Edge calculation method and device for multi-task parallel execution assisted by unmanned aerial vehicle

Abstract

The invention provides an edge calculation method and device for multi-task parallel execution assisted by an unmanned aerial vehicle, and relates to the technical field of mobile edge calculation. The method comprises the following steps: optimizing and constructing a content caching and unloading mode of a single unmanned aerial vehicle as a Markov decision process by using an optimization target of minimizing the total service duration of the unmanned aerial vehicle, classifying mobile equipment by using a deep Q network algorithm, optimizing an unloading mode of a calculation task of each mobile equipment, and calculating the task of the mobile equipment according to the optimized unloading mode. And an optimal allocation scheme under a set condition is obtained. For an optimization target of maximizing a transmission rate sum between an unmanned aerial vehicle and a mobile device, based on a game theory, unmanned aerial vehicle channel allocation is modeled as a random game model, and a channel allocation result satisfying a related equilibrium strategy is obtained through a regret value minimization algorithm. According to the method, the network cost is effectively reduced, and the service experience of the user is improved. The edge calculation method is based on unmanned aerial vehicle assistance, realizes multi-task parallelism, and is high in execution efficiency and low in network cost.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile edge computing, and particularly to an edge computing method and device for multi-task parallel execution assisted by an unmanned aerial vehicle. Background Art

[0002] With the rapid development of mobile networks and edge intelligence, various new applications such as augmented reality, virtual reality, and image recognition have emerged continuously. These applications have characteristics of being computationally intensive or latency-sensitive, which pose severe challenges to mobile devices (MDs) with limited computing capabilities. Mobile Edge Computing (MEC), as an extension of the traditional cloud data processing architecture, is an important solution to solve this problem. By deploying servers with storage and computing capabilities at the edge of the mobile network, MEC can transfer data and computing functions from the centralized cloud to the network edge. Using MEC can provide real-time and efficient computing services for the complex and diverse tasks of MDs, significantly improving the user service experience.

[0003] In traditional methods, infrastructure-dependent MEC improves the execution efficiency of complex computing tasks, but there are still significant challenges in providing generally reliable computing services for MDs. With the explosive growth in the number of MDs, static MEC servers are difficult to support the real-time and efficient large-scale service requirements. Traditional static MEC servers are usually deployed with limitations according to experience and budget, which may lead to unbalanced network loads and thus reduce the service quality. Due to the characteristics of high mobility and easy deployment, unmanned aerial vehicles have become one of the important forms of assisting edge computing, and they can provide flexible and efficient computing services for MDs on demand in complex scenarios.

[0004] Regarding the resource optimization in the unmanned aerial vehicle-assisted mobile edge computing network, many research works have provided valuable methods from multiple perspectives, such as: research methods for trajectory control and task offloading, optimization research on communication resources, joint optimization research on computing and communication, optimization research on content caching resources, etc. This has laid a foundation for subsequent research. Some key points that are crucial for improving the service quality of mobile devices and reducing network costs still need to be explored.

[0005] In the processing of most target tasks, either they are processed one by one, or only partially occupy computing and communication resources for execution. The sequential execution method leads to low resource utilization efficiency, and the resource sharing method may lead to unhealthy resource competition. The existing technology has insufficient research on the joint optimization of content caching, computing offloading, and communication resources. For complex joint optimization problems, there is a lack of effective processing methods, and it is difficult to consider the mutual relationship between multiple optimization dimensions simultaneously.

[0006] In the prior art, there is a lack of an edge computing method based on drone assistance that can achieve high execution efficiency and low network cost for multi-task parallel execution. Summary of the Invention

[0007] To solve the technical problems in the prior art, such as the low resource utilization efficiency and the long total service time of drones caused by unhealthy resource competition, and the high network computing cost caused by insufficient exploration of the joint optimization of content caching, computing offloading, and communication resources, an edge computing method and device for multi-task parallel execution assisted by drones are provided in the embodiments of the present invention. The technical solutions are as follows:

[0008] On the one hand, an edge computing method for multi-task parallel execution assisted by drones is provided. This method is implemented by an edge computing device and includes:

[0009] Obtain the first computing task set of multiple mobile devices within the service range of the drone, the environmental state set of the drone, the available channels of the drone, the action space of the drone, and the basic parameters of the drone;

[0010] Based on the first computing task set, the drone performs cache optimization to obtain the optimized content cache of the drone;

[0011] According to the optimized content cache set and the first computing task set, screen the multiple mobile devices to obtain the first type of mobile device set and the second type of mobile device set;

[0012] Taking the minimization of the total service duration of the drone as the optimization goal, construct a Markov decision process according to the optimized content cache, the first computing task set, and the preset drone action strategy;

[0013] Based on the Markov decision process and the deep Q-network algorithm, screen the second type of mobile device set according to the first computing task set to obtain the third type of mobile device set and the fourth type of mobile device set;

[0014] Taking the maximization of the drone transmission rate sum as the optimization goal, construct a stochastic game model according to the drone environmental state set and the action space;

[0015] Based on the stochastic game model, perform iterative optimization on the allocation of the available channels through the regret value minimization algorithm according to the action space to obtain the optimal channel allocation;

[0016] Based on the first type of mobile device set, the third type of mobile device set, and the fourth type of mobile device set, perform calculations according to the basic parameters of the drone, the first computing task set, and the optimal channel allocation to obtain the minimum total service duration of the drone.

[0017] On the other hand, an edge computing device for drone-assisted multi-task parallel execution is provided. This device is applied to the edge computing method for drone-assisted multi-task parallel execution, and the device includes:

[0018] A drone information acquisition module, configured to acquire a first set of computing tasks of multiple mobile devices within the service range of the drone, an environmental state set of the drone, available channels of the drone, an action space of the drone, and basic drone parameters;

[0019] A content cache optimization module, configured to perform cache optimization on the drone based on the first set of computing tasks to obtain an optimized content cache of the drone;

[0020] A first device classification module, configured to screen the multiple mobile devices according to the optimized content cache set and the first set of computing tasks to obtain a first set of mobile devices and a second set of mobile devices;

[0021] A Markov decision process construction module, configured to construct a Markov decision process with the goal of minimizing the total service duration of the drone according to the optimized content cache, the first set of computing tasks, and a preset drone action strategy;

[0022] A second device classification module, configured to screen the second set of mobile devices according to the first set of computing tasks based on the Markov decision process and the deep Q-network algorithm to obtain a third set of mobile devices and a fourth set of mobile devices;

[0023] A game model construction module, configured to construct a stochastic game model with the goal of maximizing the sum of drone transmission rates according to the drone environmental state set and the action space;

[0024] An optimal channel allocation module, configured to iteratively optimize the allocation of the available channels based on the stochastic game model and according to the action space through a regret value minimization algorithm to obtain an optimal channel allocation;

[0025] A total service duration calculation module, configured to calculate based on the first set of mobile devices, the third set of mobile devices, and the fourth set of mobile devices according to the basic drone parameters, the first set of computing tasks, and the optimal channel allocation to obtain the minimum total service duration of the drone.

[0026] On the other hand, an edge computing device is provided. The edge computing device includes: a processor; a memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, any one of the methods in the above-mentioned edge computing method for drone-assisted multi-task parallel execution is implemented.

[0027] On the other hand, a computer-readable storage medium is provided, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned edge computing methods for multi-task parallel execution assisted by drones.

[0028] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include:

[0029] The present invention proposes an edge computing method for multi-task parallel execution assisted by drones. With the optimization goal of minimizing the total service duration of drones, the content caching and offloading mode optimization of a single drone is constructed as a Markov decision process. The deep Q-network algorithm is used to classify mobile devices, and at the same time, the offloading mode of each mobile device's computing task is optimized to obtain an optimal allocation scheme under given conditions. For the optimization goal of maximizing the sum of the transmission rates between drones and mobile devices, based on game theory, the drone channel allocation is modeled as a stochastic game model, and through the regret value minimization algorithm, a channel allocation result that satisfies the relevant equilibrium strategy is obtained. This method effectively reduces the network cost and improves the user's service experience. The present invention is an edge computing method based on drone assistance that realizes high execution efficiency and low network cost for multi-task parallel execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0031] Figure 1 is a flowchart of an edge computing method for multi-task parallel execution assisted by drones provided by an embodiment of the present invention;

[0032] Figure 2 is a schematic diagram of three possible task completion sequences under multi-task parallel processing provided by an embodiment of the present invention;

[0033] Figure 3 is a comparison chart of the average service duration of each drone under different optimization algorithms provided by an embodiment of the present invention;

[0034] Figure 4 is a block diagram of an edge computing device for multi-task parallel execution assisted by drones provided by an embodiment of the present invention;

[0035] Figure 5 is a schematic structural diagram of an edge computing device provided by an embodiment of the present invention. Detailed implementation manners

[0036] The following describes the technical solutions in the present invention with reference to the accompanying drawings.

[0037] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "example" is intended to present concepts in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either of the two can be selected.

[0038] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same. "(of)", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meanings they express are the same.

[0039] In the embodiments of the present invention, sometimes subscripts such as W 1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings they express are the same.

[0040] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0041] The embodiments of the present invention provide an edge computing method for multi-task parallel execution assisted by an unmanned aerial vehicle (UAV). This method can be implemented by an edge computing device, and the edge computing device can be a terminal or a server. As Figure 1 shown in the flowchart of the edge computing method for multi-task parallel execution assisted by a UAV, the processing flow of this method can include the following steps:

[0042] S1. Obtain the first set of computing tasks of multiple mobile devices within the service range of the UAV, the environmental state set of the UAV, the available channels of the UAV, the action space of the UAV, and the UAV basic parameters.

[0043] In a feasible implementation manner, the UAV basic parameters include the operating cycle of the UAV processing CPU and the computing frequency of the UAV. The first set of computing tasks is the set of computing tasks of all UAVs.

[0044] S2. Based on the first set of computing tasks, the UAV performs cache optimization to obtain the optimized content cache of the UAV.

[0045] Optionally, based on the first set of computing tasks, the UAV performs cache optimization to obtain the optimized content cache of the UAV, including:

[0046] Ascendingly sort the first set of computing tasks according to the content quantity of each task in the first set of computing tasks to obtain the sorted set of computing tasks;

[0047] Based on the sorted set of computing tasks, the UAV performs content caching to obtain the optimized content cache set of the UAV.

[0048] In a feasible implementation manner, according to the content required by the mobile device computing tasks, sort them in ascending order of quantity, and the UAV preferentially caches the content with a small quantity. Based on this, the UAV completes the optimization of content caching and only needs to consider the optimization of the offloading mode of the associated mobile device computing tasks.

[0049] S3. According to the optimized content cache set and the first set of computing tasks, screen multiple mobile devices to obtain the first type of mobile device set and the second type of mobile device set.

[0050] Optionally, according to the optimized content cache set and the first set of computing tasks, screen multiple mobile devices to obtain the first type of mobile device set and the second type of mobile device set, including:

[0051] Obtain the uploaded data set of the mobile device according to the optimized content cache set;

[0052] Obtain the data set to be calculated of the mobile device according to the first set of computing tasks;

[0053] Use the data set to be calculated to verify the uploaded data set to obtain the verification result;

[0054] Among multiple mobile devices, select the mobile devices with consistent verification results to obtain the first type of mobile device set;

[0055] Among multiple mobile devices, select the mobile devices with inconsistent verification results to obtain the second type of mobile device set.

[0056] In a feasible implementation manner, in the present invention, the data set to be calculated required for the edge computing task execution of the mobile device and the uploaded data set cached at the UAV side are verified.

[0057] If the content of the data set to be calculated is consistent with the content of the uploaded data set, the mobile device corresponding to the data set belongs to the first type of mobile device, and all the first type of mobile devices are screened out and aggregated into the first type of mobile device set 。

[0058] Conversely, if the content of the dataset to be calculated is inconsistent with the content of the uploaded dataset, the mobile device corresponding to the dataset belongs to the second type of mobile device, and all the second type of mobile devices are screened out and aggregated into a set of the second type of mobile devices 。

[0059] S4. Taking the minimization of the total service duration of the UAV as the optimization objective, a Markov decision process is constructed according to the optimized content caching, the first calculation task set, and the preset UAV action strategy.

[0060] In a feasible implementation manner, in the transmission network of the present invention, each UAV (labeled as ) and the mobile devices randomly distributed within the signal coverage area of the UAV, the set of mobile devices is , and the state of the entire Markov decision process is represented as the following formula (1):

[0061] (1);

[0062] where, represents the calculation task generated by the mobile device , represents the storage capacity of the UAV .

[0063] The action of the UAV in the Markov decision process is represented as the following formula (2):

[0064] (2);

[0065] where, represents the content caching vector of the UAV , represents that the UAV stores the rd content, represents all content; represents the offloading mode vector of the tasks of the mobile devices belonging to the set , where, represents the set of mobile devices whose required content for task execution is not fully cached on the UAV side.

[0066] Thus, the scale of the action space A of the UAV action is as the following formula (3):

[0067] (3);

[0068] where, represents the set belonging to The quantity of content required for tasks of all mobile devices, Indicates the mobile device The content required for the submitted tasks, Indicates the drone The cached content.

[0069] Based on the optimization of content caching, the scale of the action space is simplified to the following formula (4):

[0070] (4);

[0071] S5. Based on the Markov decision process and the deep Q-network algorithm, according to the first set of computing tasks, screen the second set of mobile devices to obtain the third set of mobile devices and the fourth set of mobile devices.

[0072] Optionally, based on the Markov decision process and the deep Q-network algorithm, according to the first set of computing tasks, screen the second set of mobile devices to obtain the third set of mobile devices and the fourth set of mobile devices, including:

[0073] In the first set of computing tasks, select the tasks corresponding to the second set of mobile devices to obtain the second set of computing tasks;

[0074] Based on the Markov decision process, according to the second set of computing tasks, use the deep Q-network algorithm to select the offloading mode to obtain the second set of offloading modes;

[0075] In the second set of mobile devices, select the mobile devices whose offloading mode in the set of offloading modes is to offload to the remote base station to obtain the third set of mobile devices;

[0076] In the second set of mobile devices, select the mobile devices whose offloading mode in the set of offloading modes is not to offload to the remote base station to obtain the fourth set of mobile devices.

[0077] In a feasible implementation manner, based on the Markov decision process, use the deep Q-network algorithm to select the offloading mode for the second type of mobile devices. Among them, the offloading mode that offloads to the remote base station is the third type of mobile devices, belonging to the third set of mobile devices ; the offloading mode that is not to offload to the remote base station is the fourth type of mobile devices, belonging to the fourth set of mobile devices .

[0078] Optionally, based on the Markov decision process, according to the second set of computing tasks, the deep Q-network algorithm is used to select the offloading mode to obtain the second set of offloading modes, including:

[0079] According to the second set of computing tasks, the ε-greedy strategy is used to select the offloading mode, and the first offloading mode set is obtained;

[0080] Based on the Markov decision process, calculations are performed according to the offloading mode set, and the immediate reward set is obtained;

[0081] Based on the immediate reward set, the preset action value function, and the preset loss function, the evaluation neural network parameters of the deep Q-network algorithm are optimized using stochastic gradient descent to obtain the optimized evaluation neural network parameters;

[0082] According to the optimized evaluation neural network parameters, the target neural network parameters of the deep Q-network algorithm are updated to obtain the updated target neural network parameters;

[0083] Using the updated target neural network parameters, the preset offloading mode selection strategy is optimized to obtain the optimal selection strategy;

[0084] According to the second set of computing tasks, the optimal selection strategy is used to select the offloading mode, and the second offloading mode set is obtained.

[0085] In a feasible implementation, the reward in the Markov decision process is expressed as the following formula (5):

[0086] (5);

[0087] where, is the duration of the UAV task processing. In state , the UAV selects a suitable action according to the policy , thereby obtaining a reward , and the environment is updated to a new state. The interaction between the UAV and the environment is represented by , where, represents the new state after obtaining the reward in state through the action .

[0088] To evaluate the preset offloading mode selection strategy , the action value function is used as follows in formula (6):

[0089] (6);

[0090] where, is the discount factor, represents the number of interactions, represents the time step in the future reward calculation.

[0091] According to the optimal policy the optimal action-value function can be obtained . Among them, it can be expressed as the following formulas (7) and (8):

[0092] (7); (8);

[0093] Among them, is the policy space.

[0094] The offloading mode of the mobile device's computing tasks is optimized using the deep Q-network algorithm, and the random environment of the entire network is initialized; according to the optimized content caching result, the mobile devices associated with the drone are classified, and for the set the offloading mode of the computing tasks of the mobile devices in is selected according to the ε-greedy policy.

[0095] Calculate the immediate reward obtained by adopting this offloading action, and the environment transfers to the new state; store the historical record in the experience buffer ; if the experience buffer is full, randomly sample a small batch from the experience buffer ; calculate the target Q value; calculate the gradient of the evaluation neural network parameter and update using stochastic gradient descent to minimize the loss function; periodically update the target neural network parameter ; repeat the above steps until the preset number of iterations is reached.

[0096] The record stored in the experience buffer is ; among them, represents the environmental state at the th iteration, represents the action adopted by the drone at the th iteration, represents the immediate reward obtained by adopting the action in the state , represents a new state entered after adopting the action in the state .

[0097] The calculation formula of the target Q value is as follows formula (9):

[0098] (9);

[0099] Among them, represents the optimal Q value of the next state.​

[0100] The loss function is as shown in Equation (10) below:

[0101] (10);

[0102] Wherein, represents sampling a series of historical records from the experience buffer to calculate the expectation. Evaluate the gradients of the neural network parameters

[0103] as shown in Equation (11) below: (11); (11);

[0104] Wherein, represents the gradient of the action value function with respect to the parameter .

[0105] The update strategy for the target neural network parameters is .

[0106] S6. With maximizing the UAV transmission rate sum as the optimization goal, construct a stochastic game model according to the UAV environmental state set and the action space.

[0107] In a feasible implementation, the overall structure of the stochastic game model is expressed as as shown in Equation (12) below:

[0108] (12);

[0109] Wherein, the agent set of the stochastic game is the UAV set .

[0110] In a multi-UAV system, there is no information exchange between different UAVs. Therefore, each UAV only has local information about the environmental state as shown in Equation (13) below:

[0111] (13);

[0112] Wherein, represents the position of the UAV itself, represents the optimized UAV content caching decision, represents the optimized UAV computing task offloading mode decision, and respectively represent the position and computing task of the mobile device associated with the UAV.

[0113] The local information of all drones regarding the environmental state constitutes the environment state of the game of the game as shown in Equation (14) below:

[0114] (14);

[0115] S7. Based on the stochastic game model, according to the action space, the allocation of available channels is iteratively optimized through the regret value minimization algorithm to obtain the optimal channel allocation.

[0116] Optionally, based on the stochastic game model, according to the action space, the allocation of available channels is iteratively optimized through the regret value minimization algorithm to obtain the optimal channel allocation, including:

[0117] Based on the stochastic game model, calculate according to the actions in the action space to obtain the action values of the game action space;

[0118] According to the action values, conduct a game through the stochastic game model to obtain the historical cumulative regret;

[0119] Iteratively optimize the historical cumulative regret through the regret value minimization algorithm to obtain the minimum historical cumulative regret;

[0120] Calculate the channel policy update rule according to the minimum historical cumulative regret;

[0121] According to the channel policy update rule, optimize the preset channel allocation policy to obtain the optimal allocation policy;

[0122] Use the optimal allocation policy to allocate the available channels to obtain the optimal channel allocation.

[0123] In a feasible implementation, each drone selects an action from its action space and the action selections of all drones constitute the action set of the game ; ;

[0124] wherein, is as shown in Equation (15) below:

[0125] (15);

[0126] represents the set of utility functions of the game ; represents the utility function of drone as shown in Equation (16) below:

[0127] (16);

[0128] Among them, is the set of channels available to the UAV, represents the UAV 's selection status for the channel , indicating that the channel is selected, indicating that the channel is not selected, and represents the transmission rate between the UAV and the mobile device when the channel is selected.

[0129] In this game framework, is the channel selection strategy of the UAV, which is a probability distribution defined on the action space and is expressed as the following formula (17): (17);

[0130] (17);

[0131] Among them, represents the probability that the UAV selects the action .

[0132] The UAV interacts with the environment to update the strategy and finally reaches the correlated equilibrium, which is a set of strategies in the joint action space , and for any UAV (labeled ), this process is as follows in formula (18): (18);

[0133] (18);

[0134] Among them, represents the probability of taking the action combination , represents the action selection combination of other UAVs except the UAV , represents any other action of the UAV in its action space except .

[0135] Iteratively updating the channel selection of the UAV through the regret value minimization algorithm, obtaining the correlated equilibrium solution of the stochastic game model includes:

[0136] To characterize the local observation attribute of the UAV, the counterfactual advantage of the UAV's action is introduced, which is jointly determined by the action value and the observation value. The UAV Take Action The obtained action value is expressed as As shown in formula (19):

[0137] (19);

[0138] Environmental status The observation value Defined as the action space All action values ​​in The expectation is as follows (20):

[0139] (20);

[0140] Accordingly, drones In the environmental state No action was taken The counterfactual advantage It is expressed as follows (21):

[0141] (twenty one);

[0142] In order to make full use of past experience to learn the optimal strategy, a series of games need to be played and the drones Always take action The accumulated regrets of history As shown in formula (22):

[0143] (twenty two);

[0144] in, Indicates the current number of iterations, Indicates starting the summation from the first iteration.

[0145] Every iteration will update the drone Channel selection strategy , the update rule is as follows (23):

[0146] (twenty three);

[0147] in, Indicates selection The maximum value between 0 and Represents the sum of the non-negative regrets of all feasible actions.

[0148] S8. Based on the first category mobile device set, the third category mobile device set and the fourth category mobile device set, calculation is performed according to the basic parameters of the drone, the first computing task set and the optimal channel allocation to obtain the minimum total drone service time.

[0149] Optionally, based on the first type of mobile device set, the third type of mobile device set, and the fourth type of mobile device set, calculate according to the UAV basic parameters, the first set of computing tasks, and the optimal channel allocation to obtain the minimum total UAV service duration, including:

[0150] Based on the first type of mobile device set, calculate the UAV service duration according to the first set of computing tasks, the optimal channel allocation, and the UAV basic parameters to obtain the first type of UAV service duration;

[0151] Based on the third type of mobile device set, calculate the UAV service duration according to the first set of computing tasks, the optimal channel allocation, and the UAV basic parameters to obtain the second type of UAV service duration;

[0152] Based on the fourth type of mobile device set, calculate the UAV service duration according to the first set of computing tasks, the optimal channel allocation, and the UAV basic parameters to obtain the third type of UAV service duration;

[0153] Calculate according to the first type of UAV service duration, the second type of UAV service duration, and the third type of UAV service duration to obtain the minimum total UAV service duration.

[0154] In a feasible implementation manner, for the mobile devices belonging to the set since the UAV caches the content it needs, after receiving the task request from the mobile device , the UAV will perform edge computing and return the calculation result. The computing delay and the transmission delay are respectively expressed as the following formulas (24), (25):

[0155] (24);

[0156] (25);

[0157] where represents the CPU operation cycles required to complete the task, represents the computing frequency of the UAV , represents the data size of the calculation result, represents when selecting the channel , the transmission rate between the UAV and the mobile device .

[0158] For the mobile devices belonging to the set since the UAV Only part of the content required for its computing task is cached, so the drone needs to first fetch the missing content from the remote base station. The transmission delay for receiving this content is expressed as Equation (26):

[0159] (26);

[0160] where represents the content that needs to be retrieved, represents the drone and the channel transmission rate between the remote base station.

[0161] The drone performs edge computing, and the computing delay is expressed as Equation (27):

[0162] (27);

[0163] After the calculation is completed, the result is transmitted back to the mobile device, and the transmission delay is expressed as Equation (28):

[0164] (28);

[0165] For the mobile devices belonging to the set , the drone will directly offload the computing task to the remote base station. After the base station completes the execution, the calculation result will be transmitted to the drone , and the drone will then return the result to the mobile device. The delays of these three steps are the remote computing delay , the base station transmission delay , and the drone transmission delay , which are expressed as Equation (29):

[0166] (29);

[0167] where is the computing frequency of the remote base station.

[0168] The computing tasks of the mobile devices in the set will be completed before the computing tasks of the mobile devices in the set , while the completion order of the computing tasks of the mobile devices in the set is uncertain. It may be completed before or after the tasks of the mobile devices in the set. Therefore, there are three different task completion orders, as shown in Equation (30) below:

[0169] (30);

[0170] Under different task completion orders, the UAV has different service durations To distinguish these three cases, three service durations are used here as shown in Equation (31):

[0171] (31);

[0172] To accurately describe the service durations of the UAV server in different situations, symbols indicating the start of certain steps are introduced. Specifically: for the mobile devices in the set , and represent the start times of edge computing and result transmission respectively. For the mobile devices in the set , and represent the start times of content reception, edge computing and result transmission respectively. In addition, for the mobile devices in the set , and represent the start times of remote computing, result reception and result transmission respectively.

[0173] Figure 2 Examples of these three orders are shown. It can be seen that in different situations, the start times of the same processing step are different.

[0174] In the first case, the tasks of the mobile devices in the set are completed last, so the total service duration of the UAV depends on the task result transmission delay of the mobile devices in the set , as shown in Equation (32):

[0175] (32);

[0176] In this task completion order, the start times of the above processing steps are represented as shown in Equation (33):

[0177] (33);

[0178] In the second case, the total service duration of the UAV depends on the task result transmission delay of the mobile devices in the set , that is, Equation (34):

[0179] (34);

[0180] In this task completion order, the start times of the above processing steps are represented as shown in Equation (35):

[0181] (35);

[0182] The third case is similar to the second case. The total service duration of the UAV depends on the task result transmission delay of the mobile devices in the set , that is, Equation (36):

[0183] (36);

[0184] Under this task completion order, the start time of the above processing steps is expressed as Equation (37):

[0185] (37);

[0186] In a feasible implementation manner, the parallel execution mode proposed by the present invention allows the calculation, reception, and transmission of different tasks to be carried out simultaneously, greatly reducing the service delay. In the scenario where mobile devices are densely configured, that is, when the number of mobile devices is large and the number of UAVs is small, the gap between the parallel execution and serial execution schemes is more significant. In the case where mobile devices are relatively sparsely configured, the average delay of each UAV is reduced by 12.6% during parallel execution. In the case where mobile devices are densely configured, this advantage is extended to 21.6%.

[0187] To evaluate the system performance of content caching, the proposed joint optimization mode of caching, computing, and communication is compared with the scheme in which no content is cached in the UAV server.

[0188] Under the same parameter configuration, when content is actively cached in the UAV server, the average service duration of each UAV will be reduced. This is because when no content is cached in the UAV server, the UAV needs to obtain all the content required by the mobile devices belonging to the sets and , or offload more tasks of its associated mobile devices to the ground base station, which will result in an increase in the service duration compared with the scheme of the present invention. In addition, for a certain number of UAVs, as the number of mobile devices increases, the gap between these two schemes will widen; while for a certain number of mobile devices, the fewer the number of UAVs, the more significant the performance advantage.

[0189] To study the efficiency of hybrid computing, the hybrid computing scheme of UAVs and ground base stations is compared with two benchmark schemes, and the two benchmark schemes include: (i) fully offloaded to the ground base station, that is, all tasks of the mobile devices are executed by the ground base station, and the UAVs only act as relays to return the calculation results; (ii) fully calculated on the UAVs, that is, all tasks of the mobile devices are processed by their associated UAVs without offloading.

[0190] In terms of reducing the task completion latency, the hybrid computing mode of the drone and the ground base station outperforms these two benchmark schemes. In addition, for these two benchmark offloading schemes, the performance of computing completely on the drone is inferior to that of completely offloading to the ground base station. Especially when the number of mobile devices increases, this indicates that in scenarios with large task traffic, computing resources rather than communication resources are the main factors affecting the network service duration.

[0191] Such as Figure 3 As shown in the comparison chart of the average service duration of each drone under different optimization algorithms, the total service duration of the drone in the present invention is significantly reduced.

[0192] The present invention proposes an edge computing method for multi-task parallel execution assisted by drones. With the optimization goal of minimizing the total service duration of drones, the content caching and offloading mode optimization of a single drone is constructed as a Markov decision process. The deep Q-network algorithm is used to classify mobile devices, and at the same time, the offloading mode of the computing tasks of each mobile device is optimized to obtain the optimal allocation scheme under given conditions. For the optimization goal of maximizing the sum of the transmission rates between the drone and the mobile devices, based on game theory, the drone channel allocation is modeled as a stochastic game model, and through the regret value minimization algorithm, the channel allocation result satisfying the correlated equilibrium strategy is obtained. This method effectively reduces the network cost and improves the service experience of users. The present invention is an edge computing method based on drone assistance that realizes high execution efficiency and low network cost for multi-task parallelism.

[0193] Figure 4 It is a block diagram of an edge computing device for multi-task parallel execution assisted by drones shown according to an exemplary embodiment. This device is used for the edge computing method of multi-task parallel execution assisted by drones. Referring to Figure 4 , this device includes a drone information acquisition module 410, a content caching optimization module 420, a first device classification module 430, a Markov decision process construction module 440, a second device classification module 450, a game model construction module 460, an optimal channel allocation module 470, and a total service duration calculation module 480. Among them:

[0194] The drone information acquisition module 410 is used to acquire the first computing task set of multiple mobile devices within the service range of the drone, the environmental state set of the drone, the available channels of the drone, the action space of the drone, and the basic parameters of the drone;

[0195] The content caching optimization module 420 is used to perform caching optimization on the drone based on the first computing task set to obtain the optimized content caching of the drone;

[0196] The first device classification module 430 is configured to screen multiple mobile devices according to the optimized content cache set and the first computing task set, and obtain the first type of mobile device set and the second type of mobile device set;

[0197] The Markov decision process construction module 440 is configured to construct a Markov decision process with the goal of minimizing the total service duration of the drones, according to the optimized content cache, the first computing task set, and the preset drone action strategy;

[0198] The second device classification module 450 is configured to screen the second type of mobile device set based on the Markov decision process and the deep Q-network algorithm according to the first computing task set, and obtain the third type of mobile device set and the fourth type of mobile device set;

[0199] The game model construction module 460 is configured to construct a stochastic game model with the goal of maximizing the sum of the drone transmission rates, according to the drone environment state set and the action space;

[0200] The optimal channel allocation module 470 is configured to iteratively optimize the allocation of available channels based on the stochastic game model and the action space through the regret value minimization algorithm, and obtain the optimal channel allocation;

[0201] The total service duration calculation module 480 is configured to calculate based on the first type of mobile device set, the third type of mobile device set, and the fourth type of mobile device set, according to the drone basic parameters, the first computing task set, and the optimal channel allocation, to obtain the minimum total service duration of the drones.

[0202] Optionally, the content cache optimization module 420 is further configured to:

[0203] Sort the first computing task set in ascending order according to the content quantity of each task in the first computing task set, and obtain the sorted computing task set;

[0204] According to the sorted computing task set, the drones perform content caching to obtain the optimized content cache set of the drones.

[0205] Optionally, the first device classification module 430 is further configured to:

[0206] Obtain the uploaded data set of the mobile devices according to the optimized content cache set;

[0207] Obtain the to-be-computed data set of the mobile devices according to the first computing task set;

[0208] Use the to-be-computed data set to verify the uploaded data set, and obtain the verification result;

[0209] Among multiple mobile devices, select the mobile devices with consistent verification results to obtain the first set of mobile devices;

[0210] Among multiple mobile devices, select the mobile devices with inconsistent verification results to obtain the second set of mobile devices.

[0211] Optionally, the second device classification module 450 is further configured to:

[0212] Among the first set of computing tasks, select the tasks corresponding to the second set of mobile devices to obtain the second set of computing tasks;

[0213] Based on the Markov decision process, according to the second set of computing tasks, use the deep Q-network algorithm to perform offloading mode selection to obtain the second set of offloading modes;

[0214] Among the second set of mobile devices, select the mobile devices with the offloading mode of offloading to the remote base station in the set of offloading modes to obtain the third set of mobile devices;

[0215] Among the second set of mobile devices, select the mobile devices with the offloading mode not being offloading to the remote base station in the set of offloading modes to obtain the fourth set of mobile devices.

[0216] Optionally, the second device classification module 450 is further configured to:

[0217] According to the second set of computing tasks, use the ε-greedy strategy to perform offloading mode selection to obtain the first set of offloading modes;

[0218] Based on the Markov decision process, perform calculations according to the set of offloading modes to obtain the immediate reward set;

[0219] Based on the immediate reward set, the preset action value function, and the preset loss function, use stochastic gradient descent to optimize the evaluation neural network parameters of the deep Q-network algorithm to obtain the optimized evaluation neural network parameters;

[0220] According to the optimized evaluation neural network parameters, update the target neural network parameters of the deep Q-network algorithm to obtain the updated target neural network parameters;

[0221] Use the updated target neural network parameters to optimize the preset offloading mode selection strategy to obtain the optimal selection strategy;

[0222] According to the second set of computing tasks, use the optimal selection strategy to perform offloading mode selection to obtain the second set of offloading modes.

[0223] Optionally, the optimal channel allocation module 470 is further configured to:

[0224] Based on the stochastic game model, calculate according to the actions in the action space to obtain the action values in the game action space;

[0225] Perform a game through the stochastic game model according to the action values to obtain the historical cumulative regret;

[0226] Iteratively optimize the historical cumulative regret through the regret value minimization algorithm to obtain the minimum historical cumulative regret;

[0227] Calculate the channel policy update rule according to the minimum historical cumulative regret;

[0228] Optimize the preset channel allocation policy according to the channel policy update rule to obtain the optimal allocation policy;

[0229] Allocate the available channels using the optimal allocation policy to obtain the optimal channel allocation.

[0230] Optionally, the total service duration calculation module 480 is further configured to:

[0231] Based on the first type of mobile device set, calculate the UAV service duration according to the first calculation task set, the optimal channel allocation, and the UAV basic parameters to obtain the first type of UAV service duration;

[0232] Based on the third type of mobile device set, calculate the UAV service duration according to the first calculation task set, the optimal channel allocation, and the UAV basic parameters to obtain the second type of UAV service duration;

[0233] Based on the fourth type of mobile device set, calculate the UAV service duration according to the first calculation task set, the optimal channel allocation, and the UAV basic parameters to obtain the third type of UAV service duration;

[0234] Calculate based on the first type of UAV service duration, the second type of UAV service duration, and the third type of UAV service duration to obtain the minimum total UAV service duration.

[0235] The present invention proposes an edge computing method for multi-task parallel execution assisted by drones. With the optimization goal of minimizing the total service duration of drones, the optimization of the content caching and offloading mode of a single drone is constructed as a Markov decision process. The deep Q-network algorithm is used to classify mobile devices, and at the same time, the offloading mode of the computing tasks of each mobile device is optimized to obtain an optimal allocation scheme under given conditions. For the optimization goal of maximizing the sum of the transmission rates between drones and mobile devices, based on game theory, the drone channel allocation is modeled as a stochastic game model, and through the regret value minimization algorithm, the channel allocation result satisfying the correlated equilibrium strategy is obtained. This method effectively reduces the network cost and improves the service experience of users. The present invention is an edge computing method based on drone assistance that realizes high execution efficiency and low network cost for multi-task parallel execution.

[0236] Figure 5 FIG. is a schematic structural diagram of an edge computing device provided by an embodiment of the present invention, as Figure 5 shown. The edge computing device may include the above-mentioned Figure 4 edge computing device for multi-task parallel execution assisted by drones as shown. Optionally, the edge computing device 510 may include a first processor 2001.

[0237] Optionally, the edge computing device 510 may further include a memory 2002 and a transceiver 2003.

[0238] Among them, the first processor 2001, the memory 2002, and the transceiver 2003 may be connected through a communication bus.

[0239] Next, in combination with Figure 5 each component of the edge computing device 510 will be specifically introduced:

[0240] Among them, the first processor 2001 is the control center of the edge computing device 510, which may be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or may be an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention, such as: one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs).

[0241] Optionally, the first processor 2001 can execute various functions of the edge computing device 510 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0242] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 5 the CPU0 and CPU1 shown in

[0243] In a specific implementation, as an embodiment, the edge computing device 510 may also include multiple processors, such as Figure 5 the first processor 2001 and the second processor 2004 shown in

[0244] Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processors here may refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0245] Among them, the memory 2002 is used to store software programs for implementing the solution of the present invention and is controlled by the first processor 2001 for execution. The specific implementation manner can refer to the above method embodiments and will not be elaborated here.

[0245] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently and be coupled to the first processor 2001 through an interface circuit ( Figure 5 not shown in

[0246] The transceiver 2003 is used to communicate with network devices or terminal devices.

[0247] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 5 not shown separately). Among them, the receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0248] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently, and is coupled to the first processor 2001 through an interface circuit ( Figure 5 not shown) of the edge computing device 510. The embodiments of the present invention do not make specific limitations in this regard.

[0249] It should be noted that Figure 5 the structure of the edge computing device 510 shown does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0250] In addition, the technical effects of the edge computing device 510 can refer to the technical effects of the edge computing method for drone-assisted multi-task parallel execution described in the above method embodiments, and will not be elaborated here.

[0251] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0252] It should also be understood that the memory in the embodiments of the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink dynamic random access memory (SLDRAM), and direct rambus random access memory (DR RAM).

[0253] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0254] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this document generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood with reference to the context before and after.

[0255] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0256] It should be understood that in various embodiments of the present invention, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0257] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0258] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the devices, apparatuses, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0259] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0260] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0261] In addition, the functional units in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0262] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0263] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An edge computing method for multi-task parallel execution assisted by a drone, characterized in that: The method comprises: Obtaining a first computing task set of multiple mobile devices within the service range of the drone, an environmental state set of the drone, an available channel of the drone, an action space of the drone, and basic parameters of the drone; Based on the first computing task set, the drone performs cache optimization to obtain an optimized content cache of the drone; According to the optimized content cache set and the first computing task set, the plurality of mobile devices are screened to obtain a first category mobile device set and a second category mobile device set; Taking minimizing the total service time of the drone as the optimization goal, constructing a Markov decision process according to the optimized content cache, the first computing task set and the preset drone action strategy; Based on the Markov decision process and the deep Q network algorithm, according to the first computing task set, the second type of mobile device set is screened to obtain a third type of mobile device set and a fourth type of mobile device set; Taking maximizing the transmission rate of the drone as the optimization goal, a random game model is constructed according to the drone environment state set and the action space; Based on the random game model and according to the action space, the allocation of the available channels is iteratively optimized by a regret value minimization algorithm to obtain an optimal channel allocation; Based on the first category mobile device set, the third category mobile device set and the fourth category mobile device set, calculation is performed according to the drone basic parameters, the first computing task set and the optimal channel allocation to obtain the minimum total drone service time.

2. The edge computing method for multi-task parallel execution assisted by drone according to claim 1 is characterized in that: The drone performs cache optimization based on the first computing task set to obtain an optimized content cache of the drone, including: Sorting the first computing task set in ascending order according to the content quantity of each task in the first computing task set to obtain a sorted computing task set; According to the sorted computing task set, the drone performs content caching to obtain an optimized content cache set of the drone.

3. The edge computing method for multi-task parallel execution assisted by drone according to claim 1 is characterized in that: The step of screening the plurality of mobile devices according to the optimized content cache set and the first computing task set to obtain a first type of mobile device set and a second type of mobile device set includes: Obtaining an uploaded data set of a mobile device according to the optimized content cache set; According to the first computing task set, obtaining a data set to be calculated of the mobile device; Using the data set to be calculated, verifying the uploaded data set to obtain a verification result; Selecting mobile devices with consistent verification results from among the multiple mobile devices to obtain a first category of mobile device set; Among the multiple mobile devices, mobile devices with inconsistent verification results are selected to obtain a second type of mobile device set.

4. The edge computing method for multi-task parallel execution assisted by drone according to claim 1, characterized in that: The method of screening the second type of mobile device set based on the Markov decision process and the deep Q network algorithm according to the first computing task set to obtain the third type of mobile device set and the fourth type of mobile device set includes: Selecting tasks corresponding to the second type of mobile device set from the first computing task set to obtain a second computing task set; Based on the Markov decision process, according to the second computing task set, a deep Q network algorithm is used to select an offloading mode to obtain a second offloading mode set; From the second category mobile device set, select a mobile device whose offloading mode is offloading to a remote base station from the offloading mode set to obtain a third category mobile device set; In the second category of mobile devices, mobile devices whose offloading mode is not offloading to a remote base station in the offloading mode set are selected to obtain a fourth category of mobile devices.

5. The edge computing method for multi-task parallel execution assisted by drone according to claim 4 is characterized in that: The step of selecting an unloading mode based on the Markov decision process and the second computing task set using a deep Q network algorithm to obtain a second unloading mode set includes: According to the second computing task set, an ε-greedy strategy is used to select an offloading mode to obtain a first offloading mode set; Based on the Markov decision process, calculating according to the unloading pattern set to obtain an immediate reward set; Based on the instant reward set, the preset action value function and the preset loss function, using stochastic gradient descent to optimize the evaluation neural network parameters of the deep Q network algorithm to obtain optimized evaluation neural network parameters; According to the optimization and evaluation of the neural network parameters, the target neural network parameters of the deep Q network algorithm are updated to obtain updated target neural network parameters; Using the updated target neural network parameters, the preset unloading mode selection strategy is optimized to obtain the optimal selection strategy; According to the second computing task set, the optimal selection strategy is used to select an unloading mode to obtain a second unloading mode set.

6. The edge computing method for multi-task parallel execution assisted by drone according to claim 1, characterized in that: The iterative optimization of the allocation of the available channels based on the random game model and the action space by a regret value minimization algorithm to obtain the optimal channel allocation includes: Based on the random game model, calculation is performed according to the actions in the action space to obtain the action value of the game action space; According to the action value, the game is played through the random game model to obtain the historical accumulated regret; Iteratively optimizing the historical accumulated regret by using a regret value minimization algorithm to obtain the minimum historical accumulated regret; Calculate the channel strategy update rule according to the minimum historical accumulated regret; According to the channel strategy update rule, the preset channel allocation strategy is optimized to obtain the optimal allocation strategy; The available channels are allocated using the optimal allocation strategy to obtain optimal channel allocation.

7. The edge computing method for multi-task parallel execution assisted by drone according to claim 1, characterized in that: The calculating based on the first type of mobile device set, the third type of mobile device set and the fourth type of mobile device set according to the basic parameters of the drone, the first computing task set and the optimal channel allocation to obtain the minimum total drone service time includes: Based on the first type of mobile device set, calculating the drone service duration according to the first computing task set, the optimal channel allocation and the drone basic parameters, to obtain the first type of drone service duration; Based on the third type of mobile device set, the drone service duration is calculated according to the first computing task set, the optimal channel allocation and the drone basic parameters to obtain the second type of drone service duration; Based on the fourth type of mobile device set, the drone service duration is calculated according to the first computing task set, the optimal channel allocation and the drone basic parameters to obtain the third type of drone service duration; The minimum total drone service time is calculated based on the service time of the first category of drones, the service time of the second category of drones, and the service time of the third category of drones to obtain.

8. An edge computing device for drone-assisted multi-task parallel execution, the edge computing device for drone-assisted multi-task parallel execution is used to implement the edge computing method for drone-assisted multi-task parallel execution as claimed in any one of claims 1 to 7, characterized in that: The device comprises: A drone information acquisition module is used to acquire a first computing task set of multiple mobile devices within the drone service range, a drone environment state set, available channels of the drone, an action space of the drone, and basic parameters of the drone; A content cache optimization module, configured to optimize the cache of the drone based on the first computing task set to obtain an optimized content cache of the drone; A first device classification module, configured to screen the plurality of mobile devices according to the optimized content cache set and the first computing task set to obtain a first category mobile device set and a second category mobile device set; A Markov decision process construction module, used to construct a Markov decision process according to the optimized content cache, the first computing task set and a preset drone action strategy with minimizing the total drone service time as the optimization goal; A second device classification module is used to screen the second type of mobile device set according to the first computing task set based on the Markov decision process and the deep Q network algorithm to obtain a third type of mobile device set and a fourth type of mobile device set; A game model building module, used to build a random game model based on the combination of the drone environment state and the action space with maximizing the drone transmission rate and as the optimization goal; An optimal channel allocation module, configured to iteratively optimize the allocation of the available channels by a regret value minimization algorithm based on the random game model and the action space to obtain an optimal channel allocation; The total service time calculation module is used to calculate based on the first type of mobile device set, the third type of mobile device set and the fourth type of mobile device set, according to the basic parameters of the drone, the first computing task set and the optimal channel allocation, to obtain the minimum total service time of the drone.

9. An edge computing device, characterized in that: The edge computing device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images