Unloading calculation and cooperative calculation method based on Stackelberg game in unmanned aerial vehicle

Abstract

The invention discloses a Stackelberg game-based calculation unloading and cooperative calculation method in an unmanned aerial vehicle. The method comprises the steps of constructing a mobile edge calculation system; defining a task unloading strategy of the MUE and a cooperative forwarding strategy of the UAV; establishing system cost models including a G2A communication model, an A2A communication model, a calculation queuing model and a system energy consumption model, and establishing a task completion time delay model including constructing an optimization problem; modeling the optimization problem as a Stackelberg game; the UAV is used as a leader, and the MUE is used as a follower; alternately updating the cooperative forwarding strategy and the task unloading strategy by adopting a distributed iteration strategy solving algorithm, and outputting a balanced unloading strategy and a balanced forwarding strategy; and the MUE offloads the task to the UAV according to a balanced offload strategy, and controls the UAV to forward the task to an adjacent UAV or to process locally according to a balanced forwarding strategy. According to the method, cooperative processing of tasks is realized based on the Stackelberg game, and the comprehensive cost is effectively reduced.

Description

Technical Field

[0001] This invention relates to the fields of mobile communication and mobile edge computing technology, and in particular to a computational offloading and collaborative computing method based on Stackelberg game in unmanned aerial vehicles (UAVs). Background Technology

[0002] The development of mobile communication technology, the Internet of Things (IoT), and artificial intelligence (AI) has driven the continuous development of emerging applications such as virtual reality, natural language processing, and autonomous driving. These applications typically generate computationally intensive and latency-sensitive tasks, placing higher demands on the computing power of Mobile User Equipment (MUE). Limited by factors such as size, hardware, and power supply, MUEs struggle to independently handle the increasing computing demands, leading to higher latency and energy consumption during local task execution.

[0003] Cloud computing can offload tasks to the cloud for processing, alleviating the lack of computing power at the terminal. However, the distance between the cloud and the terminal is relatively large, and task transmission and backhaul usually result in long transmission latency and may face the risks of network congestion and data loss, thus limiting the applicability of cloud computing in low-latency services. Mobile Edge Computing (MEC) deploys computing and storage resources close to the network edge (e.g., at the base station) near the MUE, enabling the MUE to offload tasks to nearby MEC servers for processing, thereby reducing task processing latency, alleviating network congestion, and improving Quality of Service (QoS).

[0004] However, traditional MECs typically rely on fixed ground base stations, limiting coverage and deployment flexibility. Ground communication and computing services are easily disrupted in scenarios such as base station damage due to natural disasters, large-scale temporary events, or insufficient network coverage in remote areas. Combining unmanned aerial vehicles (UAVs), which offer high mobility and rapid deployment, with MECs can create UAV-assisted MEC networks. In this case, UAVs can be deployed on demand as aerial mobile base stations and edge computing nodes, providing temporary and reliable communication and computing services to ground users. Therefore, there is an urgent need to design efficient and reliable offloading schemes to meet diverse mission requirements and reduce terminal power consumption.

[0005] Existing research on computational offloading for UAV-assisted MEC is relatively abundant, but early work mostly focused on single-UAV scenarios, improving QoS by optimizing UAV trajectories or resource allocation. However, with the increase in the number of users and the expansion of their distribution areas, single UAVs are limited in terms of coverage and computing power, making it difficult to meet the demands of large-scale services. Therefore, multi-UAV collaborative deployment and resource scheduling have become important directions for improving system stability and service efficiency. Although deploying multiple UAVs can effectively improve system QoS, in practical applications, the distribution of MUEs (Multi-User UAVs) is random. This non-uniform distribution can lead to UAVs closest to hotspot areas receiving massive offloading requests, quickly creating computational bottlenecks, while distant UAVs may be idle. This load imbalance not only drastically increases the task processing latency of overloaded UAVs but may even cause service denial due to queue overflow, ultimately leading to a sharp deterioration in the QoS of the entire system. Therefore, this invention mainly studies the MEC architecture for multi-UAV collaborative operation.

[0006] In multi-UAV scenarios, the joint optimization of computation offloading and collaborative computation often results in a complex global optimization problem. Meanwhile, centralized solutions usually require the central controller to acquire information from the entire network in real time, which may lead to unbearable signaling overhead and engineering implementation difficulties. Summary of the Invention

[0007] In view of this, the purpose of this invention is to propose a computational offloading and collaborative computation method based on Stackelberg game in UAVs, which effectively reduces the overall cost based on Stackelberg game-based equilibrium offloading and equilibrium forwarding strategies.

[0008] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: This invention provides a computational offloading and collaborative computation method based on Stackelberg game in unmanned aerial vehicles (UAVs), comprising the following steps: Step 1: Construct a mobile edge computing system consisting of multiple MUEs and multiple UAVs; and define the task offloading strategy for MUEs and the collaborative forwarding strategy for UAVs; Step 2: Establish a system cost model, which includes at least a task completion delay model and a system energy consumption model. The task completion delay model includes a G2A communication model, an A2A communication model, and a computation queuing model. Construct an optimization problem with minimizing the total system cost as the optimization objective. Step 3: Based on the interaction between the cooperative forwarding strategy of UAVs and the task offloading strategy of MUEs, the optimization problem is modeled as a Stackelberg game with multiple leaders and multiple followers; where each UAV is a leader and each MUE is a follower. Step 4: Based on Stackelberg game theory, a distributed iterative strategy algorithm is used to alternately update the cooperative forwarding strategy and the task offloading strategy until the convergence condition is met, and output the balanced offloading strategy and the balanced forwarding strategy. Step 5: Each MUE offloads its tasks to the corresponding UAV according to the load balancing offload strategy, and controls each UAV to forward the received tasks to the corresponding adjacent UAV or local processing according to the load balancing forwarding strategy, so as to achieve collaborative processing of tasks.

[0009] Furthermore, step 1 specifically includes: Step 11: Construct a mobile edge computing system consisting of N UAVs and M MUEs; Step 12: Define the set of UAVs as follows The set of MUEs is ,in, n Indicates the index of the UAV. m Indicates the index of MUE, MUE m Indicates the first m MUE, UAV n Indicates the first n UAV; assuming MUE m One task is generated in each unit of time, using triples. It means that, among them, Indicates the amount of task data. This indicates the number of CPU cycles required to process a unit of data. Indicates the total number of computation cycles required for the task; Step 13: Define MUE m The feasible G2A communication set is UAV n The set of serviceable MUEs is UAV n The feasible A2A communication set is ; Step 14: Define MUE m The task uninstallation strategy is ,in, Indicates MUE m The task Uninstall to UAV n The proportion to be processed is used to form G2A subtasks. ; Indicates task The proportion processed locally; Define UAV n The collaborative forwarding strategy is ,in, , UAVn G2A subtask Forward to adjacent UAV k The proportion of tasks to be processed is used to form A2A subtasks. ;when hour, UAV n The proportion of tasks processed locally; Step 15: Based on the defined task offloading and collaborative forwarding strategies, perform task processing in two phases: G2A offloading phase: MUE m The task is split according to the task unloading strategy and divided into two parts according to the preset ratio. One part of the task is a G2A subtask, which is unloaded to the corresponding UAV for processing through the G2A link; the other part of the task is processed locally. A2A Collaboration Phase: UAV n According to the cooperative forwarding strategy, each G2A subtask received is further split into A2A subtasks, and then forwarded to the corresponding adjacent UAV via the A2A link. k Or it can be processed locally.

[0010] Furthermore, the specific process of establishing the G2A communication model in step 2 is as follows: Step 211, for MUE m With UAV n Calculate the line-of-sight propagation probability of the G2A link between them. Probability of non-line-of-sight propagation ; The line-of-sight propagation probability The calculation formula is:

[0011] in, For MUE m With UAV n The angle of elevation between them For MUE m With UAV n The Euclidean distance between them and b For environmental parameters; Indicates MUE m coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis UAV n coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV n The flight altitude, where T represents the matrix bias; The non-line-of-sight propagation probability The calculation formula is: ; Step 212: Calculate MUE separately m With UAV n Line-of-sight path loss Non-line-of-sight path loss The specific calculation formula is as follows:

[0012] in, For signal frequency, At the speed of light, This represents the average additional loss due to free-space propagation loss in a line-of-sight environment. This represents the average additional loss due to free-space propagation loss in non-line-of-sight environments. Step 213: Based on line-of-sight propagation probability Non-line-of-sight propagation probability Line-of-sight path loss Non-line-of-sight path loss Calculate MUE m With UAV n Average path loss between The calculation formula is:

[0013] Step 214: Based on the average path loss Calculate the corresponding channel gain The calculation formula is:

[0014] Step 215, based on channel gain MUE m Transmission power MUE m With UAV n Channel bandwidth during communication and UAV n Noise power at the location Calculate the G2A data transmission rate The calculation formula is:

[0015] in, For UAV n The lowest signal-to-noise ratio of the received signal, when UAV n Signal-to-noise ratio of received signal Not higher than When the UAV is active, it is considered to have established a valid G2A link; when the UAV is active... n Signal-to-noise ratio of received signal Higher than At that time, A value of 0 indicates that a valid G2A link cannot be established. For MUE m and UAV n ,like >0, then , ;in, i The variable representing the traversal of MUE. j The variable representing the traversal of the UAV; Step 216, for MUE m Uninstall to UAV n and the ratio is G2A subtask Based on its task data volume and the calculated G2A data transfer rate Calculate its G2A transmission delay The calculation formula is: .

[0016] Furthermore, the specific process of establishing the A2A communication model in step 2 is as follows: Step 221: For any UAV acting as the sender n With UAV as the receiving party k Calculate the channel gain of the A2A link between them. The calculation formula is:

[0017] in, The channel gain is given at a reference distance of 1 meter. UAV n With UAV k The Euclidean distance between them UAV k coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV kFlight altitude For UAV n With UAV k Path loss coefficient between; Step 222: Based on channel gain UAV n Transmission power UAV n With UAV k Channel bandwidth during communication and UAV k Noise power at the location Calculate the A2A data transmission rate The calculation formula is:

[0018] in, UAV k Signal-to-noise ratio of received signal , For UAV k The lowest signal-to-noise ratio (SNR) of the received signal, when the SNR Not higher than When the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established; when the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established. Higher than At that time, A value of 0 indicates that a valid A2A link cannot be established. Step 223, when UAV n Choose to process the received tasks yourself, that is At that time, the signal-to-noise ratio Consider it as infinity, that is The transmission capacity of this A2A link is considered to be in an ideal state; if >0, then UAV n A2A feasible communication set ; Step 224: Based on A2A data transmission rate Determine UAV n A2A feasible communication set ; Step 225: Place the UAV n A2A subtask forwarded to UAV k The process is based on proportion. ,Proportion Task data volume and A2A data transfer rate Calculate its A2A transmission delay Specifically: when n = k At that time, due to =+∞, =0 indicates that there is no A2A transmission delay in the self-processing part; when n ≠ k and When >0, ; when n ≠ k and When =0, It is considered to be infinite.

[0019] Furthermore, the specific process for establishing the queuing model in step 2 is as follows: Step 231: Assuming the arrival of tasks follows a Poisson process, construct the task processing queues on each UAV as independent M / M / 1-PS queuing models; Step 232: Calculate the UAV based on the task offloading policy of all MUEs and the cooperative forwarding policy of UAVs. n Task arrival rate The calculation formula is:

[0020] in, It is MUE m Uninstall to UAV k The initial task ratio, It is a UAV k MUE m Uninstalled G2A subtasks Forward to adjacent UAV n The proportion to be processed It is a UAV k The set of serviceable MUEs; Step 233: For any MUE m Generated and via UAV n Forwarded to UAV k The A2A subtask for final processing Calculate its performance in UAV k Expected processing latency The calculation formula is: in, It is a UAV k The computing power per unit time It is a UAV k Task completion rate; Step 234: Set the desired processing delay A2A transmission delay Adding them together gives the total completion time of the A2A subtask. The calculation formula is:

[0021] Step 235: For any MUE m Uninstall to UAV n The completion delay of the G2A subtask It depends on the latest completed of all A2A subtasks formed after its secondary split, while also including the initial G2A transmission delay; the calculation formula is:

[0022] Step 236: Calculate MUE m Local processing latency for tasks processed locally The calculation formula is:

[0023] in, It is MUE m Computational power; Step 237, for a complete task Its final completion delay The calculation formula depends on the maximum of the completion latency of all its G2A subtasks and the local processing latency: .

[0024] Furthermore, the specific process of establishing the system energy consumption model in step 2 is as follows: Step 241: Calculate MUE m Total energy consumption generated in completing its own task This includes the total energy consumption for offloading tasks to UAVs and the local computing energy consumption for allocating tasks to local processing; the calculation formula is:

[0025] in, For MUE m The capacitor switching coefficient; Step 242: Calculate UAV n Total energy consumption generated in completing its own task This includes the total energy consumption for forwarding tasks to adjacent UAVs and the total energy consumption for allocating tasks for local processing; the calculation formula is:

[0026] in, For UAV n The capacitor switching coefficient, For UAV n Computational power; Step 243: Sum the total energy consumption of all MUEs and the total energy consumption of all UAVs to obtain the total energy consumption of the mobile edge computing system.

[0027] Furthermore, the process of optimizing the objective in step 2 is as follows: Step 251: Define the joint strategy for the mobile edge computing system. { },in: This represents the set of all task offloading policies for MUE; This represents the set of cooperative forwarding strategies for all UAVs; Step 252: Construct an optimization problem P0 with the goal of minimizing the total system cost. The total system cost is defined as the sum of the weighted sum of the final completion delays of all MUEs and the weighted sum of the total system energy consumption; specifically, it is expressed as follows:

[0028] in, As the system delay weight, As the system energy consumption weight, and ; Step 253: Set the following three types of constraints for the optimization problem:

[0029] Among them, C1 is the task ratio constraint, which ensures that the sum of the task split ratios of any MUE does not exceed 1; C2 is the flow conservation constraint, which ensures that tasks offloaded to UAV are fully processed or forwarded; C3 is the system stability constraint, which ensures that the task arrival rate of each UAV does not exceed its maximum service capacity.

[0030] Furthermore, step 3 specifically includes: Step 31: Model the optimization problem as a multi-leader, multi-follower Stackelberg game. In this Stackelberg game architecture, the UAV, as the leader, prioritizes formulating a cooperative forwarding strategy to guide the network's load distribution; the MUE, as the follower, makes its optimal task offloading strategy after observing the UAV's cooperative forwarding strategy. The Stackelberg game is defined as follows: G ={ } in, The set of leaders, i.e., the set of all UAVs; The set of followers, that is, the set of all MUEs; For leaders The collaborative forwarding strategy, For followers Task uninstallation strategy For leaders utility function For followers Utility function; Step 32: Define the Leader utility function for:

[0031] in, Including the leader A collaborative forwarding strategy involving all leaders outside of [the previous one]. This indicates the task unloading policy for all followers; Step 33: Define Followers utility function for:

[0032] in, Including followers The task unloading strategy for all followers, This indicates a collaborative forwarding strategy among all leaders; Step 34, appoint the leader With followers The optimization problems are defined as follows:

[0033] Step 35: Define the Stackelberg equilibrium as: a strategy group ( ) constitutes a game The Stackelberg equilibrium for any leader With followers It should meet the following two conditions: (a)

[0034] (b)

[0035] in, This represents the balanced forwarding strategy for all leaders. This represents the balanced unloading strategy for all followers. Indicates the leader Balanced forwarding strategy, Including the leader A balanced forwarding strategy for all leaders except those in leadership positions; followers Balanced offloading strategy, Including followers A balanced unloading strategy for all followers except those in the same category; Condition (a) is the Nash equilibrium of the leader subgame, representing the equilibrium unloading strategy given all followers. and except leaders Balanced forwarding strategy for all leaders except There is no leader. n It can achieve greater efficiency by unilaterally changing its own strategy; condition (b) is the Nash equilibrium of the follower subgame, representing the equilibrium forwarding strategy given all leaders. and followers Balanced forwarding strategy for all followers except Below, there were no followers. m It can achieve greater efficiency by unilaterally changing its own strategy.

[0036] Furthermore, step 4 specifically includes: The distributed iterative strategy solution algorithm includes: Algorithm 1, transforming the MUE offloading decision problem into a parameter optimization problem by introducing an auxiliary variable with a delay upper limit, and equating the offloading decision problem to a standard continuous knapsack problem with bin constraints when given the auxiliary variable, and obtaining the task offloading strategy by using a greedy allocation strategy based on unit energy consumption cost sorting; Algorithm 2, using the golden section search method on the auxiliary variable with the delay upper limit to obtain the optimal task offloading strategy for MUE; Algorithm 3, when updating the task offloading strategy for MUE and the cooperative forwarding strategy for UAV, introducing a near-end regularization term to strongly concave the objective function, and obtaining the optimal task offloading strategy and the optimal cooperative forwarding strategy as the balanced offloading strategy and the balanced forwarding strategy.

[0037] Furthermore, Algorithm 1 specifically includes the following steps: (1) For MUE m optimization problem Introduce an auxiliary variable for the upper limit of latency. This represents the upper limit of the latency for the completion of all its subtasks, which will optimize this problem. Transform into about Parameter optimization problem It is represented as follows:

[0038] The constraints include: task ratio constraint C1, system stability constraint C3, and latency upper limit constraint C4: for all feasible unloading paths, Total completion time G2A transmission delay The sum shall not exceed And latency upper limit constraint C5: MUE m Local processing latency for tasks processed locally Not exceeding ; (2) When Given the parameter optimization problem The problem is transformed into a standard continuous knapsack problem with box constraints. This standard continuous knapsack problem uses a greedy strategy based on sorting by unit energy cost to obtain the optimal solution, as follows: (a) Initialize the task proportion of the local computing portion = For each drone Initialize its maximum allocable task ratio. =1; (b) Iterate through each drone : For drones n The corresponding A2A feasible communication set Each drone in k : If the drone's cooperative forwarding strategy is in Then, the positive real roots are solved according to the preset constraint equations. and update = ; (c) Determine whether the sum of the maximum percentages of all available tasks for all drones currently meets the unloading requirements: like If it is not feasible, then it is determined to be infeasible and a feasibility flag is set. Otherwise, proceed to step (d); (d) Calculate the energy consumption cost of each unit task performed by the local machine and each UAV via the G2A link. After sorting the tasks in ascending order of energy consumption cost, prioritize allocating the task proportion to the G2A link with the lowest energy consumption cost. During the allocation process, the allocation shall not exceed the upper limit of the allocable task proportion for each UAV, until the total task proportion reaches 1, and generate a task offloading strategy. And set feasibility indicators. Calculate total energy consumption ; (e) Output the feasibility flag Feasible and the task unloading strategy. Total energy consumption ; For Algorithm 2, the specific steps are as follows: Solving the parameter optimization problem using the golden section search method The details are as follows: (a) Definition feasible region , This represents the theoretical fastest time when all computing resources can be used. Indicates the time required for fully local computation; initialization : = , = And set the golden ratio coefficient ρ; (b) Perform an iterative search within the feasible region until the length of the feasible region meets the accuracy requirements: Calculate the two golden section points within the feasible region: , ; by and Using the time delay parameter, Algorithm 1 is called to obtain the feasibility flag. , and the corresponding total energy consumption and ; (c) Update the feasible region interval based on the feasibility indicators and the objective function value: like If it is False, then update. = ; like If it is False, then update. = ; like and If both are True, then compare the weighted objective function values: like Then update = Otherwise, update. = ; (d) Repeat steps (b) to (c) until the length of the feasible region is | | greater than the precision threshold δ; (e) Take the value from the current feasible region interval ( + Using 2 / 2 as the final delay parameter, Algorithm 1 is called again to obtain the optimal task unloading strategy. ; (f) Output the optimal task unloading strategy ; Algorithm 3 specifically includes the following steps: A hybrid distributed offloading and cooperative strategy iterative algorithm is used to achieve the game equilibrium between the leader and followers, as detailed below: (a) Initialization: Set the iteration count t=1; for all mobile user equipment Initialize its task unloading strategy Local computing mode; for all drones ∈ Initialize its cooperative forwarding strategy Local processing mode; (b) Iterative optimization until convergence conditions are met: 1) For each mobile user equipment Based on current environmental information Calling Algorithm 2 yields... To the optimized task unloading strategy The current task unloading strategy is updated using an inertial weighted approach. :

[0039] in, For inertial weights, Unload the strategy for the previous task; According to the updated task uninstallation policy Update the mission arrival rate of all drones; 2) For each drone ∈ Based on the previous drone... Coordination with other drones Forwarding strategy and the task unloading strategy for all current followers Solve for optimal solutions with proximal regularization terms. The modified collaborative forwarding strategy :

[0040] in, Represents the regularization parameter. Indicates the last time a drone The collaborative forwarding strategy; The current cooperative forwarding strategy is updated using inertial weighting. : ; Update the mission arrival rate of all drones according to the updated collaborative forwarding strategy; 3) Order ; (c) Stop iteration if the following condition is met: ≤μ and , μ; where μ represents the convergence accuracy; (d) Output the current optimal cooperative forwarding strategy. Task Unloading Strategy As a balanced forwarding strategy and a balanced offloading strategy.

[0041] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: This invention addresses mobile edge computing systems comprised of multiple drones and multiple mobile user devices (MUEs). It allows MUEs to access the system from multiple points within overlapping drone coverage areas, proportionally splitting individual computing tasks and offloading them in parallel across multiple drones. This improves communication and computing resource utilization and reduces end-to-end latency. Furthermore, it allows drones to further split received offloaded tasks and forward them to adjacent low-load drones for collaborative computing, achieving load balancing and improved service stability. Based on this, a system cost model is established, and the interaction between the task offloading strategy of the MUEs and the collaborative forwarding strategy of the drones is modeled as a Stackelberg game with multiple leaders and multiple followers: the UAVs, as leaders, prioritize formulating collaborative forwarding strategies to guide network load distribution, while the MUEs, as followers, make their optimal task offloading strategy after observing the collaborative forwarding strategies. This approach ensures system stability and service quality while reducing overall latency and energy consumption. Simultaneously, a distributed iterative solution algorithm is proposed. On the follower side, by introducing a delay upper limit auxiliary variable, the optimization problem is transformed into a continuous knapsack problem with bin constraints. A greedy strategy combined with golden section search is used to efficiently solve the unloading strategy. On the leader side, a proximal term is introduced to strongly concave the problem to ensure iterative stability. An interior point method is used to solve its optimal task unloading strategy. Finally, alternating updates are performed between the MUE and UAV layers to obtain the balanced forwarding strategy and balanced unloading strategy under game equilibrium, thus achieving load balancing. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is an execution flowchart of a computational offloading and collaborative computation method based on Stackelberg game in an unmanned aerial vehicle (UAV) according to an embodiment of the present invention.

[0044] Figure 2 This is an execution flowchart of the distributed iterative strategy solving algorithm provided in this embodiment of the invention.

[0045] Figure 3 This is a schematic diagram of a multi-UAV-assisted mobile edge computing system provided in an embodiment of the present invention. Detailed Implementation

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figures 1-3 The present invention provides a computational offloading and collaborative computation method based on Stackelberg game in a drone, comprising the following steps: Step 1: Construct a mobile edge computing system consisting of multiple MUEs (Mobile User Equipment) and multiple UAVs (Unmanned Aerial Vehicles); and define the task offloading strategy for MUEs and the collaborative forwarding strategy for UAVs; In this embodiment, step 1 specifically includes: Step 11: Construct a mobile edge computing system consisting of N UAVs and M MUEs. If the ground base station is damaged and unable to provide services, the UAVs are deployed in the airspace above the target area to provide temporary communication coverage and computing services to ground users. Step 12: Define the set of UAVs as follows The set of MUEs is ,in, n Indicates the index of the UAV. m Indicates the index of MUE, MUE m Indicates the first m MUE, UAV n Indicates the first n UAV; assuming MUE m One task (computationally intensive and latency-sensitive) is generated in each unit of time, using triples. It means that, among them, Indicates the amount of task data. This indicates the number of CPU cycles required to process a unit of data. This indicates the total number of computation cycles required for the task. The MUE can transmit tasks to the UAV for processing via a ground-to-air (G2A) link to reduce processing latency. Simultaneously, the UAVs can also collaborate via an air-to-air (A2A) link, further splitting received tasks and forwarding them to adjacent UAVs for collaborative computation, further reducing processing latency and alleviating single-machine congestion.

[0048] Step 13: Define MUE m The G2A feasible communication set (the G2A feasible communication set represents the set of communication that can be used with MUE) m The set of UAVs communicating via G2A (Ground to Air) links is UAV n The set of serviceable MUEs (the set of serviceable MUEs represents UAVs) n The set of MUEs responsible for the service is UAV n The set of feasible A2A communications (the set of feasible A2A communications represents the communication between the UAV and the UAV) n A collection of UAVs that cooperate via A2A (Air-to-Air) links. ; Step 14: Define MUE m The task uninstallation strategy is ,in, Indicates MUE m The task Uninstall to UAV n The proportion to be processed is used to form G2A subtasks. ; Indicates task The proportion processed locally; the task offloading strategy. Satisfying the task ratio constraint: for any MUE m The sum of the task ratios unloaded to each UAV and the locally calculated ratios is 1, and the ratios of each task are all within the range of [0,1].

[0049] Define UAV n The collaborative forwarding strategy is ,in, , UAV n G2A subtask Forward to adjacent UAV k The proportion of tasks to be processed is used to form A2A subtasks. ;when hour, UAV n The proportion of tasks processed locally; the collaborative forwarding strategy. Satisfying the flow conservation constraint: for any MUE m Uninstall to UAV n The G2A subtask, after being split twice on the UAV side, has a total forwarding ratio of 1 to each UAV, and each forwarding ratio is within the range of [0,1].

[0050] Step 15: Based on the defined task offloading and collaborative forwarding strategies, perform task processing in two phases: G2A offloading phase: MUE m The task is split according to the task offloading strategy, and divided into two parts according to a preset ratio. One part of the task is a G2A subtask, which is offloaded to the corresponding UAV for processing via the G2A link; the other part of the task is processed locally; this is called MUE. m The generated tasks will be handled according to the task unloading policy. Uninstall to Tasks that are not uninstalled will be processed locally; A2A Collaboration Phase: UAV n According to the cooperative forwarding strategy, each G2A subtask received is further split into A2A subtasks, and then forwarded to the corresponding adjacent UAV via the A2A link. k Or local processing; i.e., UAV n Received After each MUE unloads its G2A subtask, it proceeds according to the collaborative forwarding strategy. The G2A subtask is further split, and the resulting A2A subtask is forwarded to... One or more UAVs in the system perform collaborative computing, or the computing is performed by UAVs. n Retain local calculations.

[0051] Step 2: Establish a system cost model, which includes at least a task completion delay model and a system energy consumption model. The task completion delay model includes a G2A communication model, an A2A communication model, and a computation queuing model. Construct an optimization problem with minimizing the total system cost as the optimization objective. The positional information of MUE and UAV is represented using three-dimensional Cartesian coordinates. m and UAV n The coordinates are respectively represented as and To avoid co-channel interference, it is assumed that the system uses orthogonal spectrum resource allocation.

[0052] In this embodiment, the specific process of establishing the G2A communication model in step 2 is as follows: Considering the complexity of the ground environment and potential obstacle occlusion, a probabilistic path loss model is used to model G2A communication. This model takes into account the probability of occurrence and path loss of line-of-sight (LoS) and non-line-of-sight (None-LoS) channels.

[0053] Step 211, for MUE m With UAV nCalculate the line-of-sight propagation probability of the G2A link between them. Probability of non-line-of-sight propagation ; The line-of-sight propagation probability The calculation formula is:

[0054] in, For MUE m With UAV n The angle of elevation between them For MUE m With UAV n The Euclidean distance between them and b For environmental parameters; Indicates MUE m coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis UAV n coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV n The flight altitude, where T represents the matrix bias; The non-line-of-sight propagation probability The calculation formula is: ; Step 212: Calculate MUE separately m With UAV n Line-of-sight path loss Non-line-of-sight path loss The specific calculation formula is as follows:

[0055] in, For signal frequency, At the speed of light, This represents the average additional loss due to free-space propagation loss in a line-of-sight environment. This represents the average additional loss due to free-space propagation loss in non-line-of-sight environments. Step 213: Based on line-of-sight propagation probability Non-line-of-sight propagation probability Line-of-sight path loss Non-line-of-sight path loss Calculate MUE m With UAV n Average path loss between The calculation formula is:

[0056] Step 214: Based on the average path loss Calculate the corresponding channel gain The calculation formula is:

[0057] Step 215, based on channel gain MUE m Transmission power MUE m With UAV n Channel bandwidth during communication and UAV n Noise power at the location Calculate the G2A data transmission rate The calculation formula is:

[0058] in, For UAV n The lowest signal-to-noise ratio of the received signal, when UAV n Signal-to-noise ratio of received signal Not higher than When the UAV is active, it is considered to have established a valid G2A link; when the UAV is active... n Signal-to-noise ratio of received signal Higher than At that time, A value of 0 indicates that a valid G2A link cannot be established. For MUE m and UAV n ,like >0, then , ;in, i The variable representing the traversal of MUE. j The variable representing the traversal of the UAV; Step 216, for MUE m Uninstall to UAV n and the ratio is G2A subtask Based on its task data volume and the calculated G2A data transfer rate Calculate its G2A transmission delay The calculation formula is: .

[0059] In this embodiment, the specific process of establishing the A2A communication model in step 2 is as follows: Since UAVs hover in the air, their communication links with other UAVs are rarely blocked by obstacles. Therefore, the A2A channel can be reasonably simplified to a LoS channel.

[0060] Step 221: For any UAV acting as the sender n With UAV as the receiving party k Calculate the channel gain of the A2A link between them. The calculation formula is:

[0061] in, The channel gain is given at a reference distance of 1 meter. UAV n With UAV k The Euclidean distance between them UAV k coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV k Flight altitude For UAV n With UAV k Path loss coefficient between; Step 222: Based on channel gain UAV n Transmission power UAV n With UAV k Channel bandwidth during communication and UAV k Noise power at the location Calculate the A2A data transmission rate The calculation formula is:

[0062] in, UAV k Signal-to-noise ratio of received signal , For UAV k The lowest signal-to-noise ratio (SNR) of the received signal, when the SNR Not higher than When the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established; when the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established. Higher than At that time, A value of 0 indicates that a valid A2A link cannot be established. Step 223, when UAV n Choose to process the received tasks yourself, that is At that time, the signal-to-noise ratio Consider it as infinity, that is The transmission capacity of this A2A link is considered to be in an ideal state; if >0, then UAV n A2A feasible communication set ; Step 224: Based on A2A data transmission rate Determine UAV n A2A feasible communication set ; Step 225: Place the UAV n A2A subtask forwarded to UAV k The process is based on proportion. ,Proportion Task data volume and A2A data transfer rate Calculate its A2A transmission delay Specifically: when n = k At that time, due to =+∞, =0 indicates that there is no A2A transmission delay in the self-processing part; when n ≠ k and When >0, ; when n ≠ k and When =0, It is considered to be infinite.

[0063] In this embodiment, the specific process of establishing the queuing model in step 2 is as follows: Step 231: Assuming that the arrival of tasks is a Poisson process, construct the task processing queue on each UAV as an independent M / M / 1-PS queuing model to describe the computational congestion and queuing delay caused by the random arrival of tasks. Step 232: Based on the task offloading policy of all MUEs and the collaborative forwarding policy of UAVs, since the G2A subtasks offloaded from the MUE to the UAV will be further split into A2A subtasks by the UAV and forwarded to the adjacent UAV for processing, the tasks actually processed in the UAV are all regarded as A2A subtasks. n Task arrival rate For the reason The total number of computation cycles required for all forwarded A2A subtasks, and the calculation of UAV. n Task arrival rate The calculation formula is:

[0064] in, It is MUE m Uninstall to UAV k The initial task ratio, It is a UAV k MUE m Uninstalled G2A subtasks (MUE) m The task Uninstall to UAV k Processing is performed to form G2A subtasks. Forward to adjacent UAV n The proportion to be processed It is a UAV k The set of serviceable MUEs; Step 233: For any MUE m Generated and via UAV n Forwarded to UAV k The A2A subtask for final processing Calculate its performance in UAV k Expected processing latency The calculation formula is: in, It is a UAV k The computing power per unit time; It is a UAV k Task completion rate; Step 234: Since the size of the task's execution result is usually much smaller than the input data, the latency required for sending back the result data after the task is processed is ignored. Therefore, the expected processing latency is... A2A transmission delay Adding them together gives the total completion time of the A2A subtask. The calculation formula is:

[0065] Step 235: For any MUE m Uninstall to UAV n The completion delay of the G2A subtask It depends on the latest completed of all A2A subtasks formed after its secondary split, while also including the initial G2A transmission delay; the calculation formula is:

[0066] Step 236: Calculate MUE m Local processing latency for tasks processed locally The calculation formula is:

[0067] in, It is MUE m Computational power; Step 237, for a complete task Its final completion delay The calculation formula depends on the maximum of the completion latency of all its G2A subtasks and the local processing latency: .

[0068] In this embodiment, the specific process of establishing the system energy consumption model in step 2 is as follows: Step 241: Both task transmission and processing consume energy. Calculate MUE. m Total energy consumption generated in completing its own task This includes the total energy consumption for offloading tasks to UAVs and the local computing energy consumption for allocating tasks to local processing; the calculation formula is:

[0069] in, For MUE m The capacitor switching coefficient, the value of which depends on the type of processor chip used; Step 242: Calculate UAV n Total energy consumption generated in completing its own task This includes the total energy consumption for forwarding tasks to adjacent UAVs and the total energy consumption for allocating tasks for local processing; the calculation formula is:

[0070] in, For UAV n The capacitor switching coefficient, For UAV nComputational power; Step 243: Sum the total energy consumption of all MUEs and the total energy consumption of all UAVs to obtain the total energy consumption of the mobile edge computing system.

[0071] In this embodiment, the process of optimizing the target in step 2 is as follows: The optimization objective is to minimize the total system cost in order to achieve Pareto optimality in terms of computational efficiency and resource efficiency.

[0072] Step 251: Define the joint strategy for the mobile edge computing system. { },in: This represents the set of all task offloading policies for MUE. This represents the set of cooperative forwarding strategies for all UAVs; Step 252: Construct an optimization problem P0 with the goal of minimizing the total system cost. The total system cost is defined as the sum of the weighted sum of the final completion delays of all MUEs and the weighted sum of the total system energy consumption; specifically, it is expressed as follows:

[0073] in, As the system delay weight, As the system energy consumption weight, and ; Step 253: Set the following three types of constraints for the optimization problem:

[0074] Among them, C1 is the task ratio constraint, which ensures that the sum of the task split ratios of any MUE does not exceed 1; C2 is the flow conservation constraint, which ensures that the tasks offloaded to UAV are fully processed or forwarded; C3 is the system stability constraint, which ensures that the task arrival rate of each UAV does not exceed its maximum service capacity, thus ensuring the stability of the system.

[0075] Step 3: Based on the interaction between the cooperative forwarding strategy of UAVs and the task offloading strategy of MUEs, the optimization problem is modeled as a Stackelberg game with multiple leaders and multiple followers; where each UAV is a leader and each MUE is a follower. Solving the constructed optimization problem P0 faces both mathematical and engineering challenges. Mathematically, the objective function exhibits bilinear coupling terms between policy variables; specifically, the UAV task arrival rate is determined by the product of the MUE offloading ratio and the UAV forwarding ratio. This results in a bilinear fractional structure for task processing latency, with its Hessian matrix being indeterminate within the domain, thus proving the objective function is non-convex. Furthermore, from an engineering implementation perspective, implementing a centralized solution to the optimization problem P0 requires the central controller to acquire real-time information from the entire network, potentially leading to unacceptable signaling overhead. Therefore, directly seeking a universally optimal solution becomes challenging.

[0076] In this embodiment, step 3 specifically includes: Step 31: Model the optimization problem as a multi-leader, multi-follower Stackelberg game, decomposing the complex global optimization problem into individual optimization subproblems for each participant based on local information. In this Stackelberg game architecture, the UAV, as the leader, prioritizes formulating a cooperative forwarding strategy to guide the network's load distribution; the MUE, as a follower, makes its optimal task offloading strategy after observing the UAV's cooperative forwarding strategy. The Stackelberg game is defined as follows: G ={ } in, The set of leaders, i.e., the set of all UAVs; The set of followers, that is, the set of all MUEs; For leaders The collaborative forwarding strategy, For followers Task uninstallation strategy For leaders utility function For followers Utility function; Step 32: Considering that leaders focus on the overall quality of task completion, define the leader. utility function for:

[0077] in, Including the leader A collaborative forwarding strategy involving all leaders outside of [the previous one]. This indicates the task unloading policy for all followers; Step 33: Define Followers utility function for:

[0078] in, Including followers The task unloading strategy for all followers, This indicates a collaborative forwarding strategy among all leaders; Step 34: In the constructed Stark game, both the leader and followers are rational and selfish, and their goal is to maximize their own utility function. Therefore, the leader... With followers The optimization problems are defined as follows:

[0079] Based on the aforementioned individual optimization problem, the stable state of the system is described by Stackelberg equilibrium. In this state, the leader formulates the optimal cooperative forwarding strategy, and the followers, upon observing this cooperative forwarding strategy, adopt the best-responding task offloading strategy. Furthermore, any participant that unilaterally deviates from its current strategy cannot obtain higher utility.

[0080] Step 35: Define the Stackelberg equilibrium (SE) as: a strategy group ( ) constitutes a game The Stackelberg equilibrium for any leader With followers It should meet the following two conditions: (a)

[0081] (b)

[0082] in, This represents the balanced forwarding strategy for all leaders. This represents the balanced unloading strategy for all followers. Indicates the leader Balanced forwarding strategy, Including the leader A balanced forwarding strategy for all leaders except those in leadership positions; followers Balanced offloading strategy, Including followers A balanced unloading strategy for all followers except those in the same category; Condition (a) is the Nash equilibrium of the leader subgame, representing the equilibrium unloading strategy given all followers. and except leaders Balanced forwarding strategy for all leaders except There is no leader. nIt can achieve greater efficiency by unilaterally changing its own strategy; condition (b) is the Nash equilibrium of the follower subgame, representing the equilibrium forwarding strategy given all leaders. and followers Balanced forwarding strategy for all followers except Below, there were no followers. m It can achieve greater efficiency by unilaterally changing its own strategy.

[0083] Step 4: Based on Stackelberg game theory, a distributed iterative strategy algorithm is used to alternately update the cooperative forwarding strategy and the task offloading strategy until the convergence condition is met, and output the balanced offloading strategy and the balanced forwarding strategy. The existence of SE has been proven above, and for each participant, their utility function is a concave function. Solving for SE... , The corresponding results can be obtained using a general solver. However, considering the high startup overhead of calling the solver, for the low-performance MUE layer, the problem... Based on the structural properties, an unloading strategy solution algorithm based on analytical solutions was designed to eliminate the overhead required when calling the solver.

[0084] In this embodiment, step 4 specifically includes: The distributed iterative strategy solution algorithm includes: Algorithm 1, transforming the MUE offloading decision problem into a parameter optimization problem by introducing an auxiliary variable of the latency upper limit, and equating the offloading decision problem to a standard continuous knapsack problem with bin constraints when given the auxiliary variable, and obtaining the task offloading strategy by using a greedy allocation strategy based on unit energy consumption cost sorting; Algorithm 2, using the golden section search method on the auxiliary variable of the latency upper limit to obtain the optimal task offloading strategy for MUE; Algorithm 3, when updating the task offloading strategy for MUE and the cooperative forwarding strategy for UAV, introducing a near-end regularization term to strongly concave the objective function, obtaining the optimal task offloading strategy and the optimal cooperative forwarding strategy as the balanced offloading strategy and balanced forwarding strategy, so as to improve the stability of distributed iterative update and ensure convergence; For Algorithm 1, the specific steps are as follows: (1) For MUE m optimization problem To eliminate potential non-smoothness in the objective function, an auxiliary variable for the upper limit of time delay is introduced. This represents the upper limit of the latency for the completion of all its subtasks, which will optimize this problem. Transform into about Parameter optimization problem It is represented as follows:

[0085] The constraints include: task ratio constraint C1, system stability constraint C3, and latency upper limit constraint C4: for all feasible unloading paths, Total completion time G2A transmission delay The sum shall not exceed And latency upper limit constraint C5: MUE m Local processing latency for tasks processed locally Not exceeding ; Analysis of latency upper limit constraints The mathematical structure, which, through algebraic transformations, is applicable to any unloading strategy. This constraint is equivalent to the fractional inequality:

[0086] in, , , .because Constants that are all greater than 0, when When, quadratic equation There are two positive real roots The system stability constraint C3 is equivalent to... It introduces a vertical asymptote constraint. ,because Based on quadratic functions Properties, have Therefore, the feasible region that satisfies the constraints is .when At this time The feasible region that satisfies the constraints must be greater than 1 / 3. The feasible region at that time. Furthermore, the feasible region for local computation is... Solving this parameter optimization problem This needs to be performed at the tightest upper bound that satisfies all constraints. The feasible region of the unloading strategy inequality constraint can be expressed as: .

[0087] (2) When Given the parameter optimization problem The problem is transformed into a standard continuous knapsack problem with box constraints. This standard continuous knapsack problem uses a greedy strategy based on sorting by unit energy cost to obtain the optimal solution, as follows: (a) Initialize the task proportion of the local computing portion = For each drone Initialize its maximum allocable task ratio. =1; (b) Iterate through each drone : For drones n The corresponding A2A feasible communication set Each drone in k : If the drone's cooperative forwarding strategy is in Then, the positive real roots are solved according to the preset constraint equations. and update = ; (c) Determine whether the sum of the maximum percentages of all available tasks for all drones currently meets the unloading requirements: like If it is not feasible, then it is determined to be infeasible and a feasibility flag is set. Otherwise, proceed to step (d). (d) Calculate the energy consumption cost of each unit task performed by the local machine and each UAV via the G2A link. After sorting the tasks in ascending order of energy consumption cost, prioritize allocating the task proportion to the G2A link with the lowest energy consumption cost. During the allocation process, the allocation shall not exceed the upper limit of the allocable task proportion for each UAV, until the total task proportion reaches 1, and generate a task offloading strategy. And set feasibility indicators. Calculate total energy consumption ; (e) Output the feasibility flag Feasible and the task unloading strategy. Total energy consumption .

[0088]

[0089] For Algorithm 2, the specific steps are as follows: We proved in Theorem 1 It is about A strictly convex function, and It is about The function is a linear function, therefore constraint C4 is a strictly convex constraint. Based on the conclusions of perturbation analysis in convex optimization, the optimal value function of a convex problem exhibits convexity with respect to the right-hand side parameters of its constraints. At this point, the energy consumption term... It is about The objective function is a convex function, therefore the objective function about Any local minimum is the global minimum.

[0090] Solving the parameter optimization problem using the golden section search method The details are as follows: (a) Definition feasible region , This represents the theoretical fastest time when all computing resources can be used. Indicates the time required for fully local computation; initialization : = , = And set the golden ratio coefficient ρ; (b) Perform an iterative search within the feasible region until the length of the feasible region meets the accuracy requirements: Calculate the two golden section points within the feasible region: , ; by and Using the time delay parameter, Algorithm 1 is called to obtain the feasibility flag. , and the corresponding total energy consumption and ; (c) Update the feasible region interval based on the feasibility indicators and the objective function value: like If it is False, then update. = ; like If it is False, then update. = ; like and If both are True, then compare the weighted objective function values: like Then update = Otherwise, update. = ; (d) Repeat steps (b) to (c) until the length of the feasible region is | | greater than the precision threshold δ; (e) Take the value from the current feasible region interval ( + Using 2 / 2 as the final delay parameter, Algorithm 1 is called again to obtain the optimal task unloading strategy. ; (f) Output the optimal task unloading strategy .

[0091]

[0092] Algorithm 3 specifically includes the following steps: A Hybrid Distributed Offloading and Collaboration Iterative Algorithm (HDOCIA) is employed to achieve game equilibrium between the leader and followers. This algorithm, based on the block coordinate descent concept, iteratively updates the task offloading strategy of the MUE and the collaborative forwarding strategy of the UAV. Specifically: (a) Initialization: Set the iteration count t=1; for all mobile user equipment Initialize its task unloading strategy Local computing mode; for all drones ∈ Initialize its cooperative forwarding strategy Local processing mode; (b) Iterative optimization until convergence conditions are met: 1) For each mobile user equipment Based on current environmental information Calling Algorithm 2 yields... To the optimized task unloading strategy The current task unloading strategy is updated using an inertial weighted approach. :

[0093] in, For inertial weights, The uninstallation policy is based on the previous task; this is to avoid multiple users... Network oscillations that occur during decision-making, such as rows 3 to 6 in Table 3 of Algorithm 3; According to the updated task uninstallation policy Update the mission arrival rate of all drones; 2) For each drone ∈ Based on the previous drone... Coordination with other drones Forwarding strategy and the task unloading strategy for all current followers Solve for optimal solutions with proximal regularization terms. The modified collaborative forwarding strategy :

[0094] in, Represents the regularization parameter. Indicates the last time a drone The collaborative forwarding strategy; the introduction of Regularization parameters The objective function of the UAV is strongly concave to ensure the stability of the policy; The current cooperative forwarding strategy is updated using inertial weighting. : ; Update the mission arrival rate of all drones according to the updated collaborative forwarding strategy; 3) Order ; (c) Stop iteration if the following condition is met: ≤μ and , μ; where μ represents the convergence accuracy; (d) Output the current optimal cooperative forwarding strategy Task Unloading Strategy As a balanced forwarding strategy and a balanced offloading strategy.

[0095]

[0096] Step 5: Each MUE offloads its tasks to the corresponding UAV according to the load balancing offload strategy, and controls each UAV to forward the received tasks to the corresponding adjacent UAV or local processing according to the load balancing forwarding strategy, so as to achieve collaborative processing of tasks.

[0097] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A computational offloading and collaborative computation method based on Stackelberg game theory in unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: Step 1: Construct a mobile edge computing system consisting of multiple MUEs and multiple UAVs; and define the task offloading strategy for MUEs and the collaborative forwarding strategy for UAVs; Step 2: Establish a system cost model, which includes at least a task completion delay model and a system energy consumption model. The task completion delay model includes a G2A communication model, an A2A communication model, and a computation queuing model. Construct an optimization problem with minimizing the total system cost as the optimization objective. Step 3: Based on the interaction between the cooperative forwarding strategy of UAVs and the task offloading strategy of MUEs, the optimization problem is modeled as a Stackelberg game with multiple leaders and multiple followers; where each UAV is a leader and each MUE is a follower. Step 4: Based on Stackelberg game theory, a distributed iterative strategy algorithm is used to alternately update the cooperative forwarding strategy and the task offloading strategy until the convergence condition is met, and output the balanced offloading strategy and the balanced forwarding strategy. Step 5: Each MUE offloads its tasks to the corresponding UAV according to the load balancing offload strategy, and controls each UAV to forward the received tasks to the corresponding adjacent UAV or local processing according to the load balancing forwarding strategy, so as to achieve collaborative processing of tasks.

2. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, Step 1 specifically includes: Step 11: Construct a mobile edge computing system consisting of N UAVs and M MUEs; Step 12: Define the set of UAVs as follows The set of MUEs is ,in, n Indicates the index of the UAV. m Indicates the index of MUE, MUE m Indicates the first m MUE, UAV n Indicates the first n UAV; assuming MUE m One task is generated in each unit of time, using triples. It means that, among them, Indicates the amount of task data. This indicates the number of CPU cycles required to process a unit of data. Indicates the total number of computation cycles required for the task; Step 13: Define MUE m The feasible G2A communication set is UAV n The set of serviceable MUEs is UAV n The feasible A2A communication set is ; Step 14: Define MUE m The task uninstallation strategy is ,in, MUE m The task Uninstall to UAV n The proportion to be processed is used to form G2A subtasks. ; Indicates task The proportion processed locally; Define UAV n The collaborative forwarding strategy is ,in, , UAV n G2A subtask Forward to adjacent UAV k The proportion of tasks to be processed is used to form A2A subtasks. ;when hour, UAV n The proportion of tasks processed locally; Step 15: Based on the defined task offloading and collaborative forwarding strategies, perform task processing in two phases: G2A offloading phase: MUE m The task is split according to the task unloading strategy and divided into two parts according to the preset ratio. One part of the task is a G2A subtask, which is unloaded to the corresponding UAV for processing through the G2A link; the other part of the task is processed locally. A2A Collaboration Phase: UAV n According to the cooperative forwarding strategy, each G2A subtask received is further split into A2A subtasks, and then forwarded to the corresponding adjacent UAV via the A2A link. k Or it can be processed locally.

3. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, The specific process of establishing the G2A communication model in step 2 is as follows: Step 211, for MUE m With UAV n Calculate the line-of-sight propagation probability of the G2A link between them. Probability of non-line-of-sight propagation ; The line-of-sight propagation probability The calculation formula is: in, For MUE m With UAV n The angle of elevation between them For MUE m With UAV n The Euclidean distance between them and b For environmental parameters; Indicates MUE m coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis UAV n coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV n The flight altitude, where T represents the matrix bias; The non-line-of-sight propagation probability The calculation formula is: ； Step 212: Calculate MUE separately m With UAV n Line-of-sight path loss Non-line-of-sight path loss The specific calculation formula is as follows: in, For signal frequency, At the speed of light, This represents the average additional loss due to free-space propagation loss in a line-of-sight environment. This represents the average additional loss due to free-space propagation loss in non-line-of-sight environments. Step 213: Based on line-of-sight propagation probability Non-line-of-sight propagation probability Line-of-sight path loss Non-line-of-sight path loss Calculate MUE m With UAV n Average path loss between The calculation formula is: Step 214: Based on the average path loss Calculate the corresponding channel gain The calculation formula is: Step 215, based on channel gain MUE m Transmission power MUE m With UAV n Channel bandwidth during communication and UAV n Noise power at Calculate the G2A data transmission rate The calculation formula is: in, For UAV n The lowest signal-to-noise ratio of the received signal, when UAV n Signal-to-noise ratio of received signal Not higher than When the UAV is active, it is considered to have established a valid G2A link; when the UAV is active... n Signal-to-noise ratio of received signal Higher than At that time, A value of 0 indicates that a valid G2A link cannot be established. For MUE m and UAV n ,like >0, then , ;in, i The variable representing the traversal of MUE. j The variable representing the traversal of the UAV; Step 216, for MUE m Uninstall to UAV n and the ratio is G2A subtask Based on its task data volume and the calculated G2A data transfer rate Calculate its G2A transmission delay The calculation formula is: 。 4. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, The specific process of establishing the A2A communication model in step 2 is as follows: Step 221: For any UAV acting as the sender n With UAV as the receiving party k Calculate the channel gain of the A2A link between them. The calculation formula is: in, The channel gain is given at a reference distance of 1 meter. UAV n With UAV k The Euclidean distance between them UAV k coordinates express x Coordinate values ​​on the axis express y Coordinate values ​​on the axis express z The coordinate values ​​on the axis, i.e., UAV k Flight altitude For UAV n With UAV k Path loss coefficient between; Step 222: Based on channel gain UAV n Transmission power UAV n With UAV k Channel bandwidth during communication and UAV k Noise power at Calculate the A2A data transmission rate The calculation formula is: in, UAV k Signal-to-noise ratio of received signal , For UAV k The lowest signal-to-noise ratio (SNR) of the received signal, when the SNR Not higher than When the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established; when the signal-to-noise ratio is 1, it is considered that a valid A2A link has been established. Higher than At that time, A value of 0 indicates that a valid A2A link cannot be established. Step 223, when UAV n Choose to process the received tasks yourself, that is At that time, the signal-to-noise ratio Consider it as infinity, that is The transmission capacity of this A2A link is considered to be in an ideal state; if >0, then UAV n A2A feasible communication set ; Step 224: Based on A2A data transmission rate Determine UAV n A2A feasible communication set ; Step 225: Place the UAV n A2A subtask forwarded to UAV k The process is based on proportion. ,Proportion Task data volume and A2A data transfer rate Calculate its A2A transmission delay Specifically: when n = k At that time, due to =+∞, =0 indicates that there is no A2A transmission delay in the self-processing part; when n ≠ k and When >0, ; when n ≠ k and When =0, It is considered to be infinite.

5. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, The specific process for establishing the queuing model in step 2 is as follows: Step 231: Assuming the arrival of tasks follows a Poisson process, construct the task processing queues on each UAV as independent M / M / 1-PS queuing models; Step 232: Calculate the UAV based on the task offloading policy of all MUEs and the cooperative forwarding policy of UAVs. n Task arrival rate The calculation formula is: in, It is MUE m Uninstall to UAV k The initial task ratio, It is a UAV k MUE m Uninstalled G2A subtasks Forward to adjacent UAV n The proportion to be processed It is a UAV k The set of serviceable MUEs; Step 233: For any MUE m Generated and via UAV n Forwarded to UAV k The A2A subtask for final processing Calculate its performance in UAV k Expected processing latency The calculation formula is: in, It is a UAV k The computing power per unit time; It is a UAV k Task completion rate; Step 234: Set the desired processing delay A2A transmission delay Adding them together gives the total completion time of the A2A subtask. The calculation formula is: Step 235: For any MUE m Uninstall to UAV n The completion delay of the G2A subtask It depends on the latest completed of all A2A subtasks formed after its secondary split, while also including the initial G2A transmission delay; the calculation formula is: Step 236: Calculate MUE m Local processing latency for tasks processed locally The calculation formula is: in, It is MUE m Computational power; Step 237, for a complete task Its final completion delay The calculation formula depends on the maximum of the completion latency of all its G2A subtasks and the local processing latency: 。 6. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, The specific process for establishing the system energy consumption model in step 2 is as follows: Step 241: Calculate MUE m Total energy consumption generated in completing its own task This includes the total energy consumption of offloading and transporting tasks to UAVs and the local computing energy consumption of allocating tasks to local processing. The calculation formula is: in, For MUE m The capacitor switching coefficient; Step 242: Calculate UAV n Total energy consumption generated in completing its own task This includes the total energy consumption for forwarding tasks to adjacent UAVs and the total energy consumption for allocating tasks for local processing; the calculation formula is: in, For UAV n The capacitor switching coefficient, For UAV n Computational power; Step 243: Sum the total energy consumption of all MUEs and the total energy consumption of all UAVs to obtain the total energy consumption of the mobile edge computing system.

7. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, The process of optimizing the objective in step 2 is as follows: Step 251: Define the joint strategy for the mobile edge computing system. { },in: This represents the set of all task offloading policies for MUE; This represents the set of cooperative forwarding strategies for all UAVs; Step 252: Construct an optimization problem P0 with the goal of minimizing the total system cost. The total system cost is defined as the sum of the weighted sum of the final completion delays of all MUEs and the weighted sum of the total system energy consumption; specifically, it is expressed as follows: in, As the system delay weight, As the system energy consumption weight, and ; Step 253: Set the following three types of constraints for the optimization problem: Among them, C1 is the task ratio constraint, which ensures that the sum of the task split ratios of any MUE does not exceed 1; C2 is the flow conservation constraint, which ensures that tasks offloaded to UAV are fully processed or forwarded; C3 is the system stability constraint, which ensures that the task arrival rate of each UAV does not exceed its maximum service capacity.

8. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, Step 3 specifically includes: Step 31: Model the optimization problem as a multi-leader, multi-follower Stackelberg game. In this Stackelberg game architecture, the UAV, as the leader, prioritizes formulating a cooperative forwarding strategy to guide the network's load distribution; the MUE, as the follower, makes its optimal task offloading strategy after observing the UAV's cooperative forwarding strategy. The Stackelberg game is defined as follows: G ={ } in, The set of leaders, i.e., the set of all UAVs; The set of followers, that is, the set of all MUEs; For leaders The collaborative forwarding strategy, For followers Task uninstallation strategy For leaders utility function For followers Utility function; Step 32: Define the Leader utility function for: in, Including the leader A collaborative forwarding strategy involving all leaders outside of [the previous one]. This indicates the task unloading policy for all followers; Step 33: Define Followers utility function for: in, Including followers The task unloading strategy for all followers, This indicates a collaborative forwarding strategy among all leaders; Step 34, appoint the leader With followers The optimization problems are defined as follows: Step 35: Define the Stackelberg equilibrium as: a strategy group ( ) constitutes a game The Stackelberg equilibrium for any leader With followers It should meet the following two conditions: （a） （b） in, This represents the balanced forwarding strategy for all leaders. This represents the balanced unloading strategy for all followers. Indicates the leader Balanced forwarding strategy, Including the leader A balanced forwarding strategy for all leaders except those in leadership positions; followers Balanced offloading strategy, Including followers A balanced unloading strategy for all followers except those in the same category; Condition (a) is the Nash equilibrium of the leader subgame, representing the equilibrium unloading strategy given all followers. and except leaders Balanced forwarding strategy for all leaders except There is no leader. n It can achieve greater efficiency by unilaterally changing its own strategy; condition (b) is the Nash equilibrium of the follower subgame, representing the equilibrium forwarding strategy given all leaders. and followers Balanced forwarding strategy for all followers except Below, there were no followers. m It can achieve greater efficiency by unilaterally changing its own strategy.

9. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 1, characterized in that, Step 4 specifically includes: The distributed iterative strategy solution algorithm includes: Algorithm 1, transforming the MUE offloading decision problem into a parameter optimization problem by introducing an auxiliary variable with a delay upper limit, and equating the offloading decision problem to a standard continuous knapsack problem with bin constraints when given the auxiliary variable, and obtaining the task offloading strategy by using a greedy allocation strategy based on unit energy consumption cost sorting; Algorithm 2, using the golden section search method on the auxiliary variable with the delay upper limit to obtain the optimal task offloading strategy for MUE; Algorithm 3, when updating the task offloading strategy for MUE and the cooperative forwarding strategy for UAV, introducing a near-end regularization term to strongly concave the objective function, and obtaining the optimal task offloading strategy and the optimal cooperative forwarding strategy as the balanced offloading strategy and the balanced forwarding strategy.

10. The computational offloading and collaborative computation method based on Stackelberg game in UAVs as described in claim 9, characterized in that, For Algorithm 1, the specific steps are as follows: (1) For MUE m optimization problem Introduce an auxiliary variable for the upper limit of latency. This represents the upper limit of the latency for the completion of all its subtasks, which will optimize this problem. Transform into about Parameter optimization problem It is represented as follows: The constraints include: task ratio constraint C1, system stability constraint C3, and latency upper limit constraint C4: for all feasible unloading paths, Total completion time G2A transmission delay The sum shall not exceed And latency upper limit constraint C5: MUE m Local processing latency for tasks processed locally Not exceeding ; (2) When Given the parameter optimization problem The problem is transformed into a standard continuous knapsack problem with box constraints. This standard continuous knapsack problem uses a greedy strategy based on sorting by unit energy cost to obtain the optimal solution, as follows: (a) Initialize the task proportion of the local computing portion = For each drone Initialize its maximum allocable task ratio. =1; (b) Iterate through each drone : For drones n The corresponding A2A feasible communication set Each drone in k : If the drone's cooperative forwarding strategy is in Then, the positive real roots are solved according to the preset constraint equations. and update = ; (c) Determine whether the sum of the maximum percentages of all available tasks for all drones currently meets the unloading requirements: like If it is not feasible, then it is determined to be infeasible and a feasibility flag is set. Otherwise, proceed to step (d); (d) Calculate the energy consumption cost of each unit task performed by the local machine and each UAV via the G2A link. After sorting the tasks in ascending order of energy consumption cost, prioritize allocating the task proportion to the G2A link with the lowest energy consumption cost. During the allocation process, the allocation shall not exceed the upper limit of the allocable task proportion for each UAV, until the total task proportion reaches 1, and generate a task offloading strategy. And set feasibility indicators. Calculate total energy consumption ; (e) Output the feasibility flag Feasible and the task unloading strategy. Total energy consumption ; For Algorithm 2, the specific steps are as follows: Solving the parameter optimization problem using the golden section search method The details are as follows: (a) Definition feasible region , This represents the theoretical fastest time when all computing resources can be used. Indicates the time required for fully local computation; initialization : = , = And set the golden ratio coefficient ρ; (b) Perform an iterative search within the feasible region until the length of the feasible region meets the accuracy requirements: Calculate the two golden section points within the feasible region: , ; by and Using the time delay parameter, Algorithm 1 is called to obtain the feasibility flag. , and the corresponding total energy consumption and ; (c) Update the feasible region interval based on the feasibility indicators and the objective function value: like If it is False, then update. = ; like If it is False, then update. = ; like and If both are True, then compare the weighted objective function values: like Then update = Otherwise, update. = ; (d) Repeat steps (b) to (c) until the length of the feasible region is | | greater than the precision threshold δ; (e) Take the value from the current feasible region interval ( + Using 2 / 2 as the final delay parameter, Algorithm 1 is called again to obtain the optimal task unloading strategy. ; (f) Output the optimal task unloading strategy ; Algorithm 3 specifically includes the following steps: A hybrid distributed offloading and cooperative strategy iterative algorithm is used to achieve the game equilibrium between the leader and followers, as detailed below: (a) Initialization: Set the iteration count t=1; for all mobile user equipment Initialize its task unloading strategy Local computing mode; for all drones ∈ Initialize its cooperative forwarding strategy Local processing mode; (b) Iterative optimization until convergence conditions are met: 1) For each mobile user equipment Based on current environmental information Calling Algorithm 2 yields... To the optimized task unloading strategy The current task unloading strategy is updated using an inertial weighted approach. : in, For inertial weights, Unload the strategy for the previous task; According to the updated task uninstallation policy Update the mission arrival rate of all drones; 2) For each drone ∈ Based on the previous drone... Coordination with other drones Forwarding strategy and the task unloading strategy for all current followers Solve for optimal solutions with proximal regularization terms. The modified collaborative forwarding strategy : in, Represents the regularization parameter. Indicates the last time a drone The collaborative forwarding strategy; The current cooperative forwarding strategy is updated using inertial weighting. : ； Update the mission arrival rate of all drones according to the updated collaborative forwarding strategy; 3) Order ; (c) Stop iteration if the following condition is met: ≤μ and , μ; where μ represents the convergence accuracy; (d) Output the current optimal cooperative forwarding strategy Task Unloading Strategy As a balanced forwarding strategy and a balanced offloading strategy.

Images