An unloading method for air-ground collaborative edge computing server

Abstract

The present invention relates to the technical field of electronic digital data processing, and in particular to an unloading method for an air-ground collaborative edge computing server; the method comprises: constructing an edge computing scenario, wherein the edge computing scenario comprises a group of mobile devices and a group of drones, and each mobile device continuously generates computing tasks; obtaining the completion time and energy consumption cost of the mobile device when selecting a local computing task to obtain a first cost; obtaining the unloading completion time and energy consumption of the mobile device when receiving an unloading request; obtaining a second cost based on the unloading completion time and energy consumption; obtaining the remaining power of the current mobile device, the deadline of the computing task, and the local computing capacity to obtain the priority of the current mobile device; constructing a total cost model based on the first cost, the second cost, and the priority of each mobile device, and determining an unloading strategy based on the total cost model, so that the final unloading strategy has a better effect and a low total cost.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to an unloading method for an air-ground collaborative edge computing server. Background Art

[0002] In recent years, with the massive access of IoT devices, some applications in smart terminal devices, such as face recognition, unmanned driving, virtual reality and voice processing, have been widely used. Terminal devices with poor computing power often find it difficult to complete computing-intensive and delay-sensitive tasks in a short period of time. The development of multi-access edge computing is the key to solving such problems. However, edge computing servers are generally installed on geographically fixed cellular ground base stations, which can hardly meet the computing needs of highly mobile terminal devices.

[0003] In order to provide reliable and flexible offloading services, drone-assisted edge computing alleviates the computing offloading pressure of fixed ground servers; however, existing research mostly focuses on deployment optimization to reduce offloading costs, ignoring the variability of device distribution, the limited channel resources of drones and the intelligence of device association. Summary of the invention

[0004] In order to solve the technical problems of inaccurate offloading cost acquisition and poor offloading strategy effect, the purpose of the present invention is to provide an offloading method for an air-ground collaborative edge computing server. The technical solution adopted is as follows:

[0005] An embodiment of the present invention provides a method for uninstalling an air-ground collaborative edge computing server, the method comprising the following steps:

[0006] Building an edge computing scenario, wherein the edge computing scenario includes a group of mobile devices and a group of drones, each of which continuously generates computing tasks;

[0007] Obtaining the completion time and energy consumption cost of the mobile device when selecting a local computing task, and obtaining a first cost of the local computing task based on the completion time and the energy consumption cost; obtaining the number of unloadings of the mobile device when receiving an unloading request, and obtaining the unloading completion time according to the unloading number, data transmission time, and the time when the drone performs the task; obtaining energy consumption based on the transmission power of the mobile device and the computing frequency of the task assigned to the mobile device by the drone; and obtaining a second cost based on the unloading completion time and the energy consumption;

[0008] Acquire the remaining power of the current mobile device, the deadline of the computing task, and the local computing capacity, and obtain the priority of the current mobile device by performing a weighted summation according to the remaining power, the deadline, and the local computing capacity;

[0009] Obtain the drone track, device association status and resource allocation, build a total cost model based on the first cost, the second cost and the priority of each mobile device, and determine the unloading strategy based on the total cost model.

[0010] Preferably, the step of determining the unloading strategy based on the total cost model further includes:

[0011] Constructing a hierarchical reinforcement learning model, wherein the upper layer of the hierarchical reinforcement learning model is the location optimization of multi-UAV deployment, and the lower layer is the unloading optimization strategy;

[0012] The upper layer of the hierarchical reinforcement learning model receives the environment state and generates action output and the next state through a deterministic policy network; the next state is used as the input of the lower layer of the hierarchical reinforcement learning model, and the lower layer outputs the unloading strategy.

[0013] Preferably, the step of acquiring the drone track, device association status, and resource allocation, and building a total cost model in combination with the first cost, the second cost, and the priority of each mobile device includes:

[0014] The total cost model is:

[0015]

[0016] Among them, U represents the track of the UAV; C represents the device association, and α ij indicates that the value is 0 or 1; F indicates resource allocation; w t and w e Respectively represent weights; represents the first cost; represents the second cost; P i Indicates the priority of mobile device i; Indicates the completion time; represents the energy consumption cost; Indicates the time when uninstallation is completed; Indicates energy consumption.

[0017] Preferably, the step of obtaining the completion time and energy consumption cost of the mobile device when selecting a local computing task includes:

[0018] The completion time is the ratio of the computing capacity required by the mobile device to process the task to the computing power of the mobile device;

[0019] An energy coefficient of a central processing unit and a square value of computing power of a mobile device are obtained, and a product of the energy coefficient and the square value is used as the energy consumption cost.

[0020] Preferably, the step of obtaining the unloading completion time according to the unloading quantity, data transmission time and drone mission execution time includes:

[0021] The calculation method of the uninstallation completion time is:

[0022]

[0023] in, Indicates the data transmission time. r ij (t) represents the data transmission rate, D i (t) represents the size of the task data; Indicates the time when the drone performs the mission. f ij (t) Determined by the number of mobile devices connected to the drone, K represents the number of mobile devices that need to be uninstalled.

[0024] Preferably, the step of obtaining energy consumption based on the transmission power of the mobile device and the calculation frequency of the task assigned to the mobile device by the drone comprises:

[0025] Calculating the sum of the transmission energy consumption and the execution energy consumption of all offloaded tasks as the energy consumption;

[0026] The transmission energy consumption is the product of the transmission power of the mobile device and the data transmission time; the execution energy consumption is obtained by the calculation frequency of the task assigned by the drone to the mobile device.

[0027] Preferably, the upper layer of the hierarchical reinforcement learning model optimizes the positions of multiple drone deployments specifically including:

[0028] Status:s t =(D(t),x j (t),y j (t),h j ,x i (t),y i (t),Φ i (t)); where D(t) represents the amount of data within the communication range of the drone group, which is obtained by adding the amount of data from devices within the coverage range of the drone; (x j (t),y j (t),h j ) represents the three-dimensional position of the drone; (x i (t),y i (t)) represents the horizontal position of the mobile device; Φ i (t) indicates the remaining battery power of the mobile device;

[0029] Action: a t={θ j (t),l j (t)}, where 0≤θ j (t)≤2π represents the steering angle of the UAV at time t, l j (t) represents the flight distance of the UAV at time t;

[0030] award:

[0031] Objective: To maximize the amount of missions within the communication range of the drone swarm by optimizing the deployment of drones.

[0032] Preferably, the unloading optimization strategy of the lower layer of the hierarchical reinforcement learning model specifically includes:

[0033] Status:s t+η ={x j (t+1),y j (t+1),h j ,x i (t+1),y i (t+1),P i};

[0034] Action: a t+η ={α ij ,f ij};

[0035] award:

[0036] Target: The optimal offloading strategy is obtained by determining the device association and resource allocation of the output to minimize the system cost.

[0037] Preferably, the method for acquiring the data transmission rate includes:

[0038] Obtain the number of mobile devices currently connected to the drone and the bandwidth, calculate the ratio of the bandwidth to the number of mobile devices currently connected to the drone and record it as a first ratio;

[0039] Obtaining the horizontal distance between the current mobile device and the current drone and the height of the current drone, and calculating the sum of the square value of the horizontal distance and the square value of the height of the drone; obtaining the product result of the first indicator and the transmission power of the current mobile device, and recording the ratio of the product result to the sum result as the second ratio;

[0040] The second ratio is added with 1, and a logarithmic operation is performed on the addition result, and the data transmission rate is obtained by multiplying the logarithmic operation result and the first ratio.

[0041] Preferably, the first indicator is:

[0042]

[0043] Among them, g 0 Represents the channel gain per unit spatial distance; G 0 is a constant; 2 Represents the noise power.

[0044] The present invention has the following beneficial effects:

[0045] In the embodiment of the present invention, the completion time and cost under different associated situations are obtained by various parameters of the mobile device when it is working, so as to obtain the corresponding first cost and second cost. In the acquisition of the first cost and the second cost, the number of tasks and the execution time of the drone and other limitations are considered for analysis, thereby improving the accuracy of the total cost model. In addition, the priority indicator is added to the total cost model. When a conflict occurs in the offloading request, the task deadline and the remaining power and computing power of the mobile device are comprehensively considered for measurement to maximize the number of tasks completed. While ensuring a lower total cost, the efficiency of the task execution process is improved, and the final offloading strategy is more effective and more convincing.

[0046] In addition, the embodiment of the present invention also uses a hierarchical reinforcement learning model to optimize the deployment and unloading strategies of drones, making the overall solution more flexible and adaptable. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of an unloading method of an air-ground collaborative edge computing server provided by an embodiment of the present invention;

[0049] Figure 2 An integrated architecture diagram of deployment and unloading of a multi-drone-assisted edge computing system provided by an embodiment of the present invention;

[0050] Figure 3 An updating process of hierarchical reinforcement learning is provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the unloading method of an air-ground collaborative edge computing server proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0052] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] The following is a detailed description of a specific scheme of an unloading method for an air-ground collaborative edge computing server provided by the present invention in conjunction with the accompanying drawings.

[0054] See also Figure 1 , which shows a flow chart of an unloading method of an air-ground collaborative edge computing server provided by an embodiment of the present invention, the method comprising the following steps:

[0055] Step S100, constructing an edge computing scenario, the edge computing scenario includes a group of mobile devices and a group of drones, and each mobile device continuously generates computing tasks.

[0056] Specifically, we first construct an edge computing scenario as the basis for subsequent analysis. In the embodiment of the present invention, the edge computing scenario includes a mobile device and a drone equipped with an edge server. Figure 2 , which shows an integrated architecture diagram of deployment and offloading of a multi-drone-assisted edge computing system; for easy distinction, each mobile device and drone are numbered, the mobile device is numbered i∈I={1,2,…,N}, N represents the number of all mobile devices; the drone is numbered j∈J={1,2,…,M}, M represents the number of all drones; accordingly, based on the relevant configuration information of the mobile device and the drone, the computing power of each mobile device and the computing power of each drone can be known, and the computing power of mobile device i is recorded as f i loc (t), the computing power of the drone is denoted as f u .

[0057] Each mobile device corresponds to a different application form. The computing task is generated in time slot t. Each computing task includes three configuration files, namely the size of the input task data volume D in bits, the computing power F required to process the task, that is, the number of CPU cycles, and the maximum delay T max ; then the configuration file of each mobile device can be expressed as φ i (t) = {Di (t),F i (t),T i (t)}.

[0058] Step S200, obtain the completion time and energy consumption cost of the mobile device when selecting a local computing task, and obtain the first cost of the local computing task based on the completion time and the energy consumption cost; obtain the number of unloadings of the mobile device when receiving an unloading request, and obtain the unloading completion time according to the unloading number, data transmission time, and drone task execution time; obtain the energy consumption based on the transmission power of the mobile device and the computing frequency of the task assigned to the mobile device by the drone; obtain the second cost based on the unloading completion time and energy consumption.

[0059] Since the height of drone deployment is much higher than that of mobile devices, and the line-of-sight channel and other channel consumption of drone communication links are more significant, other channel consumption such as small-scale fading and shadow loss, the wireless channel between mobile devices and drones can be evaluated using the free space path loss model. The power gain of the wireless channel link between mobile devices and drones is:

[0060]

[0061] Among them, g 0 Represents the channel gain per unit spatial distance; R ij (t) represents the horizontal distance between the mobile device and the drone; h j (t) represents the altitude of the UAV.

[0062] In order to reduce the transmission delay, mobile device i always transmits data at the maximum transmission power to improve the signal-to-noise ratio of the communication link. Therefore, the data transmission rate between mobile device i and the drone is:

[0063]

[0064] Where B represents bandwidth; k represents the number of mobile devices connected to the drone; represents the transmission power of mobile device i; R ij (t) represents the horizontal distance between the mobile device and the drone; h j (t) represents the altitude of the drone; log 2 Indicates the calculation of logarithmic function with base 2; g 0 Represents the channel gain per unit spatial distance, G 0 is a constant, in the embodiment of the present invention, G 0 =2.2846,σ 2 Represents the noise power.

[0065] Based on this, the data transmission rate between each mobile device and the drone is obtained. Then the offloading selection of the mobile device is analyzed to obtain the α ij Indicates uninstall selection, α ij =0 means no unloading task; α ij =1 indicates that there is an unloading task; for α ij =0 and α ij = 1 are analyzed separately.

[0066] Specifically, when α ij =0, the mobile device selects the local computing task. At this time, the cost of the mobile device's computing is obtained by the completion time and energy consumption cost of the local computing task. The specific calculation of the completion time is:

[0067]

[0068] Among them, F i (t) The computing power required by mobile device i to process the task; f i loc (t) represents the computing power of mobile device i.

[0069] Correspondingly, the specific calculation of the energy consumption cost of a mobile device performing local computing tasks is:

[0070]

[0071] Among them, k i Indicates the energy coefficient of the central processing unit.

[0072] Therefore, the corresponding first cost can be obtained according to the completion time and energy consumption cost when the mobile device selects the local computing task. The specific calculation of the first cost is:

[0073]

[0074] Among them, w t represents the weight of the completion time; w e It represents the weight of energy consumption cost. The specific value is set by the implementer.

[0075] Furthermore, when α ij =1, it indicates that the mobile device receives an offload request. At this time, the offload situation needs to be considered. While avoiding channel waste, the offload strategy when a conflict occurs should also be considered. The offload cost at this time is related to the offload completion time and energy consumption of the offload calculation.

[0076] OFDM technology is applied in the embodiment of the present invention, which means that due to the limited number of subcarriers and the limitation of the task transmission time threshold, the maximum number of tasks that each UAV can receive in each time slot t is V. max, so the following relationship exists:

[0077]

[0078] Since the backhaul task is small, the time required can be ignored. Therefore, the offloading completion time of the offloading calculation consists of two parts: the data transmission time and the time required for the UAV to perform the task. The calculation of the offloading completion time is:

[0079]

[0080] in, Indicates the data transmission time. Indicates the time when the drone performs the mission. f ij (t) Determined by the number of mobile devices connected to the drone, K represents the number of mobile devices that need to be uninstalled.

[0081] The energy consumption of completing the offloading task also consists of two parts, namely transmission energy consumption and execution energy consumption. Therefore, the specific calculation of energy consumption is:

[0082]

[0083] in, represents the transmission energy consumption, represents the transmission power of mobile device i; represents the execution energy consumption, f ij (t) represents the computation frequency of the task assigned by the UAV to mobile device i.

[0084] Thus, the second cost can be obtained according to the unloading completion time and energy consumption, and the calculation of the second cost is:

[0085]

[0086] Among them, w t represents weight; w e Represents the weight, and the specific value is set by the implementer.

[0087] Step S300, obtaining the remaining power of the current mobile device, the deadline of the computing task, and the local computing capacity, and performing weighted summation according to the remaining power, the deadline, and the local computing capacity to obtain the priority of the current mobile device.

[0088] Since the communication resources of the drone are limited, when all channels are unloaded by mobile devices, the drone resources are effectively utilized, and when the channels are idle, resource waste will occur; however, when there are too many unloading tasks to be unloaded, certain conflicts will occur. At this time, it is necessary to consider which tasks to be unloaded. Therefore, the embodiment of the present invention sets a corresponding priority for each randomly generated task. When the number of mobile devices to be unloaded is greater than the maximum unloading number of the drone, the priority is added to the cost function, and the unloading task with a larger priority is considered first; when unloading conflicts occur, unloading is mainly divided into three situations according to the order of priority:

[0089] (1) The number of mobile devices within the communication range of the drone is less than or equal to the maximum unloading number of the drone. At this time, the first cost is compared with the second cost, and the method with the smaller cost is selected to calculate the task.

[0090] (2) The number of mobile devices within the communication range of the drone is greater than the maximum number of mobile devices to be unloaded to the drone, but the number of mobile devices that choose to be unloaded to the drone is less than or equal to the maximum number of mobile devices to be unloaded. Since the system will choose a method with a lower cost to calculate the task, there will be no unloading conflict at this time.

[0091] (3) The number of mobile devices within the communication range of the drone is greater than the maximum number of devices that can be unloaded to the drone. The number of devices that choose to unload to the drone is greater than the maximum number of devices that can be unloaded. In this case, an unloading conflict will occur. Priority is set for each mobile device, taking into account the computing power of the mobile device and the time sensitivity of the task.

[0092] In the process of task calculation offloading, the local computing power of the mobile device and the delay requirement of the task play a decisive role in the association of the mobile device, and the remaining power determines whether the task can be successfully offloaded before the local power is exhausted. Therefore, in the embodiment of the present invention, the remaining power of the mobile device, the maximum deadline of the task and the local computing power are used as the basis for measuring the priority of the mobile device. The specific calculation of the priority is:

[0093]

[0094] Among them, P i represents the priority of mobile device i; f i loc Indicates local computing power; Indicates the maximum deadline of the task; Φ i Indicates the remaining power of mobile device i; ω 1 and ω 2 They represent the weight factors corresponding to local computing power and deadline respectively, which are set by the implementer.

[0095] Priority P iThe smaller it is, the higher the probability that the local computing task cannot be completed, and the more it should be offloaded first.

[0096] Step S400, obtaining the drone track, device association status and resource allocation, building a total cost model based on the first cost, the second cost and the priority of each mobile device, and determining the unloading strategy based on the total cost model.

[0097] The first cost and the second cost are obtained in step S200, and the priority is obtained in step S300. Then, a total cost model is constructed according to the first cost, the second cost and the priority. Specifically, the total cost model is:

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] Among them, U represents the track of the drone; C represents device association, indicating whether the drone selects user access. The specific association is represented by α ij represents; F represents resource allocation, the computational completion time and energy consumption of the task are related to U, C and F, and the total cost function is minimized by optimizing these three variables; α ij (t)={0,1}, j∈J,t∈T represents the selection of mobile device association; This means that each mobile device can only be unloaded to one drone; Indicates that the number of tasks to be uninstalled must be less than the maximum number of tasks to be uninstalled from the server; It means that the computing power allocated by the drone to the mobile device task is less than the total computing power of the drone; Indicates that the calculation time of the task must be less than the maximum deadline, otherwise the task calculation fails.

[0105] Furthermore, in order to ensure the rationality and accuracy of strategy formulation, the embodiment of the present invention also includes continuous updating using hierarchical reinforcement learning, please refer to Figure 3 , which shows an update process of hierarchical reinforcement learning; that is, the original deployment and unloading of edge computing servers are decomposed into two sub-problems, one is the deployment of edge computing servers, that is, the location optimization problem, and the other is the unloading optimization problem, as follows:

[0106] P1:

[0107] P2:

[0108] Two issues are analyzed separately:

[0109] First, the location optimization problem of multi-UAV deployment in the upper layer is analyzed; since the transmission rate in the communication model and the transmission time in the calculation model are both related to the position between the UAV and the device, the unloading optimization strategy of the lower layer can only be stably trained when the deployment of the upper-layer UAVs converges to the optimal or suboptimal state.

[0110] Status:s t =(D(t),x j (t),y j (t),h j ,x i (t),y i (t),Φ i (t))

[0111] Where D(t) represents the amount of data within the communication range of the drone group, which is obtained by adding the amount of data from devices within the coverage range of the drones; (x j (t),y j (t),h j ) represents the three-dimensional position of the drone; (x i (t),y i (t)) represents the horizontal position of the mobile device; Φ i (t) indicates the remaining battery power of the mobile device.

[0112] Action: a t ={θ j (t),l j (t)}, where 0≤θ j (t)≤2π represents the steering angle of the UAV at time t, l j (t) represents the flight distance of the UAV at time t.

[0113] award:

[0114] Objective: To maximize the amount of missions within the communication range of the drone swarm by optimizing the deployment of drones.

[0115] Drone deployment optimization is a continuous variable and can be solved using a multi-agent deterministic policy gradient method. The upper layer’s goal is to maximize the number of tasks within the drone’s coverage area, that is, to maximize the upper layer’s cumulative reward, which can be expressed as a function:

[0116]

[0117] The action value function corresponding to the optimal strategy is represented by the Bellman equation, which is:

[0118]

[0119] Multi-agent deterministic policy gradient critic uses a neural network (θ c ) fits the Q value function and trains the Q function through the loss value. The loss function is:

[0120]

[0121] in, It is the state transition tuple stored in the experience replay pool.

[0122] The policy function of the multi-agent deterministic policy gradient actor adopts u j It means that it takes the state information of the environment as input and generates a deterministic action. Similarly, it passes through the neural network (θ j ) approximation; the gradient of the policy function is expressed as:

[0123]

[0124] Among them, the value of Q is updated through the loss function.

[0125] In order to make the training of the neural network more stable, two sets of Actor networks and Critic networks are established, namely: Critic network, Target Critic (Q′) and Actor network, Target Actor (u j ′); the two Target networks are also composed of neural networks (θ c′ ,θ j′ ) is constructed and is composed of (θ c ,θ j ) is updated regularly.

[0126] After sampling ε times from the experience replay pool, the loss function and gradient function are:

[0127]

[0128]

[0129] Secondly, the offloading optimization problem of the lower layer is analyzed; given the location of the drone, the choice of mobile device association is determined according to the cost function. Mobile device association is a binary variable. Due to multi-device offloading, the computing resources are calculated as the average distribution of the maximum computing resources of the drone. The output of the lower layer is a discrete variable.

[0130] Status:s t+η ={x j (t+1),y j (t+1),h j ,x i (t+1),y i (t+1),P i};

[0131] Action: a t+η ={α ij ,f ij};

[0132] award:

[0133] Target: Minimize system cost by determining device association and resource allocation for output.

[0134] Similarly, the target network Q′ uses the neural network (θ α ) and the data in the experience replay pool ξ are updated, and the learning experience ξ in the experience pool = {s t+η ,a t+η ,r lower ,s t+η+1}, update the Q-value function by sampling data from the experience replay pool, and the updated Q-value function is:

[0135]

[0136] Among them, β represents the learning rate of the neural network, and the neural network parameter θ α Updated by the loss function, specifically:

[0137]

[0138] The final Q value function (θ α ) is to obtain the optimal strategy for problem P2, that is, the optimal unloading strategy.

[0139] In summary, in the embodiment of the present invention, an edge computing scenario is constructed, and the edge computing scenario includes a group of mobile devices and a group of drones, and each mobile device continuously generates computing tasks; the completion time and energy consumption cost of the mobile device when selecting a local computing task are obtained, and the first cost of the local computing task is obtained based on the completion time and the energy consumption cost; the number of unloadings of the mobile device when receiving an unloading request is obtained, and the unloading completion time is obtained according to the unloading number, the data transmission time and the drone execution time; the energy consumption is obtained based on the transmission power of the mobile device and the computing frequency of the task assigned to the mobile device by the drone; the second cost is obtained based on the unloading completion time and the energy consumption; the remaining power of the current mobile device, the deadline of the computing task and the local computing power are obtained, and the priority of the current mobile device is obtained by weighted summation according to the remaining power, the deadline and the local computing power; a total cost model is constructed with the first cost, the second cost and the priority of each mobile device, and the unloading strategy is determined based on the total cost model, and the unloading strategy has better effect, low total cost and high efficiency.

[0140] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0141] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for uninstalling an air-ground collaborative edge computing server, characterized in that: The method comprises the following steps: Building an edge computing scenario, wherein the edge computing scenario includes a group of mobile devices and a group of drones, each of which continuously generates computing tasks; Obtaining the completion time and energy consumption cost of the mobile device when selecting a local computing task, and obtaining a first cost of the local computing task based on the completion time and the energy consumption cost; obtaining the number of unloadings of the mobile device when receiving an unloading request, and obtaining the unloading completion time according to the unloading number, data transmission time, and the time when the drone performs the task; obtaining energy consumption based on the transmission power of the mobile device and the computing frequency of the task assigned to the mobile device by the drone; and obtaining a second cost based on the unloading completion time and the energy consumption; Acquire the remaining power of the current mobile device, the deadline of the computing task, and the local computing capacity, and obtain the priority of the current mobile device by performing weighted summation according to the remaining power, the deadline, and the local computing capacity; Obtain the drone's track, device association status, and resource allocation, build a total cost model based on the first cost, the second cost, and the priority of each mobile device, determine the unloading strategy based on the total cost model, and the device association status indicates whether the drone chooses a mobile device to access.

2. The method for uninstalling an air-ground collaborative edge computing server according to claim 1, characterized in that: The step of determining the unloading strategy based on the total cost model further includes: Constructing a hierarchical reinforcement learning model, wherein the upper layer of the hierarchical reinforcement learning model is the location optimization of multi-UAV deployment, and the lower layer is the unloading optimization strategy; The upper layer of the hierarchical reinforcement learning model receives the environment state and generates action output and the next state through a deterministic policy network; the next state is used as the input of the lower layer of the hierarchical reinforcement learning model, and the lower layer outputs the unloading strategy.

3. The method for uninstalling an air-ground collaborative edge computing server according to claim 1, characterized in that: The step of acquiring the drone track, device association status, and resource allocation, and building a total cost model in combination with the first cost, the second cost, and the priority of each mobile device includes: The total cost model is: ； in, Indicates the track of the drone; Indicates the device association status, using Indicates that the value is 0 or 1; Indicates resource allocation; and Respectively represent weights; represents the first cost; represents the second cost; Represents a mobile device Priority; Indicates the completion time; represents the energy consumption cost; Indicates the time when uninstallation is completed; Indicates energy consumption.

4. The method for uninstalling an air-ground collaborative edge computing server according to claim 1, characterized in that: The step of obtaining the completion time and energy consumption cost of the mobile device when selecting a local computing task includes: The completion time is the ratio of the computing capacity required by the mobile device to process the task to the computing power of the mobile device; An energy coefficient of a central processing unit and a square value of computing power of a mobile device are obtained, and a product of the energy coefficient and the square value is used as the energy consumption cost.

5. The method for uninstalling an air-ground collaborative edge computing server according to claim 3, characterized in that: The step of obtaining the unloading completion time according to the unloading quantity, data transmission time and the UAV execution task time includes: The calculation method of the uninstallation completion time is: ； in, Indicates the data transmission time. , Indicates the data transmission rate, Indicates the size of the task data; Indicates the time when the drone performs the mission. , Determined by the number of mobile devices connected to the drone, , Indicates the number of mobile devices that need to be uninstalled. Represents the computing power of the drone.

6. The method for uninstalling an air-ground collaborative edge computing server according to claim 1, characterized in that: The step of obtaining energy consumption based on the transmission power of the mobile device and the calculation frequency of the task assigned by the drone to the mobile device comprises: Calculating the sum of the transmission energy consumption and the execution energy consumption of all offloaded tasks as the energy consumption; The transmission energy consumption is the product of the transmission power of the mobile device and the data transmission time; the execution energy consumption is obtained by the calculation frequency of the task assigned by the drone to the mobile device.

7. The method for uninstalling an air-ground collaborative edge computing server according to claim 2, characterized in that: The upper layer of the hierarchical reinforcement learning model specifically optimizes the position of multiple drone deployments by: state: ;in, It represents the amount of data within the communication range of the drone group, which is obtained by adding the amount of data from devices within the drone coverage area; Indicates the three-dimensional position of the drone; Indicates the horizontal position of the mobile device; Indicates the remaining battery power of the mobile device; action: ,in, Indicates that the drone is at time The steering angle, Indicates that the drone is at time Flight distance; award: ; Objective: To maximize the amount of missions within the communication range of the drone swarm by optimizing the deployment of drones.

8. The method for uninstalling an air-ground collaborative edge computing server according to claim 7, characterized in that: The unloading optimization strategy of the lower layer of the hierarchical reinforcement learning model specifically includes: state: ; action: ; award: ; Target: , to determine the output device association and resource allocation to minimize the system cost and obtain the optimal offloading strategy.

9. The method for uninstalling an air-ground collaborative edge computing server according to claim 5, characterized in that: The method for acquiring the data transmission rate comprises: Obtain the number of mobile devices currently connected to the drone and the bandwidth, calculate the ratio of the bandwidth to the number of mobile devices currently connected to the drone and record it as a first ratio; Obtaining the horizontal distance between the current mobile device and the current drone and the height of the current drone, and calculating the sum of the square value of the horizontal distance and the square value of the height of the drone; obtaining the product result of the first indicator and the transmission power of the current mobile device, and recording the ratio of the product result to the sum result as the second ratio; The second ratio is added with 1, and a logarithmic operation is performed on the addition result, and the data transmission rate is obtained by multiplying the logarithmic operation result and the first ratio.

10. The method for uninstalling an air-ground collaborative edge computing server according to claim 9, characterized in that: The first indicator is: ； in, Represents the channel gain per unit spatial distance; is a constant; Represents the noise power.

Images