing CPU frequency F for users and drones * Optimal user and drone transmission power P * And the optimal drone trajectory Q * .

[0185] Step i: The system selects working parameters based on the optimal optimization variables to achieve optimal system performance.

[0186] This invention addresses the problems of severe channel fading and limited computing resources of a single mobile edge computing server in traditional mobile edge computing networks. First, it leverages the advantages of UAVs, such as ease of deployment, flexible mobility, and line-of-sight link connections, to establish a reliable channel connection with ground users. Second, it utilizes multiple mobile edge computing servers to enhance network computing resources, achieving stability in environments with high computational workloads, while simultaneously reducing energy consumption for both UAVs and users. This invention can be used in edge computing systems to improve task completion rates.

[0187] The effects of the present invention will be further illustrated below with simulation experiments:

[0188] A. Simulation conditions

[0189] The simulation was conducted using computer simulation software, setting the drone's flight start and end points to (0,0) and (40,50) respectively, considering three scenarios with 1, 2, and 3 base stations. When there is 1 base station, the base station location is set to (0,0); when there are 2 base stations, the base station locations are set to (0,0) and (40,50); when there are 3 base stations, the base station locations are set to (0,0), (20,25), and (40,50). The simulation will analyze the impact of key parameters, including the user's computational workload I. k Number of base stations M, system bandwidth B, mission completion time T, and UAV energy consumption weight ω U Total energy consumption weight ω E Unless otherwise stated, the basic simulation parameters are listed in Table 1.

[0190] Table 1 Simulation Parameters

[0191]

[0192]

[0193] B. Simulation Content

[0194] Simulation 1: UAV trajectories corresponding to different task completion times. Simulation results are as follows: Figure 4 As shown;

[0195] Simulation 2: UAV speeds corresponding to different task completion times; simulation results are as follows. Figure 5 As shown;

[0196] Simulation 3: The impact of task load and number of base stations on the weighted sum of task completion rate and energy consumption. Simulation results are as follows: Figure 6 As shown;

[0197] Simulation 4: The impact of task load and number of base stations on task completion rate and energy consumption, respectively. Simulation results are as follows: Figure 7 As shown;

[0198] Simulation 5: The impact of UAV weights and system bandwidth on the weighted sum of task completion rate and energy consumption. Simulation results are as follows: Figure 8 As shown;

[0199] Simulation 6: The impact of UAV weights and system bandwidth on mission completion rate and energy consumption, respectively. Simulation results are as follows: Figure 9 As shown;

[0200] Simulation 7: The impact of UAV weight and energy consumption weight on mission completion rate and energy consumption weighted sum. Simulation results are as follows: Figure 10 As shown;

[0201] Simulation 8: The impact of UAV weight and energy consumption weight on mission completion rate and energy consumption, respectively. Simulation results are as follows: Figure 11 As shown;

[0202] C. Simulation Results

[0203] Depend on Figure 4 It is evident that as the task completion time T increases, the drone can utilize its maneuverability to approach the user's location more closely. This is because flying closer to the user reduces path loss. Combined with... Figure 5 It can also be observed that for longer task completion times (e.g., T = 7.5s and T = 10.5s), the drone trajectory tends to stabilize, initially flying at maximum speed, then decelerating, and even tending to hover at a fixed point, which optimally allows the user to perform task offloading and thus reduce the user's computational energy consumption. Furthermore, in the... Figure 4 In time slots 11 and 12, it can be observed that as the drone gradually approaches the base station, its speed begins to decrease and it even hovers. This is because by approaching the base station, the drone can reduce its own computational and offloading energy consumption.

[0204] Depend on Figure 6 It is evident that as the user's task load increases, the weighted sum of task completion rate and energy consumption also increases. Furthermore, as the number of base stations increases, the weighted sum of task completion rate and energy consumption decreases, and this effect becomes more pronounced with increasing task load.

[0205] Depend on Figure 7 It is evident that when the user's computational workload is relatively small (e.g., I...), k With a base station capacity of 12Mbit, the combined energy consumption of the drone and the user actually increases with the presence of three base stations. This is because the drone needs to offload the user's task input data to more base stations, increasing its communication energy consumption. However, when the task load is large, the drone can reduce its own computing energy consumption by offloading the user's task input data to more base stations. Therefore, the drone needs to find a balance between its own computing energy consumption and communication energy consumption. Furthermore, compared to a single-base station scenario, a multi-base station scenario can significantly improve the user's completion rate.

[0206] Depend on Figure 8 It is evident that the weighted sum of task completion rate and energy consumption increases with the increase of the drone's weight, and decreases with the increase of system bandwidth. Figure 9 It is evident that increasing the drone's weight leads to a continuous decrease in both the drone's and the user's energy consumption. This is because the drone's energy consumption is primarily flight energy, and increasing the drone's weight makes the optimization objective function focus more on the drone's flight energy consumption. Therefore, the drone will reduce its close-to-user flight and hovering, which also reduces the task completion rate as the drone's weight increases. Furthermore, increased system bandwidth allows the user to offload more task input data to the drone, increasing the drone's computational and communication energy consumption. However, with increased system bandwidth, the drone can offload more data to the ground base station, improving the task completion rate.

[0207] Depend on Figure 10 It is evident that as the weight of the drone and its energy consumption increases, the weighted sum of mission completion rate and energy consumption also increases. Figure 11 It is evident that the increased weighting of energy consumption means that the optimization objective function, while focusing on the drone's energy consumption, also places greater emphasis on the user's energy consumption. Due to increased bandwidth, both the user and the drone have more decision-making space regarding power. As the drone reduces its flight energy consumption, the user will also reduce their own computing and communication energy consumption. Therefore, the task completion rate decreases as the energy consumption weighting increases. Thus, the objective function needs to find a balance between these two factors: energy consumption weighting and task completion rate.

[0208] The above simulation analysis proves the correctness and effectiveness of the method proposed in this invention.

[0209] The parts of this invention not described in detail are common knowledge to those skilled in the art.

[0210] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Obviously, those skilled in the art, after understanding the content and principle of the present invention, may make various modifications and changes in form and detail without departing from the principle and structure of the present invention. However, these modifications and changes based on the concept of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for maximizing task completion rate of a UAV-assisted multi-MEC server, the method comprising: Includes the following steps: (1) Building an edge computing network model: Set up a drone individual users and An edge computing network model consisting of base stations equipped with mobile edge computing (MEC) servers; and Let represent the sets of users and base stations, respectively. and They represent the first The user and the first A base station equipped with a mobile edge computing server; (2) Dividing the time slot structure for users and drones: Limited task completion time Discretize into An equal time slot, let express A set of time slots; the duration of each time slot is Assuming the drone's position remains constant during each time slot; The user uses a time division multiple access (TDMA) protocol to divide each time slot into... Equal sub-time slots, The duration of each sub-slot is And users In its corresponding first Unload its task input data in each sub-slot; The drone also uses A time-division multiple access protocol with equal time division divides each sub-slot into... Each sub-slot is a sub-slot, the size of which is determined by the offloading decision variable. Sure; (3) Obtain the optimal optimization variables in the scenario of UAV-assisted multi-mobile edge computing server: (3.1) Using a three-dimensional Euclidean coordinate system, assuming that the base station and all users are located on the ground at zero elevation, the base station is obtained. and users The horizontal positions are respectively and Assuming the drone completes its mission within the specified time... With fixed height inside flight, Furthermore, the flight start and end points are set as follows: and ; Based on the discrete path planning method, we obtain the result at the first... The horizontal position of the time slot drone ,in , Assuming that the wireless channels between the drone and the base station, as well as between the drone and the user, are dominated by line-of-sight links, the results in time slots are obtained. Internal drones and users Between and between drones and base stations Channel power gain between and ; (3.2) Assume that each user has a specific computation task to be executed, and the computation task consists of triples. It means that among them This indicates the size of the input data for the computation task. This represents the computational resources required to input 1 bit of data. For users The maximum tolerable latency; ensuring that all users have the same maximum tolerable latency, i.e. Assume each user's task completion rate is... ; (3.3) Each user uses the time division multiple access protocol to offload part of the computing task to the UAV for computing, and performs the other part of the computing locally; The drone collects computational tasks offloaded by the user by optimizing its flight trajectory. It also employs a time-division multiple access (TDMA) protocol to offload a portion of the collected computational tasks to multiple ground base stations for computation, while the remaining portion is computed locally by the drone; thus obtaining the computational results within the time slot. Inner The amount of task data executed by a user's local computing and energy consumption , No. The amount of task data that a user unloads onto the drone and transmission energy consumption The drone's local computing execution Task data volume per user and energy consumption Drones to base stations Uninstall the Task data volume per user and transmission energy consumption and the flight energy consumption of drones ; (3.4) Obtain the user according to the following formula In each time slot Total energy consumption and drones in each time slot Total energy consumption : ； ； (3.5) Constructing the optimal task completion rate and user and drone weighted energy consumption The expression: ； The optimization variable is the user uninstallation decision. User task completion rate Local computing CPU frequency of users and drones Transmission power of users and drones and drone trajectories ; and These represent the energy consumption weight of the drone and the total energy consumption weight, respectively. The user and the drone are set to meet the following constraints: and Indicates task completion constraints; This represents the causal constraint of information, where, ,in, ; , , , , , , , , , , , , , , , , , , ,in , , and The first Maximum available CPU frequency and maximum transmit power for each user and drone. ; , This indicates the maximum speed constraint for the drone; Indicates user In the time slot CPU clock frequency; Indicates user In the time slot Internal transmission power; Indicates that the drone is in the time slot Internal for users The allocated CPU clock frequency; Indicates that the drone is in the time slot Inward base station Uninstall user The transmission power of the task; (3.6) Maximize the task completion rate and minimize the weighted energy consumption of the UAV and the user, and obtain the task completion rate through an alternating optimization algorithm. and the weighted energy consumption of users and drones The corresponding optimal optimization variable; (4) The optimal task completion rate obtained in step (3.6) and the weighted energy consumption of users and drones The corresponding optimal optimization variables are used to set the system operating parameters, and the system is made to run under these parameters to achieve optimization of system task completion rate and energy consumption.

2. The method according to claim 1, characterized in that: In step (3.3), the time slot is described Inner The amount of task data executed by a user's local computing and energy consumption , No. The amount of task data that a user unloads onto the drone and transmission energy consumption The drone's local computing execution Task data volume per user and energy consumption Drones to base stations Uninstall the Task data volume per user and transmission energy consumption and the flight energy consumption of drones The results are obtained from the following calculations. ； ； ； ； ； ； ； ； ； in, Indicates user The bandwidth to the drone and the bandwidth from the drone to the base station, This represents the noise power of any node in the system. and Representing users respectively And the chip capacitance coefficient of the drone, This represents the computing resources required for a drone to input 1 bit of data. The power consumption of the rotary-wing drone during flight is as follows: ； in, and These represent the airfoil power and induced power in hovering mode, respectively. , , , , and These represent the rotor blade tip velocity, average rotor induced velocity during hovering, fuselage drag ratio, air density, rotor solidity, and rotor area, respectively.

3. The method according to claim 1, characterized in that: The task completion rate is obtained through the alternating optimization algorithm described in step (3.6). and the weighted energy consumption of users and drones The corresponding optimal optimization variables are implemented as follows: (3.6.1) Fix the UAV trajectory and use variable substitution to convert the expression constructed in step (3.5) into a convex expression; (3.6.2) The transformed convex expression is solved iteratively using the Lagrange dual decomposition algorithm and the subgradient algorithm to obtain the optimal user unloading decision. Optimal user task completion rate Optimal local computing CPU frequency for users and drones And optimal transmission power for users and drones ; (3.6.3) Fixed , , and The optimal UAV trajectory is obtained by solving the UAV trajectory using a continuous convex approximation algorithm. .

Images