Optimization method for minimizing the computation time of integrated tasks in MEC networks under thermal-sensitive conditions

Abstract

The present application discloses an optimization method for minimizing the computing time of integrated tasks in a MEC network under thermal-sensitive conditions. The optimization method is for a MEC network supported by a drone, in which the drone provides computing services to users during cruising, so that the drone acts as an aerial MEC server to expand the coverage of the MEC network. The optimization method minimizes the average computing time of users by jointly optimizing task scheduling, trajectory, computing and communication resource allocation strategies, while taking into account the CPU thermal constraints of the airborne MEC server. The resulting optimization method is a low-complexity algorithm based on task urgency, which is processed using the iterative algorithm of the successive convex approximation (SCA) technology and the block coordinate descent method. The output average computing time minimum system optimization model can take into account CPU frequency, CPU temperature and average computing time, improve the balance of the strategy obtained by the model, and effectively maintain the average computing time and service stability of the drone airborne MEC network.

Description

Technical Field

[0001] The present invention relates to the field of information and communication engineering technology, and in particular to orthogonal frequency division multiple access technology, partial offloading technology and resource allocation technology of mobile edge computing in wireless communication systems, and in particular to a method for minimizing the computing time of integrated tasks in a mobile edge computing (MEC) network under thermal-sensitive conditions. Background Art

[0002] Mobile edge computing (MEC) provides users with a high quality of experience (QoE) by placing servers close to end users. Compared with local computing, MEC helps save energy, but it also causes communication delays. There have been many related studies on the problem of meeting user quality experience requirements in a multi-user environment while minimizing the energy consumption of mobile devices.

[0003] Dai Yueyue studied the minimization of total energy consumption of MEC system from the perspective of user association optimization in the paper "Joint computation offading and user association in multitask mobile edge". Ren Jinke realized the minimization of energy consumption of MEC network based on time division multiple access in the paper "Latency Optimization for Resource Allocation in Mobile-Edge Computation Offloading". Laizhong Cui considered the balance between energy consumption minimization and latency in the paper "Joint Optimization of Energy Consumption and Latency in Mobile Edge Computing for Internet of Things" and modeled the scenario as a constrained multi-objective optimization problem. Pouria Paymard studied the operator's profit maximization, latency minimization and energy consumption minimization with orthogonal frequency division multiple access technology in the paper "Joint Optimization of Energy Consumption and Latency in Mobile Edge Computing for Internet of Things". Khalili and Ata considered the QoE requirements for time-sensitive computing tasks in the paper "Joint resource allocation and offloading decision in mobile edge computing", but only considered full offloading techniques.

[0004] The existing technologies do not consider the CPU temperature constraint, which affects the computing resources and device performance stability of the MEC network onboard the unmanned aerial vehicle (UAV), and do not propose a method to minimize the centralized computing time of the MEC network onboard the UAV. Summary of the invention

[0005] In response to the above technical problems, the present application provides a method for minimizing the integrated task computing time in the MEC network under thermal-sensitive conditions, and a method for minimizing the user's average computing time under CPU thermal constraints by jointly optimizing the UAV trajectory, task scheduling strategy, and computing and communication resource allocation strategy.

[0006] The present application provides a method for minimizing the integrated task computing time in a mobile edge computing (MEC) network under thermal conditions, comprising the following steps: Step S1: Initialization phase: In this phase, nodes in the network obtain basic configuration information of the network and initialize relevant parameters, including: local computing resources and computing power, initial values ​​of drone flight trajectories, initial values ​​of user bandwidth allocation, initial values ​​of task planning strategies, and initial values ​​of computing resource allocation variables;

[0007] Step S2: Establishing a system optimization model: According to the overall goal of minimizing the average computing time of the user and the constraints, an average computing time minimum system optimization model is established. The constraints include: latency, CPU temperature, maximum computing resources and processing speed. According to the constraints, the relevant parameters are substituted into the average computing time minimum system optimization model to obtain the initial optimal UAV trajectory. Based on the initial optimal UAV trajectory, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K and the computing resource allocation are calculated;

[0008] Average computing time minimum system optimization model:

[0009]

[0010]

[0011]

[0012]

[0013] S≤S max , (12e)

[0014]

[0015]

[0016]

[0017]

[0018]

[0019] c k [n] represents the computing planning strategy of the kth user in the nth time slot, represents the position of the drone in the nth time slot, where z = H 1 , a k [n] = 1 means that the user offloads the task to the drone in the nth time slot, b k [n] represents the bandwidth portion allocated by UVA to the kth user in the nth time slot, f k[n] represents the CPU resources allocated to the kth user at the nth time slot, r k [n] represents the unloading rate of the kth user at the nth time slot, F max represents the maximum computing capacity of the UAV, S max

[0020] is the maximum temperature of the processor, where

[0021] Using the BCD method Decoupling is:

[0022] 1) Trajectory optimization;

[0023] 2) Communication resource allocation;

[0024] 3) Task scheduling;

[0025] 4) Computational resource allocation;

[0026] For k users Decouple and calculate the optimal UAV trajectory corresponding to user k, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K, and the computing resource allocation;

[0027] Step S3: Iterative loop calculation: While satisfying the constraints, the optimal UAV trajectory, optimal bandwidth, mission planning strategy and computing resources obtained in the current round are used to enter the next round of iterative loop calculation according to the block coordinate descent method;

[0028] Step S4: When the result of the cyclic calculation reaches the set target accuracy, the optimal drone trajectory, optimal bandwidth, task planning strategy and computing resource allocation for different computing tasks of each user node are obtained.

[0029] Preferably, in step S2 Trajectory optimization includes the following steps: Solved by CVX method:

[0030]

[0031]

[0032]

[0033] Get the best feasible trajectory Q for user k ＊ .

[0034] Preferably, in step S2, communication resources are allocated Follow these steps to solve:

[0035] According to the feasible trajectory Q obtained by solving the optimal trajectory ＊ , feasible bandwidth allocation B ＊ Solved using CVX method To get the uninstall delay that meets all users' needs:

[0036]

[0037] st (12f), (12i), .

[0038] Preferably, in step S2 The task scheduling is solved by the following steps:

[0039] The feasible trajectory Q is obtained based on the optimal trajectory obtained by solving ＊ , feasible bandwidth allocation B ＊ And given F, in order to meet the requirement of making full use of computing resources to reduce the average time consumption of users, the optimal task planning strategy C is obtained ＊ .

[0040] Preferably, in step S2 Task scheduling solution includes the following steps:

[0041] Step S21: According to the urgency: Calculate the scheduling order of each task in descending order, given as

[0042] Among them, x k yes The index of the user in;

[0043] Step S22: sorting the task scheduling order according to the above urgency;

[0044] Step S23: Use an integer programming solver to solve the xth task in the task sequence k User Task Scheduling No. x k The task planning problem for each user is:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] Preferably, in step S2 Computer resource allocation is solved by the following steps:

[0052] Based on the solution The optimal task planning strategy C obtained by task scheduling ＊ , solved using CVX:

[0053]

[0054] stS≤S max , (22b)

[0055]

[0056] get Computer resource allocation.

[0057] The beneficial effects of this application include:

[0058] 1) The optimization method for minimizing the computing time of integrated tasks in MEC networks under thermal-sensitive conditions provided in this application is aimed at drone-assisted MEC networks. By jointly optimizing task scheduling, trajectory, computing and communication resource allocation strategies, the average computing time of users is minimized while considering the CPU thermal constraints of the onboard MEC server. The average computing time minimum system optimization model output by the optimization method can take into account CPU frequency, CPU temperature and average computing time, improve the balance of the strategy obtained by the model, and effectively maintain the average computing time and service stability of the drone-borne MEC network.

[0059] 2) The optimization method for minimizing the computing time of integrated tasks in MEC networks under thermal-sensitive conditions provided in this application takes into account the delay requirements of the tasks and the temperature constraints of the drones. The numerical simulation results using the improved method of this application show that the WTC scheme proposed in this application is superior to the current benchmark scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the process flow of the method for minimizing the computing time of integrated tasks in MEC networks under thermal sensitive conditions provided in this application

[0061] Figure 2 A schematic diagram of the trajectory of a drone with different source points and end points according to the simulation results in the embodiment of the present application;

[0062] Figure 3 A discount graph showing the relationship between the number of users and the average time used for calculation results in each implementation method of the present application;

[0063] Figure 4This is a diagram showing the relationship between CPU frequency, CPU temperature and average computing time for each implementation in the embodiments of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0065] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Technical features that are not used to solve the technical problems of the present application are all set or installed according to the commonly used methods in the prior art and will not be described here.

[0067] See also Figure 1 The present application provides a method for minimizing the computing time of integrated tasks in a MEC network under thermal-sensitive conditions, comprising the following steps:

[0068] Step S1: Initialization phase: In this phase, the nodes in the network obtain the basic configuration information of the network and initialize the relevant parameters, including: local computing resources and computing power, initial values ​​of UAV flight trajectory, initial values ​​of user bandwidth allocation, initial values ​​of mission planning strategy and initial values ​​of computing resource allocation variables;

[0069] Step S2: Establishing a system optimization model: According to the overall goal of minimizing the average computing time of the user and the constraints, an average computing time minimum system optimization model is established. The constraints include: latency, CPU temperature, maximum computing resources and processing speed. According to the constraints, the relevant parameters are substituted into the average computing time minimum system optimization model to obtain the initial optimal UAV trajectory. Based on the initial optimal UAV trajectory, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K and the computing resource allocation are calculated;

[0070] Average computing time minimum system optimization model:

[0071]

[0072]

[0073]

[0074]

[0075] S≤S max , (12e)

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] c k [n] represents the computing planning strategy of the kth user in the nth time slot, represents the position of the drone in the nth time slot, where z = H 1 , a k [n] = 1 means that the user offloads the task to the drone in the nth time slot, b k [n] represents the bandwidth portion allocated by UVA to the kth user in the nth time slot, f k [n] represents the CPU resources allocated to the kth user at the nth time slot, r k [n] represents the unloading rate of the kth user at the nth time slot, F max represents the maximum computing capacity of the UAV, S max is the maximum temperature of the processor, where

[0082] Using the BCD method Decoupling is:

[0083] 1) Trajectory optimization;

[0084] 2) Communication resource allocation;

[0085] 3) Task scheduling;

[0086] 4) Computational resource allocation;

[0087] For k users Decouple and calculate the optimal UAV trajectory corresponding to user k, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K, and the computing resource allocation;

[0088] Step S3: Iterative loop calculation: While satisfying the constraints, the optimal UAV trajectory, optimal bandwidth, mission planning strategy and computing resources obtained in the current round are used to enter the next round of iterative loop calculation according to the block coordinate descent method (BCD);

[0089] Step S4: When the result of the cyclic calculation reaches an acceptable accuracy (the set target accuracy), the optimal drone trajectory, optimal bandwidth, task planning strategy and computing resource allocation for different computing tasks of each user node are obtained.

[0090] Preferably, in step S2 Trajectory optimization includes the following steps: Solved by CVX method:

[0091]

[0092]

[0093]

[0094] Get the best feasible trajectory Q for user k ＊ .

[0095] Specifically, since users need to offload tasks to drones within a given time, optimizing the drone trajectory to improve the communication channel between users and drones is crucial to increasing the data transmission rate. Therefore, the relevant problem is formulated as

[0096]

[0097] st (12i), (12j).

[0098] It is non-convex with respect to Q, and uses an SCA-based method to convert the non-convex constraints into a suitable convex constraint approximation, relaxing the variables is introduced to solve the problem about r in (12i) k The non-convex part of [n], that is:

[0099]

[0100] in ”r k [n]For is convex, and given the trajectory at the i-th iteration In the case of k The lower bound of [n] is:

[0101]

[0102] in and The problem is reformulated as:

[0103]

[0104]

[0105]

[0106]

[0107] It is convex and can be solved directly using CVX methods.

[0108] Preferably, in step S2, communication resources are allocated Follow these steps to solve:

[0109] According to the feasible trajectory Q obtained by solving the optimal trajectory ＊ , feasible bandwidth allocation B ＊ Solved using CVX method To get the uninstall delay that meets all users' needs:

[0110]

[0111] st (12f), (12i), .

[0112] Specifically, with the feasible trajectory Q obtained ＊ , feasible bandwidth allocation B ＊ By solving To meet the uninstall delay needs of all users.

[0113]

[0114] st (12f), (12i),

[0115] It is convex with respect to B and can be solved directly by CVX methods.

[0116] Preferably, in step S2 The task scheduling is solved by the following steps:

[0117] The feasible trajectory Q is obtained based on the optimal trajectory obtained by solving ＊ , feasible bandwidth allocation B ＊ And given F, in order to meet the requirement of making full use of computing resources to reduce the average time consumption of users, the optimal task planning strategy C is obtained ＊ .

[0118] Preferably, in step S2 Task scheduling solution includes the following steps:

[0119] Step S21: According to the urgency: Calculate the scheduling order of each task in descending order, given as

[0120] Among them, x k yes The index of the user in;

[0121] Step S22: sorting the task scheduling order according to the above urgency;

[0122] Step S23: Use an integer programming solver to solve the xth task in the task sequence k User Task Scheduling No. x k The task planning problem for each user is:

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] Specifically, Task scheduling is accomplished by solving the following problems:

[0130]

[0131] st (12b)-(12e), (12g), (12h).

[0132] It can be seen that It is an integer programming about C, which is difficult to solve. Although the B&B method can solve this problem, it is difficult to obtain the optimal strategy in polynomial time as the number of users and time slots increase. In order to solve this problem, a low-complexity task scheduling method is proposed according to the urgency of the task, which is defined as:

[0133]

[0134] First, according to the above urgency, the scheduling order of tasks is calculated in descending order, given as where x k yes The index of the user in the sequence. Second, sort the scheduling order of the tasks according to the above urgency. Third, deploy an integer programming solver to solve the xth task in the sequence. k User In particular, the xth k The task planning problem for each user can be given as

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141] Next, give a solution details.

[0142] Step S231: Initialize dynamic power supply and mission planning strategies

[0143] Step S232: For all x k ∈X, using an integer programming solver And get for the xth k Locally optimal computation allocation strategy for each user according to Dynamic energy consumption through

[0144] Step S233: Update dynamic power supply and mission planning strategies

[0145] Step S234: Repeat steps S232 to S233 until all users have obtained the task planning strategy, and output C ＊ .

[0146] Preferably, in step S2 Computer resource allocation is solved by the following steps:

[0147] Based on the solution The optimal task planning strategy C obtained by task scheduling ＊ , solved using CVX:

[0148]

[0149] stS≤S max , (22b)

[0150]

[0151] get Computer resource allocation.

[0152] Specifically, based on the solution The optimal task planning strategy C obtained by task scheduling ＊ , optimal CPU resource allocation F ＊ By solving To further reduce the average computing time for users:

[0153]

[0154] st (12e), (12g), (12h),

[0155] Its objective equation is a constant with respect to F, inspired by satisfying all constraints to increase f in a given time slot. k [n] to compute more task bits.

[0156] Therefore, for The sum of the ratios of computational data and task bits can be maximized, which will Convert to Right now:

[0157]

[0158] stS≤S max , (22b)

[0159]

[0160] The (12h) is further relaxed to allow more task bits to be computed using fewer time slots, so the average computation time for the user is further reduced. F is convex and is solved using the CVX method.

[0161] At this point, the optimal drone trajectory, optimal bandwidth, mission planning strategy, and computing resource allocation values ​​for this cycle are obtained.

[0162] The CVX method used in this application is operated according to the common steps in the prior art and will not be described here.

[0163] The abbreviations used in this application correspond to the full names in the table below.

[0164]

[0165] Example

[0166] In the following embodiments, the drone MEC network used consists of a fixed-wing drone, denoted as u, and a set of ground users, denoted as The total service time of the drone is recorded as T, which can be divided into |N| time slots, and the size of each time slot is and Since δ is small enough, the position of the UAV in a time slot remains roughly unchanged compared to the change in the distance from the user to the UAV.

[0167] Assume that the user is active in a specific time period (including a series of consecutive time slots) and loads the task to the drone. The time slot is recorded as T 1，k ={n k，s , ..., n k，e},in as well as is the index of the time slot used from the start to the completion of data transmission, t k，start and t k，end represent the start and end times of the transmission respectively. The task of the kth user can be expressed as Where D k represents the amount of input data to be processed (in bits), and θ k is the number of CPU cycles required to process one bit of input data, and is the delay required to complete the task, including transmission and computation delays, It is the moment when the processing is completed.

[0168] In addition, a 3D Cartesian coordinate system model is used to describe the positions of the drone and the user. Specifically, w k =(x k ,y k ,0) T is the location of the kth user, and represents the position of the drone in the nth time slot, where z = H 1 Therefore, the distance between the user and the drone in the nth time slot is In addition, let V max is the maximum speed of the drone, then the speed limit of the drone is

[0169]

[0170] For the sake of clarity, the binary quantity a is introduced k [n] describes the load decision of the kth user in the nth time slot. The task is assigned to the drone within the time, so the load decision is

[0171]

[0172] where a k [n] = 1 means that the user loads the task to the drone in the nth time slot. In addition, a set of time slots will be arranged to calculate the kth task on the drone, and another binary variable c k [n] is introduced to represent the computational planning strategy of the kth user in the nth time slot. More specifically, c k [n] = 1 means that the task of the kth user will be scheduled for calculation in the nth time slot, otherwise c k [n] = 0, and the task or the part that has not been executed and the context of the related process will be temporarily stored. Further assume that the drone is equipped with a single-core processor, and each time slot allows at most one task to be executed, that is, Note that due to the completeness of the mission, the drone can only calculate the time interval The tasks within

[0173]

[0174] A. Communication model

[0175] The UAV mobile edge computing system considered is based on FDMA (frequency division multiple access), and the bandwidth portion allocated by UVA to the kth user in the nth time slot is denoted as b k [n], which satisfies

[0176] Since the wireless channel between the UAV and the user is mainly controlled by the line-of-sight (LoS) link

[13] , the load rate of the kth user at the nth time slot can be written as

[0177]

[0178] Where B represents the total bandwidth, p k is the transmission power of the kth user, and N0 is the noise power spectral density at the drone. More importantly, is the channel gain, where β0 represents the information gain at the reference distance d0 = 1m. Due to the completeness of the task, the task of the kth user must be completed at the nth user. k，e time slots ago, which satisfies

[0179]

[0180] B. Computational Model

[0181] Note that the drone must complete the calculation process within the delay requirement corresponding to the task, denoted as

[0182]

[0183] Among them, f k [n] represents the CPU resources allocated to the kth user at the nth time slot. In addition, the allocated computing resources cannot exceed the maximum computing capacity of the drone, denoted as F max , which satisfies

[0184]

[0185] In this paper, the average computing time of users will be used as an indicator to measure user experience, expressed as

[0186]

[0187] C. Energy and Thermal Model of UAV

[0188] In practice, performing some heavy computing tasks, such as image processing and virtual reality, can cause the CPU to overheat. In addition, since it is difficult to install a powerful cooling device on a drone, the drone's heat dissipation capacity is relatively weak, which leads to extremely high temperatures and permanent damage to the CPU. In order to protect the processor, the CPU temperature page will also be considered in this article.

[0189] Energy model: Since a processor is either in idle or executing state, dynamic energy management is considered. Specifically, if a processor is executing a task, it is said to be active, otherwise it is said to be idle [4]. The total power consumption of a processor is the sum of dynamic power consumption and leakage power consumption, i.e.

[0190] P pro =P leak +P dyn , (9)

[0191] Among them, P leak is the leakage power consumption of the processor when waiting for computing tasks. For simplicity, P leak Treated as a constant.

[0192]

[0193] Thermal Model: Inspired by previous work, this paper uses a well-known thermal circuit model to convert the processor's energy consumption to its temperature. pro (t) and S amb (t) is the average power consumption and ambient temperature of the processor during a period of time t. Therefore, the temperature S(t) of the processor at the end time is

[0194]

[0195] Where R and C are the thermal resistance and thermal capacitance respectively, and T(0) is the initial temperature of the processor. It is observed that the temperature will increase / decrease and eventually reach S amb (t)+P pro (t)·R. The stable temperature of the CPU is defined as

[0196]

[0197] In order to meet the thermal constraints, the stable temperature should satisfy S≤S max , where S max is the maximum temperature of the processor.

[0198] D. Problem Modeling

[0199] Will is the trajectory of the drone, is the bandwidth allocated by the drone to the user, is the planning strategy for the computation, is the frequency of the drone assigned to the user. In this paper, the above variables are jointly optimized to minimize the average computation time of the user, which is formulated as

[0200]

[0201]

[0202]

[0203]

[0204] S≤S max , (12e)

[0205]

[0206]

[0207]

[0208]

[0209]

[0210] Among them, the stable temperature of the CPU (12e), computing resources (12g) and maximum speed (12j) are limited on the drone due to hardware limitations. Constraints (12i) and (12h) ensure that all tasks can be fully allocated and calculated to meet the corresponding latency requirements.

[0211] Formulated Problem is a mixed integer nonlinear programming problem. It is difficult to solve due to the strongly coupled binary variables and nonlinear constraints. In order to solve this problem, the BCD method is used to convert Decompose into the following sub-problems:

[0212] 1) Trajectory optimization,

[0213] 2) Communication resource allocation,

[0214] 3) Task scheduling,

[0215] 4) Computational resource allocation.

[0216] A. Trajectory Optimization

[0217] The solution is as follows:

[0218] Since users need to load tasks to drones within a given time, optimizing the drone trajectory to improve the communication channel between users and drones is crucial to increasing the data transmission rate. Therefore, the relevant problem is formulated as

[0219]

[0220] st (12i), (12j).

[0221] It is non-convex with respect to Q. To solve this problem, an SCA-based approach is used to transform the non-convex constraints into a suitable convex constraint approximation. In particular, the slack variables is introduced to solve the problem about r in (12i) k The non-convex part of [n], i.e.

[0222]

[0223] in Although r k [n] is either concave or convex with respect to q[n], but it is is convex. Because the first-order Taylor approximation of a convex function is a global underestimation. Therefore, given the trajectory at the i-th iteration In the case of k The lower bound of [n] is

[0224]

[0225] in and Therefore, the problem can be reformulated as:

[0226]

[0227]

[0228]

[0229]

[0230] It is convex and can be solved directly by CVX methods.

[0231] B. Communication resource optimization

[0232] The solution is as follows:

[0233] With the feasible trajectory Q obtained ＊ , feasible bandwidth allocation B ＊ By solving This is achieved by satisfying the load delay requirements of all users.

[0234]

[0235] st (12f), (12i),

[0236] It is convex with respect to B and can be solved directly by CVX methods.

[0237] C. Mission Planning Strategy

[0238] The solution is as follows:

[0239] With the newly acquired {Q ＊ , B ＊ And given F, the optimal task planning strategy C is obtained by making full use of computing resources to reduce the average time consumption of users ＊ , which is done by solving the following problem

[0240]

[0241] st(12b)-(12e), (12g), (12h).

[0242] It can be seen that is an integer programming about C, which is difficult to solve. Although the B&B method can solve this problem, it is difficult to obtain the optimal strategy in polynomial time as the number of users and time slots increase. In order to solve this problem, a low-complexity task scheduling method is proposed according to the urgency of the task, which is defined as

[0243]

[0244] First, according to the above urgency, the scheduling order of tasks is calculated in descending order, given as where x k yes The index of the user in the sequence. Second, sort the scheduling order of the tasks according to the above urgency. Third, deploy an integer programming solver to solve the xth task in the sequence. k User In particular, the xth k The task planning problem for each user can be given as

[0245]

[0246]

[0247]

[0248]

[0249]

[0250]

[0251] D. Computing resource allocation

[0252] The solution is as follows:

[0253] Step 1: Initialize dynamic power supply and mission planning strategies

[0254] Step 2: For all x k ∈X, using an integer programming solver And get for the xth k Locally optimal computation allocation strategy for each user according to Dynamic energy consumption through

[0255] Step 3: Update dynamic power supply and mission planning strategies

[0256] Step 4: Repeat steps 2 and 3 until all users have obtained the task planning strategy.

[0257] D.CPU resource optimization

[0258] Based on the newly obtained C ＊ , optimal CPU resource allocation F ＊ By solving To further shorten the user's average computing time.

[0259]

[0260] st (12e), (12g), (12h),

[0261] Its objective equation is a constant with respect to F. Inspired by

[15] that satisfies all constraints, f can be increased within a given time slot. k [n] to calculate more task bits. Therefore, for The sum of the ratios of computational data to task bits can be maximized, which will Convert to Right now

[0262]

[0263] stS≤S max , (22b)

[0264]

[0265] The (12h) is further relaxed to allow more task bits to be computed using fewer time slots, so the average computation time for the user is further reduced. F is convex and is solved using the CVX method.

[0266] In the UAV mobile edge computing network, we consider a 300m×300m area with K=20 ground nodes and N=90 time slots, where the UAV Fly to Unless otherwise specified, the remaining simulation parameters are shown in Table 1.

[0267] Table 1 Simulation parameters

[0268]

[0269]

[0270] The simulation results obtained by using the above method are as follows Figures 2 to 4 As shown, Figure 2 The figure shows the trajectory of the drone with different source points and end points. The horizontal axis is the X coordinate of the drone (unit: meter), and the vertical axis is the Y coordinate of the drone (unit: meter). Figure 2The trajectories of drones with different starting and destination points are plotted in the figure. It can be seen that a shorter distance between the user and the drone will provide a higher quality communication channel for the user, and flying the drone closer to the area where users gather will increase the offloading rate and ensure that the task is completely offloaded within a given time.

[0271] Figure 3 The figure is a comparison of the average computing time of different numbers of users in different schemes. The figure shows the average unloading rate of the method WTC proposed by the present invention and the equivalent frequency allocation method EFA, the method without thermal constraints WTTC and the solver method SDS. The horizontal axis is the number of users, and the vertical axis is the average computing time of users (unit: seconds). Figure 3 It can be seen that the average calculation time of the method provided by the present application increases with the increase of users, but the average calculation time is lower than that of the EFA method, but higher than that of the WTTC method. Although higher than the WTTC method, since the method does not take the CPU temperature into consideration, the accuracy of the calculation results cannot be effectively guaranteed.

[0272] exist Figure 3 In , we evaluated the impact of the number of users on the average computing time performance. The results show that the average computing time of a user increases with the number of users. This can be explained by the fact that the computing resources available to each user decrease as the number of users increases, resulting in higher computing time and a natural extension of the average computing time of the user. Compared with EFA, the average computing time saved by WTC increases from 3% to 62.5% because a more reasonable task scheduling strategy can be obtained, thereby reducing the average computing time of the user. In addition, WTTC obtains a shorter average computing time for users compared with WTC and EFA. This is because all computing resources will be allocated without temperature constraints. It should be noted that when the number of users is relatively small, that is, N≤8, SDS can provide more scheduling strategies and leads to a long time to obtain a solution, which impairs its practicality in large-scale user situations and cannot obtain acceptable results within an acceptable time.

[0273] Figure 4 The temperature and average computing time of different CPU frequencies in different schemes are shown in the figure. The average user computing time and CPU temperature of the method WTC proposed by the invention, the equal frequency allocation method EFA, and the method WTTC without thermal constraints are shown in the figure. The horizontal axis is the CPU frequency (unit: GHz), the average user computing time (unit: seconds), and the right vertical axis is the CPU temperature (unit: Celsius). Figure 4 It can be seen that the CPU temperature of the method provided in the present application is lower than that of all the compared methods, indicating that the method provided in the present application can effectively take into account the average calculation time and CPU temperature, avoid the decrease in calculation frequency due to excessively high CPU temperature, and obtain a more reasonable average calculation time.

[0274] Figure 4 The trade-off between average computing time and CPU temperature is highlighted. Note that the average computing time of users will decrease with the increase of CPU frequency. This is because when the computing power of the UAV is enhanced, the UAV can allocate more CPU resources to users, reduce computing delays, and save users’ average computing time. In addition, compared with the significant impact of WTTC on users’ average computing time, the increase of UAV CPU frequency does not significantly reduce users’ average computing time for WTC and ETA. In WTC (or ETA), the CPU temperature will be controlled and remain constant near the peak value (60℃) to protect the computing platform, although it will sacrifice up to 66% (or 77%) of users’ average computing time compared with WTTC. In addition, in WTTC, the change of average computing time is more obvious as the frequency increases from 1.1GHz to 1.7GHz, because when F = 1.1GHz, each user needs three to four time slots to compute the task, while when F = 1.4GHz, most users only need two time slots. Therefore, users’ average computing time will decrease rapidly. However, in WTTC, as the frequency increases from 1.7 GHz to 2.6 GHz, all users still need two time slots to compute the task, so it changes slowly. In ETA, it can be seen that as the CPU frequency increases, the average computing time of the user and the CPU temperature remain constant. In this regard, the allocable F can be calculated as F = 1 GHz to satisfy constraint (12e), where the allocated CPU frequencies are equal, resulting in a constant average computing time for the user.

[0275] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for minimizing the computing time of integrated tasks in a MEC network under thermal sensitive conditions, characterized in that: The following steps are involved: Step S1: Initialization phase: In this phase, the nodes in the network obtain the basic configuration information of the network and initialize the relevant parameters, including: local computing resources and computing power, initial values ​​of UAV flight trajectory, initial values ​​of user bandwidth allocation, initial values ​​of mission planning strategy and initial values ​​of computing resource allocation variables; Step S2: Establishing a system optimization model: According to the overall goal of minimizing the average computing time of the user and the constraints, an average computing time minimum system optimization model is established. The constraints include: latency, CPU temperature, maximum computing resources and processing speed. According to the constraints, the relevant parameters are substituted into the average computing time minimum system optimization model to obtain the initial optimal UAV trajectory. Based on the initial optimal UAV trajectory, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K and the computing resource allocation are calculated; Average computing time minimum system optimization model: S≤S max , (12e) c k [n] represents the computing planning strategy of the kth user in the nth time slot, represents the position of the drone in the nth time slot, where z = H 1 , a k [n] = 1 means that the user offloads the task to the drone in the nth time slot, b k [n] represents the bandwidth portion allocated by UVA to the kth user in the nth time slot, f k [n] represents the CPU resources allocated to the kth user at the nth time slot, r k [n] represents the unloading rate of the kth user at the nth time slot, F max represents the maximum computing capacity of the UAV, S max is the maximum temperature of the processor, where Using the BCD method Decoupling is: 1) Trajectory optimization; 2) Communication resource allocation; 3) Task scheduling; 4) Computational resource allocation; For k users Decouple and calculate the optimal UAV trajectory corresponding to user k, the optimal bandwidth allocated to user K, the task planning strategy allocated to user K, and the computing resource allocation; Step S3: Iterative loop calculation: While satisfying the constraints, the optimal UAV trajectory, optimal bandwidth, mission planning strategy and computing resources obtained in the current round are used to enter the next round of iterative loop calculation according to the block coordinate descent method; Step S4: When the result of the cyclic calculation reaches the set target accuracy, the optimal drone trajectory, optimal bandwidth, task planning strategy and computing resource allocation for different computing tasks of each user node are obtained.

2. The method according to claim 1, characterized in that In step S2 Trajectory Optimization The solution is done by CVX method including the following steps: Get the best feasible trajectory Q for user k ★ .

3. The method according to claim 1, characterized in that Communication resource allocation in step S2 Solve it in the following steps: Based on the feasible trajectory Q obtained by solving the optimal trajectory ★ , feasible bandwidth allocation B ★ Solved using CVX method To get the uninstall delay that meets all users' needs: st(12f),(12i),.

4. The method according to claim 1, characterized in that: In step S2 The task scheduling is solved by the following steps: The feasible trajectory Q is obtained based on the optimal trajectory obtained by solving ★ , feasible bandwidth allocation B ★ And given F, in order to meet the requirement of making full use of computing resources to reduce the average time consumption of users, the optimal task planning strategy C is obtained ★ .

5. The method according to claim 4, characterized in that In step S2 Task scheduling solution includes the following steps: Step S21: According to the urgency: Calculate the scheduling order of each task in descending order, given as Among them, x k yes The index of the user in; Step S22: sorting the task scheduling order according to the above urgency; Step S23: Use an integer programming solver to solve the xth task in the task sequence k User Task Scheduling No. x k The task planning problem for each user is:

6. The method according to claim 1, characterized in that In step S2 Computer resource allocation is solved in the following steps based on the solution The optimal task planning strategy C obtained by task scheduling ★ , solved using CVX: stS≤S max , (22b) get Computer resource allocation.

Images