A method, system, device and medium for offloading edge computing tasks from unmanned aerial vehicles (UAVs).

Abstract

This invention belongs to the field of mobile edge computing and discloses a method, system, device, and medium for offloading edge computing tasks from unmanned aerial vehicles (UAVs). The method includes: constructing a UAV edge computing model; determining multiple decision variables for the inspection task to be performed by the UAV; establishing a communication model and a computation and energy consumption model based on the decision variables and the UAV edge computing model; constructing a multi-objective optimization model based on the communication model and the computation and energy consumption model; and iteratively solving the multi-objective optimization model using a potential-driven multi-learning particle swarm optimization algorithm to output the optimal UAV offloading scheme as the global optimal solution. This invention effectively reduces the overall energy consumption of UAV swarms during task execution while meeting the real-time requirements of the task.

Description

Technical Field

[0001] This invention belongs to the field of mobile edge computing and relates to a method, system, device and medium for unloading edge computing tasks from unmanned aerial vehicles. Background Technology

[0002] Unmanned aerial vehicles (UAVs), with their advantages of maneuverability, flexibility, and wide field of view, play a vital role in fields such as power line inspection, agricultural monitoring, and environmental monitoring. Particularly in power line inspection, UAVs can be equipped with high-definition cameras to achieve efficient inspections of critical facilities such as transmission lines and substations, significantly improving the level of automation and operational safety. However, as the complexity of inspection tasks increases, UAVs often face the challenge of limited onboard computing resources when performing computationally intensive tasks such as real-time image recognition and defect detection. Simultaneously, such tasks also lead to rapid battery drain. If these tasks are performed locally, not only will there be significant processing latency, but the power consumption of the UAV will also increase rapidly, thus limiting its continuous operation capability.

[0003] Mobile edge computing (MEC) technology deploys computing resources at the network edge, such as base stations or substations, close to the drone's operating area, providing low-latency and low-energy near-end computing support for drones. This effectively avoids the high latency problem caused by long-distance backhaul in traditional cloud computing models, providing a feasible path to solve the aforementioned contradictions. However, the total resources of mobile edge computing servers are limited. How to intelligently formulate offloading strategies for multiple drones operating in parallel, and collaboratively optimize multiple variables such as offloading decisions, resource allocation, and power control, while meeting the real-time requirements of the task, minimizing system energy consumption, and balancing server load, has become crucial for improving the overall performance of drone edge computing systems. This problem can be formulated as a joint optimization problem involving multiple decision variables and constraints, and its solution process has high computational complexity. Therefore, researching efficient and intelligent offloading decision-making methods has significant theoretical and practical application value.

[0004] Scholars have extensively studied the task offloading problem in edge computing environments, and existing work mostly employs intelligent optimization algorithms for solving it, such as genetic algorithms, gray wolf optimization (GWO), and standard particle swarm optimization (PSO). Among these, particle swarm optimization has attracted much attention due to its simple principle, few parameters, and fast convergence speed. However, these existing methods, especially standard particle swarm optimization, still have significant shortcomings when dealing with complex optimization problems involving high dimensions, mixed variables, and multiple strict constraints: first, they are prone to getting trapped in local optima, and their global search capability is limited; second, their ability to handle complex constraints is weak, making it difficult to effectively guide the search process towards the feasible solution region, resulting in low solution efficiency, and the final offloading scheme still has considerable room for improvement in key performance indicators such as total system energy consumption, task latency, and load balancing. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system, device and medium for unloading edge computing tasks of UAVs, which effectively reduces the overall energy consumption of UAV swarms during task execution while meeting the real-time requirements of the tasks.

[0006] To achieve the above objectives, the present invention employs the following technical solution: A method for offloading edge computing tasks from a drone includes the following steps: Constructing an edge computing model for drones; For the inspection tasks to be performed by the drone, multiple decision variables are identified, including unloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation. Establish communication models and computing and energy consumption models based on decision variables and UAV edge computing models; A multi-objective optimization model is constructed based on the communication model and the computing and energy consumption model. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computing and communication energy consumption, and the server load balancing index. A potential-driven multi-learning particle swarm optimization algorithm is used to iteratively solve the multi-objective optimization model, and the optimal unloading scheme of the UAV is output as the global optimal solution.

[0007] Optionally, establish a communication model, including: The uplink transmission rate is determined based on transmit power, air-to-ground channel gain, and co-channel interference. The uplink transmission delay is determined by combining the uplink transmission rate, offload decision, base station selection decision, and task data volume. The backhaul link delay is determined based on base station selection decision, the connection relationship between the base station and the mobile edge computing server, server selection decision, task data volume, and the fixed transmission rate of the backhaul link. The total communication delay is calculated by adding uplink transmission delay, backhaul link delay, and base station processing delay, and a communication model is constructed based on the total communication delay.

[0008] Optionally, the computation and energy consumption model includes the total edge computing latency. The calculation process for the total edge computing latency includes: The local computing latency of the drone is determined based on the task's computational load and the drone's local processor clock frequency. Queuing latency and server computing latency are determined based on server computing resource allocation, task computing volume, average queuing wait time of priority tasks, and service rate. The total edge computing latency is calculated by adding the total communication latency, queuing latency, and server computing latency together.

[0009] Optionally, the calculation and energy consumption model includes energy consumption, and the energy consumption calculation process includes: The local computing power consumption is determined based on the local processor clock frequency, chip power consumption coefficient, and task computation load of the drone. The communication energy consumption is determined by combining base station selection decisions, transmission power, and uplink transmission delay. Hovering energy consumption is determined based on drone hovering power, unloading decision, drone local computing latency, and uplink transmission latency. The flight maneuver energy consumption is determined based on air density, total rotor area, drag coefficient, UAV cruise speed, and Euclidean distance between inspection points. Battery power constraints are established based on local computing energy consumption, communication energy consumption, hovering energy consumption, flight maneuvering energy consumption, and unloading decisions, and are used as constraints for the multi-objective optimization model.

[0010] Optionally, the multi-objective optimization model includes constraints, which include: Offloading-Base Station Selection Logic Constraints: When the inspection task is executed locally, no base station is allocated; when the inspection task is executed offloading, only one base station is allocated. Network topology validity constraint: When the inspection task is transmitted through the base station and the base station is directly connected to the mobile edge computing server, the inspection task will be assigned to the mobile edge computing server for execution. Base station wireless resource capacity constraint: The total number of inspection tasks transmitted through the base station shall not exceed the base station's wireless resource limit. Server computing resource capacity constraints: The sum of the computing frequencies allocated to mobile edge computing servers shall not exceed the upper limit of the computing frequency of mobile edge computing servers; Task delay constraint: The total delay of the inspection task shall not exceed the maximum tolerable delay; Drone battery safety constraints: The total energy consumption of the drone in performing all inspection tasks shall not exceed the product of the total capacity of the drone battery and the safety factor; Transmission power constraint: The transmission power of the inspection mission shall not exceed the maximum transmission power of the UAV, and the sum of the transmission power of all inspection missions shall not exceed the total power budget; Domain constraints for variables: Base station selection decisions and server selection decisions are integer variables, while server computing resource allocation is a non-negative continuous variable.

[0011] Optionally, a potential-driven multi-learning particle swarm optimization algorithm is used to iteratively solve the multi-objective optimization model, outputting the optimal unloading scheme for the UAV as the global optimal solution, including: Randomly generate an initial particle population and set algorithm parameters; The decision variables are divided into binary variables and continuous variables, and mixed encoding is performed to obtain the particle position vector. The binary variables include unloading decision, base station selection decision and server selection decision, while the continuous variables include transmission power and server computing resource allocation. The particle position vector is decoded into a decision variable, and the fitness value of the particle is calculated by combining the objective function of the multi-objective optimization model and the dynamic penalty coefficient. Update the individual particle's historical best position and the global historical best position based on the fitness value; The particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position. The iteration stops when the maximum number of iterations is reached, and the global historical best position is output as the global optimal solution. The global optimal solution is then decoded to obtain the optimal unloading scheme for the UAV.

[0012] Optionally, the particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position, including: A random elite guidance strategy is obtained, and the particle velocity is updated by combining the individual particle's historical best position, the global historical best position, and the guidance position randomly selected from the current elite particles. The elite particles refer to the particles with the top 20% fitness values. Update the particle position vector based on the updated particle velocity; Calculate the constraint violation degree of the updated particle position vector corresponding to the constraint condition, and calculate the mutation probability based on the constraint violation degree; Get a random number. If the random number is less than the mutation probability, apply a Gaussian perturbation to the continuous variable of the selected particle position vector and perform a bit flip operation on the binary variable of the selected particle position vector. Return to the step that calculates the fitness value of the particle, until the maximum number of iterations is reached.

[0013] A drone edge computing task offloading system includes: The UAV edge computing model building module is used to build UAV edge computing models. The decision variable determination module is used to determine multiple decision variables for the inspection tasks to be performed by the UAV. The decision variables include unloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation. The communication model and computation and energy consumption model building module is used to build communication models and computation and energy consumption models based on decision variables and UAV edge computing models; The multi-objective optimization model construction module is used to build a multi-objective optimization model based on the communication model and the computing and energy consumption model. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computing and communication energy consumption, and the server load balancing index. The optimal unloading scheme output module is used to iteratively solve the multi-objective optimization model using a potential-driven multi-learning particle swarm optimization algorithm, and output the optimal unloading scheme of the UAV as the global optimal solution.

[0014] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the UAV edge computing task offloading method.

[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the UAV edge computing task offloading method.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a drone edge computing model encompassing drones, base stations, and servers, and jointly models multi-dimensional heterogeneous variables such as offloading decisions, resource allocation, and power control, thus reconstructing the coupling relationships in the edge computing environment. A potential-driven multi-learning particle swarm optimization algorithm is employed to uniformly solve a multi-objective optimization model coupling latency, energy consumption, and load balancing. Utilizing the algorithm's global search mechanism, it directly seeks Pareto optimal solutions in a massive solution space, overcoming the inherent weakness of traditional separate optimization methods that easily get trapped in local optima. This ensures that the final joint decision-making scheme maximizes the reduction of total system energy consumption and achieves load balancing among multiple servers while guaranteeing task real-time performance. Attached Figure Description

[0017] Figure 1 This is a flowchart of the UAV edge computing task offloading method according to Embodiment 1 of the present invention; Figure 2This is a schematic diagram illustrating an application scenario of the UAV edge computing task offloading method according to Embodiment 2 of the present invention; Figure 3 This is a flowchart of the UAV edge computing task offloading method according to Embodiment 2 of the present invention; Figure 4 This is a flowchart of the PDML-PSO solution process in Embodiment 2 of the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] Example 1 This embodiment provides a method for offloading edge computing tasks from a drone, such as... Figure 1 As shown, the process includes the following: Construct an edge computing model for drones.

[0020] For the inspection tasks to be performed by the drone, multiple decision variables are identified, including unloading decisions, base station selection decisions, transmission power, server selection decisions, and server computing resource allocation.

[0021] A communication model and a computing and energy consumption model are established based on decision variables and UAV edge computing models.

[0022] A multi-objective optimization model is constructed based on the communication model and the computing and energy consumption model. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computing and communication energy consumption, and the server load balancing index.

[0023] A potential-driven multi-learning particle swarm optimization algorithm is used to iteratively solve the multi-objective optimization model, and the optimal unloading scheme of the UAV is output as the global optimal solution.

[0024] Specifically, a communication model is established, including: determining the uplink transmission rate based on transmit power, air-to-ground channel gain, and co-channel interference; determining the uplink transmission delay by combining the uplink transmission rate, offload decision, base station selection decision, and task data volume; determining the backhaul link delay based on base station selection decision, the connection relationship between the base station and the mobile edge computing server, server selection decision, task data volume, and the fixed backhaul link transmission rate; and calculating the total communication delay by adding the uplink transmission delay, backhaul link delay, and base station processing delay, and constructing the communication model based on the total communication delay.

[0025] This model not only considers the basic data transmission volume and bandwidth, but also incorporates the real-time impact of air-to-ground channel gain, co-channel interference, and dynamic changes in transmit power on the uplink rate. Combined with the fixed transmission rate of the wired backhaul link, it comprehensively depicts the end-to-end communication flow. This communication model, which closely reflects actual channel fading and network topology limitations, can accurately quantify the total communication delay during the offloading process, effectively avoiding the risk of severe communication congestion or high latency in actual deployments due to overly idealized channel state assessments.

[0026] Specifically, the computing and energy consumption model includes the total edge computing latency. The calculation process of the total edge computing latency includes: determining the local computing latency of the UAV based on the task computing load and the clock frequency of the UAV's local processor; determining the queuing latency and server computing latency based on server computing resource allocation, task computing load, average queuing waiting time of priority tasks, and service rate; and calculating the total edge computing latency by adding the total communication latency, queuing latency, and server computing latency together.

[0027] Based on queuing theory and the principle of computing resource allocation, the time overhead of edge computing is separated and quantified. It not only calculates the absolute processing time of local execution and edge execution, but also introduces a queuing delay mechanism in multi-task concurrency scenarios. By taking task priority and server service rate into consideration, it can predict the congestion level of different computing nodes in advance. By dynamically balancing the high time consumption of local computing with the high time consumption of offloading queuing, it guides computing tasks to avoid high-load server nodes, thereby significantly reducing the waiting time of tasks in the queue and the total end-to-end processing time at the global level.

[0028] Specifically, the computation and energy consumption model includes energy consumption. The energy consumption calculation process includes: determining local computation energy consumption based on the UAV's local processor clock frequency, chip energy consumption coefficient, and task computation load; determining communication energy consumption by combining base station selection decisions, transmission power, and uplink transmission delay; determining hovering energy consumption based on UAV hovering power, unloading decisions, UAV local computation delay, and uplink transmission delay; determining flight maneuver energy consumption based on air density, total rotor area, drag coefficient, UAV cruise speed, and Euclidean distance between inspection points; and establishing battery power constraints based on local computation energy consumption, communication energy consumption, hovering energy consumption, flight maneuver energy consumption, and unloading decisions, using battery power constraints as constraints in the multi-objective optimization model.

[0029] By dividing the energy consumption of drones into four dimensions—local computing, communication, hovering, and flight maneuvering—and establishing a direct mathematical mapping between them and unloading decisions, this full-lifecycle energy consumption modeling method can perceive the energy game brought about by each unloading decision (e.g., unloading saves computing power but increases communication and hovering power). By establishing mandatory battery power safety constraints and following the law of conservation of energy, it eliminates the risk of drones crashing due to improper task scheduling and running out of power in the air.

[0030] Specifically, the multi-objective optimization model includes the following constraints: Offload-Base Station Selection Logic Constraint: When an inspection task is executed locally, no base station is allocated; when an inspection task is offloaded, only one base station is allocated. Network Topology Validity Constraint: When an inspection task is transmitted through a base station and the base station is directly connected to the mobile edge computing server, the inspection task is assigned to the mobile edge computing server for execution. Base Station Wireless Resource Capacity Constraint: The total number of inspection tasks transmitted through the base station does not exceed the base station's wireless resource limit. Server Computing Resource Capacity Constraint: The sum of computing frequencies allocated to the mobile edge computing server does not exceed the mobile edge computing server's computing frequency limit. Task Latency Constraint: The total latency of the inspection tasks does not exceed the maximum tolerable latency. UAV Battery Safety Constraint: The total energy consumption of the UAV executing all inspection tasks does not exceed the product of the UAV's total battery capacity and safety factor. Transmission Power Constraint: The transmission power of the inspection tasks does not exceed the UAV's maximum transmission power, and the sum of the transmission power of all inspection tasks does not exceed the total power budget. Variable Domain Constraint: Base station selection decisions and server selection decisions are integer variables, and server computing resource allocation is a non-negative continuous variable.

[0031] Various constraints narrow the meaningless search space, forcing the particle swarm algorithm to iterate only within a safe and executable effective boundary, ensuring that the optimal unloading scheme of the decoding output is feasible for engineering implementation, and avoiding invalid schemes that output theoretically optimal but cannot be run in practice.

[0032] Specifically, a potential-driven multi-learning particle swarm optimization algorithm is used to iteratively solve a multi-objective optimization model, outputting the optimal unloading scheme for the UAV as the global optimal solution. This includes: randomly generating an initial particle swarm and setting algorithm parameters; dividing decision variables into binary and continuous variables, and performing mixed encoding to obtain particle position vectors. Binary variables include unloading decisions, base station selection decisions, and server selection decisions, while continuous variables include transmission power and server computing resource allocation; decoding the particle position vectors into decision variables, and calculating the particle fitness values ​​based on the objective function of the multi-objective optimization model and dynamic penalty coefficients; updating the individual particle's historical best position and the global historical best position based on the fitness values; iteratively updating the particle position vectors based on the individual particle's historical best position and the global historical best position, stopping iteration when the maximum number of iterations is reached, outputting the global historical best position as the global optimal solution, and decoding the global optimal solution to obtain the optimal unloading scheme for the UAV.

[0033] By hybrid encoding discrete binary variables (decision-making and routing) and continuous variables (power and resource allocation), the mathematical challenge of handling mixed variable spaces, which is difficult for traditional algorithms, is solved. Simultaneously, a dynamic penalty coefficient related to the iteration number is introduced into the fitness function, transforming the optimization problem with stringent constraints into an unconstrained optimization problem. This mechanism allows particles some exploration ability in infeasible solution regions in the early stages of iteration to maintain population diversity, while imposing a heavy penalty in the later stages of iteration to force particles to converge to feasible solution regions, thus improving the algorithm's convergence speed and optimization accuracy.

[0034] Specifically, the particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position. This includes: obtaining a random elite guidance strategy; updating the particle velocity by combining the individual particle's historical best position, the global historical best position, and a guidance position randomly selected from the current elite particles; updating the particle position vector based on the updated particle velocity; calculating the constraint violation degree of the constraint conditions corresponding to the updated particle position vector; calculating the mutation probability based on the constraint violation degree; obtaining a random number; when the random number is less than the mutation probability, applying a Gaussian perturbation to the continuous variable of the selected particle position vector and performing a bit flip operation on the binary variable of the selected particle position vector; and returning to the step of calculating the particle's fitness value until the maximum number of iterations is reached.

[0035] Abandoning the traditional single global optimum guidance, a random elite guidance strategy is introduced to solve the problem of homogeneous evolution of the population, effectively curbing premature convergence of the algorithm. Constraint violation is used as a trigger mechanism to perform adaptive mutations on particles using Gaussian perturbations (continuous variables) and bit flips (discrete variables). This feedback-based error correction mechanism endows the potential-driven multi-learning particle swarm optimization algorithm with the ability to escape local optima and cross infeasible regions, and can still robustly search for the true global optimum even when facing high-dimensional and severe constraints.

[0036] Example 2 This embodiment provides a method for offloading edge computing tasks from a drone, such as... Figure 2 As shown, the application scenario of this method is an unmanned aerial vehicle (UAV) flight system. The UAV flight system includes a UAV, a base station (BS), and a mobile edge computing server (MEC). The UAV connects to the base station via a wireless link, and the base station connects to the mobile edge computing server (referred to as the server in this embodiment) via a wired backhaul link. The computing tasks initiated by the UAV can be offloaded to a specific server connected to the selected base station for execution, thereby forming a complete edge computing service chain.

[0037] like Figure 3 As shown, the UAV edge computing task offloading method described in this embodiment mainly includes the following processes: Step S100: Establish a three-layer system model.

[0038] Build as Figure 2 The three-layer UAV edge computing model shown includes: The drone layer consists of N drones. The drones are assembled. Each drone performs a set of inspection tasks in sequence, and the results are recorded. This is the set of inspection tasks that UAV i is to perform.

[0039] Base station layer: contains B base stations, denoted as It serves as a base station cluster to provide wireless access services for drones.

[0040] Server layer: Deploy S-servers, and remember... It is a collection of servers connected to a base station via a wired link, providing computing power for offloading tasks.

[0041] Define the connection relationship: ∈{0,1}, The connection relationship between base station b and server s is as follows: when base station b is directly connected to server s, =1.

[0042] Step S200: Define the task and decision variables.

[0043] Each inspection task is described by a 5-tuple:

[0044] Represents a quintuple. Indicates the amount of task data. Indicates computational complexity. Indicates the maximum tolerable delay. Indicates the task type. Indicates priority. In this formula and all formulas below, i represents the i-th drone and k represents the k-th task.

[0045] For each inspection task, the drone makes five types of joint decisions: 1. Uninstallation decision (0: execute locally, 1: uninstall).

[0046] 2. Base station selection decision (0: transmission not through base station b, 1: transmission through base station b).

[0047] 3. Transmission power .

[0048] 4. Server selection decision (0: Not assigned to server s, 1: Assigned to server s).

[0049] 5. Server computing resource allocation .

[0050] Step S300: Establish a communication model.

[0051] This communication model is used to describe the data transmission process between the UAV and the base station, and between the base station and the server. It includes four parts: uplink transmission rate calculation, uplink transmission delay calculation, backlink delay calculation, and total communication delay calculation, which correspond to steps S301 to S304 respectively.

[0052] Step S301: Determine the uplink transmission rate :

[0053] in, Indicates the air-to-ground channel gain. , Channel gain at reference distance, Let be the Euclidean distance between drone i and base station b. This is the path loss index. This indicates small-scale Rayleigh fading. The fixed uplink bandwidth allocated to the drone. For noise power spectral density, Indicates the transmission power. This indicates co-channel interference from other drones. , Indicates the j-th drone. Indicates that drone j executes the first... Transmission power for each mission This represents the channel gain between drone j and base station b.

[0054] Step S302: Determine the uplink transmission delay :

[0055] Where B represents the number of base stations.

[0056] Step S303: Determine the backhaul link delay :

[0057] Where S represents the number of servers. This indicates a fixed transmission rate for the backhaul link.

[0058] Step S304: Determine the total communication delay :

[0059] in, To handle latency for base stations.

[0060] Step S400: Establish a calculation and energy consumption model.

[0061] Step S401: Determine the local computational latency of the drone. Energy consumption :

[0062]

[0063] This indicates the local processor clock frequency of drone i. This represents the chip energy consumption coefficient of drone i.

[0064] Step S402: Determine the total edge computing latency :

[0065] in, This indicates a queue delay. This indicates server computation latency. , , This indicates the computation frequency assigned by server s to task k of drone i. Indicates priority as The average queuing time of a task on server s is calculated based on the M / M / 1 queuing theory model. This indicates that server s has a priority of The service rate of the task.

[0066] Step S403: Determine the UAV energy consumption model.

[0067] Drone energy consumption includes communication energy consumption Hovering energy consumption Local computing power consumption Flight maneuver energy consumption .

[0068] Communication power consumption Energy consumption generated when the drone uploads data during mission unloading.

[0069]

[0070] Hovering energy consumption Energy consumption during drone hovering during mission execution.

[0071]

[0072] in, Here, represents the hovering power of the drone, which is a fixed constant.

[0073] Local computing power consumption Energy consumption generated when the task is performed locally on the drone is calculated using formula (7) in step S401.

[0074] Flight maneuver energy consumption Energy consumption generated by drones flying between inspection points.

[0075]

[0076] in, Let A be the air density and A be the total rotor area. The drag coefficient, For the drone's cruising speed, The distance between inspection points is the Euclidean distance.

[0077] Four energy consumption parameters are determined based on offloading decisions. Combination, local execution ( This includes computational energy consumption, hovering energy consumption, and flight maneuvering energy consumption, during unloading execution ( This includes communication energy consumption, hovering energy consumption, and flight maneuvering energy consumption, as detailed in step S404, battery power constraint.

[0078] Step S404: Determine battery power constraints:

[0079] in This indicates the total battery capacity of the drone. Indicates the safety factor. =0.8, with 20% of the power reserved for emergency return.

[0080] Step S500: Construct a multi-objective optimization model.

[0081] In step S100 Based on S400, the unmanned aerial vehicle (UAV) mission offloading problem is modeled as the following multi-objective optimization model: Objective function:

[0082] in: , For maximum task delay, The total delay for drone i to perform the k-th task is defined as:

[0083] , U represents the set of drones, representing the sum of energy consumption for computation and communication.

[0084] , For mobile edge computing server load balancing metrics, This indicates the upper limit of the server's calculation frequency. γ and γ are preset weighting coefficients. For the target load rate parameter, Represents the weighting coefficient, and , , .

[0085] Constraints (C1) C8) includes the range of decision variables, resource constraints, delay constraints, and power constraints.

[0086]

[0087] Indicates the upper limit of base station wireless resources. This indicates the task deadline that meets the task timeline. This indicates the maximum transmission power of the drone. This represents the total launch power budget for drone i across all missions.

[0088] The constraints are explained as follows: C1 (Offload-Base Station Selection Logic Constraint): This constraint ensures a strict correspondence between offload decisions and base station selection. Specifically, if the task... Select local execution ( If no base station is allocated, then no base station will be assigned. If the task is selected to be uninstalled and executed (); Therefore, only one base station must be allocated. ).

[0089] C2 (Network Topology Validity Constraint): This constraint guarantees that the selected offloading path is physically feasible. Task Only through base stations Transmission, and the base station and server Directly connected ( Only when this is done can the task be assigned to the server. Execution. That is: At that time, at least one satisfy and .

[0090] C3 (Base Station Radio Resource Capacity Constraint): This constraint limits the number of drones accessing the same base station to no more than the base station's maximum radio resource capacity. That is, through the base station... Total number of tasks transmitted Must not exceed the base station upper limit of wireless resources .

[0091] C4 (Server Computing Resource Capacity Constraint): This constraint limits the total computing resources allocated to the same server to no more than the server's maximum computing capacity. In other words: server The sum of the calculation frequencies of the allocation Must not exceed the server upper limit of calculation frequency .

[0092] C5 (Task Delay Constraint): This constraint ensures that the actual completion time of each task meets real-time requirements. That is: Task Total delay It must not exceed its maximum tolerable delay. .

[0093] C6 (Drone Battery Safety Constraint): This constraint limits the total energy consumption of the drone for all missions, reserving 20% ​​of the battery capacity for emergency return. In other words: Drone The total energy consumption for performing all tasks must not exceed the total battery capacity. 80% ).

[0094] C7 (Transmit Power Constraint): This constraint controls the UAV's communication power consumption and ensures it conforms to the device's physical limitations. Specifically, for each mission... Transmission power The maximum transmit power of the drone must not be exceeded. Furthermore, the sum of the transmit power of all missions must not exceed the total power budget. .

[0095] C8 (Domain Constraint): This constraint specifies the mathematical range of values ​​for each decision variable. Among them, the base station selection variable... Server selection variables For 0-1 integer variables, server computing resource allocation It is a non-negative continuous variable.

[0096] Step S600: Solve PDML-PSO.

[0097] like Figure 4 As shown, the multi-objective optimization model constructed in step S500 is solved using the potential-driven multi-learning particle swarm optimization (PDML-PSO) algorithm, and the process is as follows: Step S601: Initialize the population and parameters.

[0098] An initial particle population is randomly generated in the solution space, with each particle corresponding to a potential solution of the multi-objective optimization model. The population size and maximum number of iterations are set. Initial inertia weights Termination inertia weight Learning factors attenuation coefficient and coefficient of variation .

[0099] Step S602: Mix and encode particle positions.

[0100] Encode the five types of decision variables in step S200 as particle positions: binary variables. (Uninstallation decision) Base station selection decision Server selection decision ) and continuous variables (transmission power) Computing resource allocation The particle position vector is represented as:

[0101] in Represents binary variables. This represents a continuous variable.

[0102] Step S603: Calculate fitness value .

[0103] Decode the particle position as a decision variable and substitute it into the objective function to calculate the fitness value:

[0104] in This represents the dynamic penalty coefficient that increases with the number of iterations. Indicates the first The degree of violation of each constraint (when the constraint is satisfied) When violated ), This represents the position vector of the particle. The lower the fitness value, the higher the quality of the solution.

[0105] Step S604: Update And Gbest.

[0106] Update the individual best historical position of each particle based on its fitness value. And the global historical best position Gbest for the entire population.

[0107] Step S605: Update particle velocity Rbest.

[0108] Introducing a random elite-guided strategy to update particle velocity:

[0109] Represents particles In the The speed of generation Represents particles In the The speed of generation Indicates inertia weight, express Random numbers within the interval Represents particles The individual's historical best position, Represents particles In the The position of the generation, This indicates a guiding position randomly selected from the current elite particles, where elite particles refer to particles with the top 20% fitness values.

[0110] Step S606: Update particle positions.

[0111] Update particle positions based on the updated velocity.

[0112]

[0113] in: Indicates the position of the next generation of particles. Indicates the current particle position.

[0114] For hybrid encoded particles: Continuous variables : Update directly using the above formula (19).

[0115] binary variables The same formula (19) is used to update, but the updated value is a continuous value, which needs to be converted into a binary value.

[0116] Step S607: Calculate the degree of constraint violation .

[0117] Calculate the degree to which each particle violates the constraints:

[0118] This represents the position vector of particle m.

[0119] Step S608: Determine whether to perform adaptive mutation.

[0120] Calculate the probability of variation based on the degree of constraint violation. :

[0121] Where the coefficient of variation , Represents particles The degree of constraint violation This represents the maximum constraint violation rate among all particles in the current population.

[0122] Decision logic: Generate a random number for each particle. .

[0123] like If the particle is not selected to undergo mutation, proceed to step S609. If the particle is selected, mutation will be performed, and the process will proceed to step S610.

[0124] Step S609: Determine whether the maximum number of iterations has been reached.

[0125] If the fitness value is not reached, return to step S603 to calculate the fitness value and continue iterating; if the fitness value has been reached, proceed to step S611.

[0126] Step S610: Gaussian perturbation / bit flip.

[0127] Apply a Gaussian perturbation to the continuous variable of the selected particle, and the binary variable is permuted by probability. Flip it. Then proceed to step S609.

[0128] Step S611: Output the global optimal solution.

[0129] Decode the globally optimal location, which is the optimal joint decision-making scheme for UAV edge computing tasks (including optimal offloading decision, base station selection, transmission power, server selection, and computing resource allocation).

[0130] The beneficial effects of this embodiment are mainly reflected in the following aspects: Coordinated optimization of total energy consumption: By integrating a multi-objective optimization model for communication, computing, hovering and flight energy consumption, and utilizing the stochastic guidance and adaptive mutation mechanism of the PDML-PSO algorithm, the global search capability for energy-saving strategies such as low-power transmission and efficient resource allocation is enhanced within all possible decision variable values. This effectively reduces the overall energy consumption of the UAV swarm during mission execution while meeting the real-time requirements of the mission.

[0131] Joint reduction of end-to-end task latency: Since the optimization objective directly includes the maximum task completion time, and the multi-objective optimization model couples communication and computing latency, and the PDML-PSO algorithm achieves global dynamic load balancing through joint iterative optimization of offloading decisions, path selection (base station / server) and computing resources, it effectively reduces the total time of task queuing and processing.

[0132] Server load balancing: Since a load balancing penalty term is explicitly introduced into the optimization objective, it provides a clear direction for the algorithm search. Furthermore, by processing server selection and resource allocation variables, the PDML-PSO algorithm can actively explore feasible solutions for migrating tasks from high-load nodes to low-load nodes. Therefore, it significantly improves the load balancing among multiple servers, enhances resource utilization, and improves system robustness.

[0133] In summary, this embodiment explicitly models the relationship between performance indicators and decision variables, and leverages the powerful global search and constraint handling capabilities of the PDML-PSO algorithm to efficiently solve the multi-objective optimization model. Therefore, the obtained offloading scheme can directly and synergistically optimize the aforementioned key performance indicators, ultimately achieving a comprehensive performance improvement characterized by low total system energy consumption, short task processing latency, and good server load balancing.

[0134] Example 3: In this embodiment, a drone edge computing task offloading system is provided. This drone edge computing task offloading system can be used to implement the above-mentioned drone edge computing task offloading method. Specifically, the drone edge computing task offloading system includes a drone edge computing model construction module, a decision variable determination module, a communication model and a computing and energy consumption model construction module, a multi-objective optimization model construction module, and an optimal offloading scheme output module.

[0135] The UAV edge computing model building module is used to build UAV edge computing models.

[0136] The decision variable determination module is used to determine multiple decision variables for the inspection tasks to be performed by the UAV. These decision variables include unloading decisions, base station selection decisions, transmission power, server selection decisions, and server computing resource allocation.

[0137] The communication model and computation and energy consumption model building module is used to build communication models and computation and energy consumption models based on decision variables and UAV edge computing models.

[0138] The multi-objective optimization model building module is used to build multi-objective optimization models based on communication models and computation and energy consumption models. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computation and communication energy consumption, and the server load balancing index.

[0139] The optimal unloading scheme output module is used to iteratively solve the multi-objective optimization model using a potential-driven multi-learning particle swarm optimization algorithm, and output the optimal unloading scheme of the UAV as the global optimal solution.

[0140] Example 4: This embodiment provides a computer device. The terminal device includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), or field-programmable gate arrays (FPGAs). Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., are the computing and control core of the terminal. They are suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function. The processor described in this embodiment of the invention can be used for the operation of the UAV edge computing task offloading method, including: constructing a UAV edge computing model; determining multiple decision variables for the inspection task to be performed by the UAV, including offloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation; establishing a communication model and a computing and energy consumption model based on the decision variables and the UAV edge computing model; constructing a multi-objective optimization model based on the communication model and the computing and energy consumption model, the objective function of the multi-objective optimization model including the maximum task delay, the sum of computing and communication energy consumption, and the server load balancing index; using a potential-driven multi-learning particle swarm optimization algorithm to iteratively solve the multi-objective optimization model, and outputting the optimal UAV offloading scheme as the global optimal solution.

[0141] Example 5: This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device.

[0142] One or more instructions stored in a computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the UAV edge computing task offloading method in the above embodiments. One or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: constructing a UAV edge computing model; determining multiple decision variables for the inspection task to be performed by the UAV, including offloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation; establishing a communication model and a computing and energy consumption model based on the decision variables and the UAV edge computing model; constructing a multi-objective optimization model based on the communication model and the computing and energy consumption model, wherein the objective function of the multi-objective optimization model includes the maximum task delay, the sum of computing and communication energy consumption, and the server load balancing index; using a potential-driven multi-learning particle swarm optimization algorithm to iteratively solve the multi-objective optimization model and output the optimal UAV offloading scheme as the global optimal solution.

[0143] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0144] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0145] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0146] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0147] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0148] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0149] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and applications beyond the examples provided will become apparent to those skilled in the art upon reading the above description.

Claims

1. A method for offloading edge computing tasks from a drone, characterized in that, Includes the following processes: Constructing an edge computing model for drones; For the inspection tasks to be performed by the drone, multiple decision variables are identified, including unloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation. Based on decision variables and UAV edge computing models, establish communication, computing, and energy consumption models; A multi-objective optimization model is constructed based on the communication model and the computing and energy consumption model. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computing and communication energy consumption, and the server load balancing index. A potential-driven multi-learning particle swarm optimization algorithm is used to iteratively solve the multi-objective optimization model, outputting the optimal unloading scheme for the UAV as the global optimal solution, including: Randomly generate an initial particle population and set algorithm parameters; The decision variables are divided into binary variables and continuous variables, and mixed encoding is performed to obtain the particle position vector. The binary variables include unloading decision, base station selection decision and server selection decision, while the continuous variables include transmission power and server computing resource allocation. The particle position vector is decoded into a decision variable, and the fitness value of the particle is calculated by combining the objective function of the multi-objective optimization model and the dynamic penalty coefficient. Update the individual particle's historical best position and the global historical best position based on the fitness value; The particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position. The iteration stops when the maximum number of iterations is reached, and the global historical best position is output as the global optimal solution. The global optimal solution is then decoded to obtain the optimal unloading scheme for the UAV. The particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position, including: A random elite guidance strategy is obtained, and the particle velocity is updated by combining the individual particle's historical best position, the global historical best position, and the guidance position randomly selected from the current elite particles. The elite particles are the particles with the top 20% fitness values. Update the particle position vector based on the updated particle velocity; Calculate the constraint violation degree of the updated particle position vector corresponding to the constraint condition, and calculate the mutation probability based on the constraint violation degree; Get a random number. If the random number is less than the mutation probability, apply a Gaussian perturbation to the continuous variable of the selected particle position vector and perform a bit flip operation on the binary variable of the selected particle position vector. Return to the step that calculates the fitness value of the particle, until the maximum number of iterations is reached.

2. The UAV edge computing task offloading method according to claim 1, characterized in that, Establish a communication model, including: The uplink transmission rate is determined based on transmit power, air-to-ground channel gain, and co-channel interference. The uplink transmission delay is determined by combining the uplink transmission rate, offload decision, base station selection decision, and task data volume. The backhaul link delay is determined based on base station selection decision, the connection relationship between the base station and the mobile edge computing server, server selection decision, task data volume, and the fixed transmission rate of the backhaul link. The total communication delay is calculated by adding uplink transmission delay, backhaul link delay, and base station processing delay, and a communication model is constructed based on the total communication delay.

3. The UAV edge computing task offloading method according to claim 2, characterized in that, The computation and energy consumption model includes the total edge computing latency. The calculation process for the total edge computing latency includes: The local computing latency of the drone is determined based on the task's computational load and the drone's local processor clock frequency. Queuing latency and server computing latency are determined based on server computing resource allocation, task computing volume, average queuing wait time of priority tasks, and service rate. The total edge computing latency is calculated by adding the total communication latency, queuing latency, and server computing latency together.

4. The UAV edge computing task offloading method according to claim 3, characterized in that, The computational and energy consumption model includes energy consumption, and the energy consumption calculation process includes: The local computing power consumption is determined based on the local processor clock frequency, chip power consumption coefficient, and task computation load of the drone. The communication energy consumption is determined by combining base station selection decisions, transmission power, and uplink transmission delay. Hovering energy consumption is determined based on drone hovering power, unloading decision, drone local computing latency, and uplink transmission latency. The flight maneuver energy consumption is determined based on air density, total rotor area, drag coefficient, UAV cruise speed, and Euclidean distance between inspection points. Battery power constraints are established based on local computing energy consumption, communication energy consumption, hovering energy consumption, flight maneuvering energy consumption, and unloading decisions, and are used as constraints for the multi-objective optimization model.

5. The UAV edge computing task offloading method according to claim 1, characterized in that, The multi-objective optimization model includes constraints, which include: Offloading-Base Station Selection Logic Constraints: When the inspection task is executed locally, no base station is allocated; when the inspection task is executed offloading, only one base station is allocated. Network topology validity constraint: When the inspection task is transmitted through the base station and the base station is directly connected to the mobile edge computing server, the inspection task will be assigned to the mobile edge computing server for execution. Base station wireless resource capacity constraint: The total number of inspection tasks transmitted through the base station shall not exceed the base station's wireless resource limit. Server computing resource capacity constraints: The sum of the computing frequencies allocated to mobile edge computing servers shall not exceed the upper limit of the computing frequency of mobile edge computing servers; Task delay constraint: The total delay of the inspection task shall not exceed the maximum tolerable delay; Drone battery safety constraints: The total energy consumption of the drone in performing all inspection tasks shall not exceed the product of the total capacity of the drone battery and the safety factor; Transmission power constraint: The transmission power of the inspection mission shall not exceed the maximum transmission power of the UAV, and the sum of the transmission power of all inspection missions shall not exceed the total power budget; Domain constraints for variables: Base station selection decisions and server selection decisions are integer variables, while server computing resource allocation is a non-negative continuous variable.

6. A drone edge computing task offloading system, characterized in that, include: The UAV edge computing model building module is used to build UAV edge computing models. The decision variable determination module is used to determine multiple decision variables for the inspection tasks to be performed by the UAV. The decision variables include unloading decision, base station selection decision, transmission power, server selection decision, and server computing resource allocation. The communication model and computation and energy consumption model building module is used to build communication models and computation and energy consumption models based on decision variables and UAV edge computing models; The multi-objective optimization model construction module is used to build a multi-objective optimization model based on the communication model and the computing and energy consumption model. The objective function of the multi-objective optimization model includes the maximum task latency, the sum of computing and communication energy consumption, and the server load balancing index. The optimal unloading scheme output module is used to iteratively solve a multi-objective optimization model using a potential-driven multi-learning particle swarm optimization algorithm, and outputs the optimal unloading scheme for the UAV as the global optimal solution, including: Randomly generate an initial particle population and set algorithm parameters; The decision variables are divided into binary variables and continuous variables, and mixed encoding is performed to obtain the particle position vector. The binary variables include unloading decision, base station selection decision and server selection decision, while the continuous variables include transmission power and server computing resource allocation. The particle position vector is decoded into a decision variable, and the fitness value of the particle is calculated by combining the objective function of the multi-objective optimization model and the dynamic penalty coefficient. Update the individual particle's historical best position and the global historical best position based on the fitness value; The particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position. The iteration stops when the maximum number of iterations is reached, and the global historical best position is output as the global optimal solution. The global optimal solution is then decoded to obtain the optimal unloading scheme for the UAV. The particle position vector is iteratively updated based on the individual particle's historical best position and the global historical best position, including: A random elite guidance strategy is obtained, and the particle velocity is updated by combining the individual particle's historical best position, the global historical best position, and the guidance position randomly selected from the current elite particles. The elite particles are the particles with the top 20% fitness values. Update the particle position vector based on the updated particle velocity; Calculate the constraint violation degree of the updated particle position vector corresponding to the constraint condition, and calculate the mutation probability based on the constraint violation degree; Get a random number. If the random number is less than the mutation probability, apply a Gaussian perturbation to the continuous variable of the selected particle position vector and perform a bit flip operation on the binary variable of the selected particle position vector. Return to the step that calculates the fitness value of the particle, until the maximum number of iterations is reached.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the UAV edge computing task offloading method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the UAV edge computing task offloading method as described in any one of claims 1 to 5.

Images