A robust task offloading method for drone-assisted mobile edge computing

Abstract

This paper proposes a robust task offloading method for drone-assisted mobile edge computing. It establishes a scenario model for a drone-assisted mobile edge computing system and constructs a problem model for minimizing the maximum task completion time of a drone-assisted mobile edge computing system while satisfying energy consumption constraints and time slot restrictions. The problem model is decomposed, non-convex constraints are approximated using a continuous convex approximation method, and the solution is solved through alternating iterations in a block coordinate descent framework. This method not only fully exploits the potential of partial drone offloading but also effectively reduces system latency and improves overall energy efficiency under realistic NLOS channel conditions.

Description

Technical Field

[0001] The present invention relates to the combination of information and communication engineering technology and Internet of Things technology, and specifically provides a robust task offloading method for drone-assisted mobile edge computing. Background Art

[0002] The development of sixth-generation (6G) mobile network technology will usher in new mobile devices such as augmented reality (AR), virtual reality (VR), and intelligent navigation. These applications place extremely high demands on computing resources and latency. However, ground-based user devices (such as smartphones and IoT devices) are constrained by limited computing power and battery life, making it difficult for them to independently complete complex computing tasks. Drone-assisted mobile edge computing (MEC) systems, which utilize drones as airborne mobile edge servers to provide flexible computing offload services to ground-based devices, are proving to be an effective solution to this problem.

[0003] Numerous studies have focused on optimizing drone-assisted MEC systems, primarily by jointly optimizing drone trajectories, task offloading decisions, and resource allocation to reduce overall task completion latency and energy consumption. For example, in the context of a surge in the number of ground devices or sparse infrastructure, some studies minimize the latency between all devices and drones; others minimize system energy consumption while satisfying task latency and drone dynamic constraints. However, existing research suffers from the following problems: many studies assume that task offloading is binary, ignoring the potential for partial offloading; and existing algorithms, to reduce device energy consumption, force drones to frequently adjust their trajectories, resulting in a decrease in overall system energy efficiency. Summary of the Invention

[0004] This paper proposes a robust task offloading method for UAV-assisted mobile edge computing.

[0005] The technical solution for achieving the purpose of the present invention is: a robust task offloading method for drone-assisted mobile edge computing, comprising:

[0006] Establish a scenario model for a drone-assisted mobile edge computing system and, while meeting energy consumption constraints and time slot restrictions, construct a problem model for minimizing the maximum task completion time of the drone-assisted mobile edge computing system.

[0007] The problem of minimizing the maximum task completion time of a UAV-assisted mobile edge computing system is decomposed into three sub-problems: the UAV trajectory optimization sub-problem, the ground equipment transmission power and data offload ratio optimization sub-problem, the binary offload decision sub-problem and the UAV computing power optimization sub-problem.

[0008] The three sub-problems are solved sequentially and iteratively, and the iterations are repeated until the convergence conditions are met and the robust task offloading is completed. The continuous convex approximation method is used to solve each sub-problem.

[0009] Compared with the prior art, the present invention has the following significant advantages:

[0010] This paper breaks the limitation of the binary assumption of task offloading in traditional research, explores the potential of partial offloading of UAVs, and optimizes coupling variables such as the task offloading rate, enabling the system to flexibly allocate computing tasks according to actual conditions, thereby improving the efficiency and rationality of task processing.

[0011] To address the issue of frequent drone trajectory adjustments leading to a decrease in overall system energy efficiency, this invention innovatively constructs a three-dimensional joint optimization model by fully considering the strong coupling relationship between trajectory, offloading, power, and computing power. By nesting the block coordinate descent method (BCD) and the successive convex approximation method (SCA), and jointly optimizing parameters such as drone trajectory, computing power, and device transmission power, it avoids the waste of drone energy consumption caused by excessive pursuit of device energy saving. While ensuring reasonable device energy consumption, it also significantly improves the overall energy efficiency of the system.

[0012] Research based on NLOS channels is more in line with actual deployment scenarios. Compared with existing methods based on ideal LOS channels, it is more in line with communication conditions in real environments, effectively reduces system latency, and enhances the practicality and reliability of the system.

[0013] The present invention will be described in further detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a model system diagram of a robust task offloading method for drone-assisted mobile edge computing described in one embodiment of the present invention.

[0015] Figure 2 This is an algorithm flow chart of a robust task offloading method for drone-assisted mobile edge computing according to an embodiment of the present invention.

[0016] Figure 3 This is a diagram showing the algorithm convergence effect of a robust task offloading method for drone-assisted mobile edge computing described in one embodiment of the present invention.

[0017] Figure 4 This is a comparison chart of total system energy consumption corresponding to different numbers of ground devices in a robust task offloading method for drone-assisted mobile edge computing described in one embodiment of the present invention. DETAILED DESCRIPTION

[0018] A robust task offloading method for drone-assisted mobile edge computing is designed for scenarios where ground user devices have limited computing power and battery budget, and where existing drone-assisted MEC system research has many limitations. Taking into account energy constraints and time slot restrictions, this study investigates how to minimize maximum system latency and improve overall energy efficiency by jointly optimizing drone trajectories, task offloading rates, drone computing power, device transmit power, and binary offloading decisions. The method primarily decomposes the problem, utilizes continuous convex approximation (SCA) to approximate non-convex constraints, and combines this with a block coordinate descent (BCD) framework for an iterative solution. This method not only fully exploits the potential of drone partial offloading but also effectively reduces system latency and improves overall energy efficiency under realistic NLOS channel conditions.

[0019] (1) Establish a UAV-assisted MEC system model based on NLOS channel, analyze and derive the objective function of minimizing the maximum delay of the system under the conditions of meeting energy consumption constraints and time slot restrictions; (2) Decompose the original problem into three sub-problems, namely, optimizing the UAV trajectory , Ground equipment transmission power and the proportion of data offloaded to drones Optimization, binary uninstallation decision and drone computing power Optimize the three sub-problems; (3) Sequential iterative optimization is repeated until the convergence conditions are met and the final optimization result is obtained. The specific implementation process is as follows:

[0020] Step 1: Establish a UAV-assisted mobile edge computing (MEC) system scenario model, namely the channel model, local computing model, and UAV-assisted edge computing model, and determine the problem model that minimizes the maximum task completion time of the UAV-assisted mobile edge computing (MEC) system under the conditions of meeting energy consumption constraints and time slot restrictions.

[0021] like Figure 1 The drone-assisted mobile edge computing (MEC) system model shown in the figure consists of a drone integrated with an edge server as an aerial base station and terrestrial users, denoted as The drone is able to communicate with the ground mobile device and provide computing services at the same time. Each user can choose whether to offload part of its computing tasks to the drone and execute the rest locally. Assume that the drone is at a fixed altitude. flight, is the length of equidistant time slots, , the horizontal coordinates of the UAV are , the ground user horizontal coordinate is The maximum speed of the drone is , then the drone trajectory constraints can be as follows:

[0022] ;

[0023] ;

[0024] Among them, the constraints Indicates that the drone's flight speed cannot exceed its maximum speed, constraint Indicates that the drone is in a cycle At the end, it returns to its initial position. The energy consumption caused by the UAV flight is related to its speed. The UAV flight energy consumption model can be expressed as:

[0025]

[0026] in Related to the payload of the drone.

[0027] Channel model:

[0028] The device With drones in The distance between time slots is expressed as:

[0029]

[0030] equipment The average channel gain with the drone is expressed as:

[0031]

[0032]

[0033] in Indicates that under NLoS link, the device With drones in The probability of a LoS channel appearing at a time slot is, It is a device With drones in The probability of an NLoS channel appearing at a time slot, and It is a parameter determined by the external environment. is the elevation angle in degrees.

[0034] Assume that the maximum transmit power of the device is , the power constraint is as follows:

[0035]

[0036] equipment The information rate between the UAV and the GNSS can be expressed as:

[0037]

[0038] in is the total channel bandwidth in Hertz (Hz), represents the noise power spectral density, and due to the actual modulation and coding scheme, there is a gap in channel capacity, which is expressed as express.

[0039] The above distance formula, the channel gain between the device and the drone, the transmission power, and the information rate constitute the channel model of the present invention.

[0040] Local computing model:

[0041] use Indicates that the mobile device The tasks to be performed. is the proportion of data offloaded to the drone, Discrete binary user scheduling variables, No. The amount of computational tasks in each time slot, is the number of CPU cycles required for each user to execute each bit. ,equipment The local computation delay and local energy consumption are as follows:

[0042]

[0043]

[0044] in is a user Maximum computing power (cycles / second), is a constant that depends on the chip architecture of the mobile device.

[0045] The local computing delay and local computing energy consumption of the above-mentioned devices constitute the local computing model of the present invention.

[0046] Drone-assisted edge computing model:

[0047] Based on the above model, in the time slot Medium users The latency and energy consumption of task offloading are as follows:

[0048]

[0049]

[0050] After the task is unloaded, the time and energy consumption of the UAV to perform the task are:

[0051]

[0052]

[0053] in Depends on the chip architecture of the drone. is the maximum computing energy of the UAV, and the computing capacity constraint, time constraint, and energy constraint are as follows:

[0054]

[0055]

[0056]

[0057]

[0058]

[0059] The offloading computational delay and computational energy consumption of the above-mentioned devices, as well as the time and energy consumption of the UAV performing the task after the task is offloaded, constitute the UAV-assisted computational offloading model of the present invention.

[0060] Problem description:

[0061] Since the computation results and task uploads are relatively small, and the transmission bandwidth allocated to the uplink is much larger than that of the downlink, the downlink transmission of the computation results to the ground equipment is ignored. Based on the above model, the following problem can be formulated:

[0062] ;

[0063] Among them, the objective function It is the minimum of the maximum of local processing time and task completion time.

[0064] Step 2: Based on the alternating optimization algorithm, the original problem is decomposed into three sub-problems.

[0065] The algorithm decomposes the original problem into three sub-problems: using the successive convex approximation method (SCA) to transform the non-convex constraints into convex constraints, and combining the block coordinate descent method (BCD) to decompose the original problem into three sub-problems, namely optimizing the UAV trajectory , Ground equipment transmission power and the proportion of data offloaded to drones Optimization, binary uninstallation decision and drone computing power Optimize three sub-problems.

[0066] Step 3: Solve the three sub-problems (steps 3.1-3.3) sequentially and iteratively, and repeat the iterative solution process until the convergence conditions are met to obtain the final optimization result.

[0067] This paper proposes an optimization algorithm for establishing a drone-assisted MEC system model based on NLOS channels, and analyzes and derives the objective function of minimizing the maximum delay of the system under the conditions of meeting energy consumption constraints and time slot restrictions. The algorithm decomposes the original problem into three sub-problems: drone trajectory , Ground equipment transmission power and the proportion of data offloaded to drones Optimization, binary uninstallation decision and drone computing power .

[0068] First, initialize the maximum number of time slots for each variable , the maximum number of ground users , the proportion of data offloaded to drones , binary user scheduling variables , No. The amount of computing tasks in each time slot , the number of CPU cycles required per user to execute each bit , maximum transmit power , the maximum flight speed of the drone , the current number of iterations ;

[0069] 1) Given The transmit power at the iteration , the proportion of data offloaded to drones , binary user scheduling variables , UAV computing power , optimize the solution trajectory ;

[0070] 2) Using the first The trajectory at the iteration , given the Binary user scheduling variable at iteration , UAV computing power , optimize and solve the transmission power of ground equipment and the proportion of data offloaded to drones ;

[0071] 3) Using the first The trajectory at the iteration , transmit power and the proportion of data offloaded to drones , optimize the solution of discrete binary user scheduling variables and drone computing power ;

[0072] 4) Determine whether the convergence conditions are met: If not satisfied, , return 1), otherwise the current result As the final optimization result, is the convergence threshold.

[0073] In a further embodiment, given other variables , optimize the solution trajectory The specific method is:

[0074] Original question Can be converted into :

[0075]

[0076] constraint and is a non-convex constraint and contains multiple variables (such as and ). To solve this problem, By considering the most likely elevation angle, the uniform approximation is fixed to Then, the device The information rate between the UAV and the GNSS can be approximately expressed as:

[0077]

[0078] in , . Introducing auxiliary variables , Set to 2, so , this time it is non-concave, and the following transformation is performed . Then Performing Taylor expansion yields:

[0079]

[0080] in . The problem at this time It can be restated as As shown below:

[0081]

[0082] at this time, The objective function and constraints are all convex and can be solved using the relevant CVX solvers.

[0083] In a further embodiment, the above steps are used to obtain , and given other variables , optimize the transmission power of ground equipment and the proportion of data offloaded to drones The specific method is:

[0084] Optimized the transmission power of ground equipment and the proportion of data offloaded to drones Other variables are given below At this point, the original question Can be converted into .

[0085]

[0086] Same as subproblem 1, introducing auxiliary variables , making ,Right now , and because . The problem at this time It can be restated as As shown below:

[0087]

[0088] in is jointly convex. At this time, The objective function and constraints are all convex and can be solved using the relevant CVX solvers.

[0089] In a further embodiment, the above two steps are used to obtain , optimize discrete binary user scheduling variables and drone computing power The specific method is:

[0090] Scheduling variables for discrete binary devices and drone computing power Optimize, given other variables At this point, the original question Can be converted into .

[0091]

[0092] Introducing continuous variables , that is Continuous variables, that is, introduce constraints:

[0093]

[0094] in ,Right now , at this time the time constraint can be (replace the other constraints Replace with ;

[0095]

[0096] Defining auxiliary variables , , then the energy constraint can be expressed as:

[0097]

[0098] The problem at this time It can be restated as As shown below:

[0099]

[0100] at this time, The objective function and constraints are all convex and are solved using the CVX solver.

[0101] In a further embodiment, it is determined whether the convergence condition is met: If not satisfied, , return 1), otherwise the current result As the final optimization result, is the convergence threshold.

[0102] Result description:

[0103] Assume that the distribution is random and uniform in an area of 500m x 500m. Mobile device, drone at fixed height , the flight speed does not exceed 50m / s, and the elevation angle is fixed at 90 o , path loss index Set to 2.5, the user's maximum transmit power =1W, total channel bandwidth =1MHZ, the maximum computing power of a mobile device Randomly distributed within [300,400] MHZ, the maximum computing power of the drone =1200 MHZ, noise power spectral density =-100dBm, UAV payload =6kg, at reference distance At this point, the channel power gain is set to =-40dB, task size follow (bits) are uniformly distributed, let cycles / bit, , the battery energy of the device =0.01KWh, the battery energy of the drone =0.01KWh.

[0104] The performance of the proposed joint optimization strategy (Algorithm 1) is verified through simulation experiments. Three benchmark schemes are selected for comparison: an algorithm without device transmit power and drone computing energy optimization (Algorithm 2), a genetic algorithm (Algorithm 3), and a full offloading (Algorithm 4).

[0105] Figure 3 This shows that our proposed solution gradually converges to a suboptimal solution after several iterations. A closer look at the figure reveals that compared to Algorithm 3, our proposed solution converges faster, meaning that under the same conditions, it can more efficiently approach the ideal result, saving significant time and computational cost for practical applications.

[0106] Figure 4 The total energy consumption of the drone is shown, including flight energy consumption, edge computing energy consumption, and device energy consumption. For example, the four algorithms for uploading energy consumption and local computing energy consumption are compared. It can be clearly seen that the drone energy consumption is lower than that of other methods under this method, and the device energy consumption is lower than that of the other two algorithms except Algorithm 4. However, the total energy consumption of the drone and device is much lower than that of Algorithm 4. This is because Algorithm 4 chooses to offload all tasks to the drone for execution, and the device only consumes energy for uploading.

[0107] This paper addresses the challenges of multivariable coupled optimization in drone-assisted mobile edge computing systems by proposing a joint optimization method based on continuous convex approximation (SCA) and block coordinate descent (BCD). This joint optimization method balances energy efficiency with real-time performance, providing theoretical support and technical reference for IoT scenarios with high user density and dynamic task demands in drone-assisted edge computing systems.

Claims

1. A robust task offloading method for drone-assisted mobile edge computing, characterized in that: include: A scenario model of a drone-assisted mobile edge computing system is established. Under the conditions of meeting energy consumption constraints and time slot restrictions, a problem model for minimizing the maximum task completion time of the drone-assisted mobile edge computing system is constructed. Specifically: ； in, ; Where, is the time slot length, is the maximum computing power of the drone, The computing power of drones, is the battery energy of the device, It is the battery energy of the drone, is the maximum tolerable time for task offloading and computing task completion, is the maximum tolerable time for local computation, is the maximum flight speed of the drone, It is the trajectory of the drone. Maximum transmit power of ground equipment, The transmission power of the ground equipment, The proportion of data offloaded to the drone, Discrete binary user scheduling variables, and are the time and energy consumption of equipment unloading, and are the time and energy consumption of local computing of the device, and are the time and energy consumption of the drone in processing the unloading task, The energy consumed by the drone flight, , , , is the set of ground user numbers, is the maximum number of devices, is the set of time slot numbers, Maximum number of time slots; The problem of minimizing the maximum task completion time of a UAV-assisted mobile edge computing system is decomposed into three sub-problems: the UAV trajectory optimization sub-problem, the ground equipment transmission power and data offload ratio optimization sub-problem, the binary offload decision sub-problem and the UAV computing power optimization sub-problem. The three sub-problems are solved sequentially and iteratively, and the iterations are repeated until the convergence conditions are met and the robust task offloading is completed. The continuous convex approximation method is used to solve each sub-problem.

2. The robust task offloading method for drone-assisted mobile edge computing according to claim 1 is characterized in that: The UAV-assisted mobile edge computing system includes a UAV integrated with an edge server and terrestrial users, denoted as ,The drone acts as an air base station, and is able to simultaneously communicate with ,ground mobile devices and provide computing services. Each user chooses whether to offload a portion of the ,computing tasks to the drone and execute the rest locally.

3. The robust task offloading method for drone-assisted mobile edge computing according to claim 1 is characterized in that: The established drone-assisted mobile edge computing system scenario model includes a channel model, a local computing model and a drone-assisted edge computing model. The channel model includes the channel gain, transmission power and information rate between the device and the drone; the local computing model includes the local computing delay and the energy consumption of the local computing; the drone-assisted edge computing model includes the unloading computing delay and the computing energy consumption, as well as the time and energy consumption of the drone to perform the task after the task is unloaded.

4. The robust task offloading method for drone-assisted mobile edge computing according to claim 1, characterized in that: The specific process of sequentially iteratively solving the three sub-problems is: Initialize the maximum number of time slots for each variable , the maximum number of ground users , the proportion of data offloaded to drones , binary user scheduling variables , No. The amount of computing tasks in each time slot , the number of CPU cycles required per user to execute each bit , maximum transmit power , the maximum flight speed of the drone , the current number of iterations ; 1) Given The transmit power at the iteration , the proportion of data offloaded to drones , binary user scheduling variables , UAV computing power , optimize the solution trajectory ; 2) Using the obtained The trajectory at the iteration , given the Binary user scheduling variable at iteration , UAV computing power , optimize and solve the transmission power of ground equipment and the proportion of data offloaded to drones ; 3) Using the obtained The trajectory at the iteration , transmit power and the proportion of data offloaded to drones , optimize the solution of discrete binary user scheduling variables and drone computing power ; 4) Determine whether the convergence conditions are met: If not satisfied, , return 1), otherwise the current result As the final optimization result, is the convergence threshold.

5. The robust task offloading method for drone-assisted mobile edge computing according to claim 4 is characterized in that: Given variable The value of , optimize the solution trajectory The specific process is: Given variable , the problem Convert to , specifically: ； The device With drones in The probability of a LoS channel appearing at a time slot The uniform approximation is fixed to ,equipment The information rate between the drone and the ； in , ; Introducing auxiliary variables ,Will Set to 2.5, so , this time it is non-concave, and the following transformation is performed ; Again Performing Taylor expansion yields: ； in ; question Re-expressed as , specifically: ； at this time, The objective function and constraints are all convex and are solved using the CVX solver.

6. The robust task offloading method for drone-assisted mobile edge computing according to claim 4 is characterized in that: The first The trajectory at the iteration , and given the variable value, optimize the transmission power of ground equipment and the proportion of data offloaded to drones The specific method is: Optimized the transmission power of ground equipment and the proportion of data offloaded to drones Other variables are given below ; At this point, the original question Convert to : ； Introducing auxiliary variables , making ,Right now , and because ;The problem at this time Re-expressed as , specifically: ； in is jointly convex, then, The objective function and constraints are all convex and are solved using the CVX solver.

7. The robust task offloading method for drone-assisted mobile edge computing according to claim 4, characterized in that: Use what you get , optimize discrete binary user scheduling variables and drone computing power The specific method is: Scheduling variables for discrete binary devices and drone computing power Optimize, given other variables In this case, the original question Convert to , specifically: ； Introducing continuous variables , that is continuous variables, that is, introducing constraints ； in ,Right now , at this time the time constraint is: ； Defining auxiliary variables , , then the energy constraint is expressed as: ； The problem at this time Re-expressed as , specifically: ； at this time, The objective function and constraints are all convex and are solved using the CVX solver.

Images