A robustness-based multi-unmanned aerial vehicle assisted mobile edge computing method

Abstract

The application relates to a robust multi-unmanned aerial vehicle assisted mobile edge computing method. The method comprises the following steps: a terminal acquires task related information currently needing to be calculated, inputs the terminal task unloading strategy analysis model, outputs the best local calculation and task unloading strategy, the terminal requests to unload the task to the MEC server of the connected unmanned aerial vehicle for edge calculation through an uplink channel according to the best local calculation and task unloading strategy, the MEC server of the unmanned aerial vehicle returns the calculation result to the terminal, the environment parameters acquired by the MEC server of the unmanned aerial vehicle are input into the unmanned aerial vehicle edge calculation strategy analysis model, and a CPU calculation resource allocation result and a flight trajectory are output, the MEC server of the unmanned aerial vehicle allocates the CPU calculation resource of edge calculation to the terminal according to the CPU calculation resource allocation result, and controls the flight trajectory of the unmanned aerial vehicle according to the flight trajectory. The energy efficiency and performance of the multi-unmanned aerial vehicle assisted MEC system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mobile edge computing, in particular to a multi-unmanned aerial vehicle assisted mobile edge computing method based on robustness. BACKGROUND

[0002] With the rapid development of Internet of Things, cloud computing and big data, the number of computationally intensive applications is gradually increasing, such as autonomous driving, virtual or augmented reality tasks, etc. However, due to limited computing resources and battery capacity, the local computing of mobile devices is difficult to provide satisfactory service quality. Giving the computing task to a centralized server can alleviate this problem, but network congestion and high delay cannot be avoided. In view of this, mobile edge computing (MEC) as a forward-looking solution has attracted great attention. Using MEC technology to offload tasks to edge servers can save energy consumption and improve computing speed. In addition, unmanned aerial vehicles (UAVs) are considered to be used in the field of wireless communication due to their low cost and high mobility. Combined with multi-unmanned aerial vehicle assisted MEC, more flexible mobile computing services are supported, and part of the user's computing task is offloaded to the edge server of the unmanned aerial vehicle to solve the delay and energy consumption problem caused by the distance between the cloud computing data center and the terminal device. Therefore, the combination of unmanned aerial vehicles and MEC is of great significance to the development of Internet of Things.

[0003] In the related art, in the current multi-unmanned aerial vehicle assisted MEC system, most of the inventions focus on the optimization target of task offloading on the theoretical energy consumption and delay of the system, and ignore the uncertainty factors such as the complexity of the task and the stability of the communication in actual application. Among them, the incompleteness of channel state information (CSI) and the inaccuracy of task complexity estimation will have an adverse effect on the system, which is easy to cause the energy efficiency and performance of the multi-unmanned aerial vehicle assisted MEC system to be low. SUMMARY

[0004] Therefore, it is necessary to provide a multi-unmanned aerial vehicle assisted mobile edge computing method based on robustness, which can improve the energy efficiency and performance of the multi-unmanned aerial vehicle assisted MEC system.

[0005] A multi-unmanned aerial vehicle assisted mobile edge computing method based on robustness, the method comprising:

[0006] The terminal obtains task-related information currently required for calculation and inputs the terminal task offloading strategy analysis model. The terminal task offloading strategy analysis model solves a target function by using a MAPPO algorithm according to the task-related information, and outputs an optimal local calculation and task offloading strategy in combination with a constraint condition.

[0007] The terminal offloads a task required for calculation to a MEC server of a UAV according to the optimal local calculation and task offloading strategy, and requests the task to be offloaded to the connected MEC server of the UAV for edge calculation through an uplink channel.

[0008] The MEC server of the UAV returns a calculation result to the terminal through a downlink after completing the task calculation.

[0009] The MEC server of the UAV obtains environmental parameters and inputs the UAV edge calculation strategy analysis model. The UAV edge calculation strategy analysis model solves the target function by using the MAPPO algorithm according to the environmental parameters, and outputs a CPU calculation resource allocation result and a flight trajectory in combination with the constraint condition.

[0010] The MEC server of the UAV allocates edge calculation CPU calculation resources to terminals in a coverage range according to the CPU calculation resource allocation result, controls a flight trajectory of the UAV according to the flight trajectory, and changes the coverage range of the MEC server of the UAV.

[0011] In one of the embodiments, the terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model each use an Actor-Critic policy network, and are trained by using a CTDE policy. The training is performed by using global information of a terminal cluster and a UAV cluster, and the training manner is as follows:

[0012] During the training process, the Actor-Critic policy networks of the terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model share a global Critic network for estimating a state value function. The global Critic network is a centralized training part in the CTDE policy.

[0013] During the generation of actions, the Actor networks of the terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model each observe local information of itself, which is a decentralized execution part in the CTDE policy.

[0014] For the terminal cluster, observation information of a kth user at a tth moment is , and a corresponding action is . ​ ; For drone swarms, A drone in The observation information at the moment is , the corresponding action is ;

[0015] The terminal task offloading strategy analysis model and the drone edge computing strategy analysis model update their own strategies based on the global Critic network. The strategies updated by the terminal offloading strategy analysis model are local computing and task offloading strategies, while the strategies updated by the drone edge computing strategy analysis model are CPU computing resource allocation results and flight trajectories.

[0016] In one embodiment, the objective function is:

[0017]

[0018] in, is the weighted energy consumption, For the The energy consumption generated by local calculation of the user terminal in a time slot is: For the In a time slot, the energy consumption generated by task offloading during communication and the energy consumption generated by MEC server calculation are For the In a time slot, the energy consumption generated by task offloading during communication and the energy consumption generated by MEC server calculation are is the energy consumption generated by the UAV flight, and N is the time period The total number of time slots within.

[0019] In one embodiment, the constraint condition is:

[0020]

[0021]

[0022] ,

[0023] in, For the The task calculation delay of each MU, is the estimation error of task complexity, is the random error term, For a given acceptable range of overall delay, For terminal collection, For time period The set of time slots within is the estimated value of the channel gain, is a range of random error terms, is a true task complexity, is an estimated task complexity, is an error interval of the task complexity, is a set of different types of tasks.

[0024] The above robust-based multi-unmanned aerial vehicle assisted mobile edge computing method, through the terminal, acquires task-related information input into a terminal task offloading strategy analysis model, which, according to the task-related information, solves a target function by using a MAPPO algorithm, and, in combination with a constraint condition, outputs an optimal local computing and task offloading strategy. The terminal, according to the optimal local computing and task offloading strategy, requests, through an uplink channel, to offload a task that needs to be calculated on a MEC server of an unmanned aerial vehicle to the connected MEC server of the unmanned aerial vehicle for edge computing. The MEC server of the unmanned aerial vehicle returns the calculation result to the terminal through a downlink after completing the task calculation. The environment parameters acquired by the MEC server of the unmanned aerial vehicle are input into an unmanned aerial vehicle edge computing strategy analysis model. The unmanned aerial vehicle edge computing strategy analysis model, according to the environment parameters, solves the target function by using the MAPPO algorithm, and, in combination with the constraint condition, outputs a CPU computing resource allocation result and a flight trajectory. The MEC server of the unmanned aerial vehicle allocates the CPU computing resource for edge computing according to the CPU computing resource allocation result for the terminals within the coverage range, controls the flight trajectory of the unmanned aerial vehicle according to the flight trajectory, and changes the coverage range of the MEC server of the unmanned aerial vehicle. In this way, the energy efficiency and performance of the multi-unmanned aerial vehicle assisted MEC system are improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a schematic diagram of an application environment of the robust-based multi-unmanned aerial vehicle assisted mobile edge computing method in an embodiment;

[0026] Figure 2 is a schematic diagram of a flow of the robust-based multi-unmanned aerial vehicle assisted mobile edge computing method in an embodiment;

[0027] Figure 3 is a schematic diagram of a flow of training the terminal task offloading strategy analysis model and the unmanned aerial vehicle edge computing strategy analysis model by using the CTDE strategy in an embodiment. DETAILED DESCRIPTION

[0028] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0029] The application provides a robust multi-UAV assisted mobile edge computing method, which can be applied to a multi-UAV assisted MEC network environment, an edge server computing environment, a UAV and terminal communication and computing environment, and the like.

[0030] The multi-UAV assisted MEC network environment includes a terminal set and an edge computing server set; the terminal set includes terminal users such as smart phones, mobile tablets, unmanned vehicles, virtual or augmented reality devices, and the like; and the edge computing server set includes a UAV group carrying an MEC server. In each time slot, the UAV carrying the MEC server can communicate with multiple terminals on the ground, and each terminal communicates with at most one UAV. The edge computing provided by the UAV MEC server can be performed simultaneously with the local computing of the terminal.

[0031] The edge server computing environment includes an edge computing server and a UAV multi-agent system (multi-agent system). In the multi-agent reinforcement learning (MARL) environment, two types of agents are included, namely, a UAV cluster and a terminal cluster. The optimal computing task offloading strategy is given by a deep reinforcement learning method of multi-agent proximal policy optimization (MAPPO).

[0032] The UAV and terminal communication environment includes uplink of the terminal offloading a task to the MEC server of the UAV, and downlink of the UAV returning a computing result to the terminal. The computing environment includes local computing of the terminal and offloading to the UAV for edge computing. In the environment, there is non-negligible communication and computing uncertainty.

[0033] In one embodiment, as shown in Figure 1 The multi-UAV assisted MEC network environment includes, but is not limited to, terminal devices such as smart phones and mobile tablets, and a UAV group carrying an MEC server. The process of communication between the ground terminal and the UAV is divided into multiple time slots with equal intervals, which are used as the basic unit of communication and computing between the UAV and the terminal. In each time slot, the UAV can communicate with multiple terminal users on the ground at the same time, and in the time slot, each terminal communicates with at most one UAV.

[0034] In one embodiment, as shown in Figure 2 A robust multi-UAV assisted mobile edge computing method is provided, including the following steps:

[0035] In step S220, the terminal obtains information related to the task that needs to be calculated and inputs it into the terminal task offloading strategy analysis model. The terminal task offloading strategy analysis model uses the MAPPO algorithm to solve the objective function based on the task related information, and combines the constraints to output the optimal local calculation and task offloading strategy.

[0036] In one embodiment, the objective function is:

[0037]

[0038] in, is the weighted energy consumption, For the The energy consumption generated by local calculation of the user terminal in a time slot is: For the In a time slot, the energy consumption generated by task offloading during communication and the energy consumption generated by MEC server calculation are For the In a time slot, the energy consumption generated by task offloading during communication and the energy consumption generated by MEC server calculation are is the energy consumption generated by the UAV flight, and N is the time period The total number of time slots within.

[0039] In one embodiment, the constraints are:

[0040]

[0041]

[0042] ,

[0043] in, For the The task calculation delay of each MU, is the estimation error of task complexity, is the random error term, For a given acceptable range of overall delay, For terminal collection, For time period The set of time slots within is the estimated value of the channel gain, is the range of the random error term, For the actual task complexity, is the estimated task complexity, is the error interval of task complexity, A collection of tasks of different types.

[0044] Step S240, the terminal requests to offload the task to the MEC server of the connected UAV for edge computing according to the optimal local computation and task offloading strategy.

[0045] Step S260, the MEC server of the UAV returns the calculation result to the terminal through the downlink after completing the task computation.

[0046] Step S280, the MEC server of the UAV inputs the environment parameters into the UAV edge computing strategy analysis model, and the UAV edge computing strategy analysis model solves the objective function by using the MAPPO algorithm according to the environment parameters, and outputs the CPU computing resource allocation result and the flight trajectory in combination with the constraint condition.

[0047] In one embodiment, the objective function is:

[0048]

[0049] wherein, is the weighted energy consumption, is the energy consumption generated by the local computation of the user terminal in the th time slot, is the energy consumption generated by the task offloading and the energy consumption generated by the MEC server computation in the communication process in the th time slot, is the energy consumption generated by the task offloading and the energy consumption generated by the MEC server computation in the communication process in the th time slot, is the energy consumption generated by the UAV flight, and N is the total number of time slots in the time period.

[0050] In one embodiment, the constraint condition is:

[0051]

[0052]

[0053] ,

[0054] wherein, is the task computation delay of the th MU, is the estimation error of the task complexity, is a random error term, is the given overall delay acceptable range, is the terminal set, is the time period ​The set of time slots within is the estimated value of the channel gain, is the range of the random error term, For the actual task complexity, is the estimated task complexity, is the error interval of task complexity, A collection of tasks of different types.

[0055] In one embodiment, the terminal task offloading strategy analysis model and the drone edge computing strategy analysis model each use an Actor-Critic strategy network, and the CTDE strategy is used to train the terminal task offloading strategy analysis model and the drone edge computing strategy analysis model. During training, the global information of the terminal cluster and the drone cluster is used for training. The training method is:

[0056] During the training process, the Actor-Critic strategy network of the terminal task offloading strategy analysis model and the UAV edge computing strategy analysis model shares a global Critic network for estimating the state value function. The global Critic network is the centralized training part of the CTDE strategy. When generating actions, the Actor network of the terminal task offloading strategy analysis model and the UAV edge computing strategy analysis model each observes its own local information, which is the decentralized execution part of the CTDE strategy. For the terminal cluster, Users in The observation information at the moment is , the corresponding action is ; For drone swarms, A drone in The observation information at the moment is , the corresponding action is The terminal task offloading strategy analysis model and the drone edge computing strategy analysis model update their own strategies based on the global Critic network. The strategies updated by the terminal offloading strategy analysis model are local computing and task offloading strategies, and the strategies updated by the drone edge computing strategy analysis model are CPU computing resource allocation results and flight trajectories.

[0057] It should be understood that the terminal task offloading strategy analysis model and the drone edge computing strategy analysis model use the MAPPO algorithm to weight the energy consumption of terminals and drones, improving the energy efficiency and performance of multi-drone-assisted MEC systems. These models also consider various errors and uncertainties that exist in actual multi-drone-assisted MEC systems, improving the robustness and reliability of the system optimization algorithm in complex environments.

[0058] In step S300, the MEC server of the UAV allocates CPU computing resources for edge computing for terminals within the coverage range according to the CPU computing resource allocation result, controls the flight trajectory of the UAV according to the flight trajectory, and changes the coverage range of the MEC server of the UAV.

[0059] The above robust-based multi-UAV assisted mobile edge computing method, by the terminal obtaining the task-related information currently needing to be calculated, inputs the terminal task offloading strategy analysis model, the terminal task offloading strategy analysis model solves the objective function by using the MAPPO algorithm according to the task-related information, and outputs the best local computing and task offloading strategy in combination with the constraint condition, the terminal according to the best local computing and task offloading strategy, the task needing to be offloaded to the MEC server of the UAV for calculation is requested to be offloaded to the connected MEC server of the UAV through the uplink channel for edge computing, the MEC server of the UAV returns the calculation result to the terminal through the downlink after completing the task calculation, the environment parameters obtained by the MEC server of the UAV are input into the UAV edge computing strategy analysis model, the UAV edge computing strategy analysis model solves the objective function by using the MAPPO algorithm according to the environment parameters, and outputs the CPU computing resource allocation result and the flight trajectory in combination with the constraint condition, the MEC server of the UAV allocates the CPU computing resources for edge computing for terminals within the coverage range according to the CPU computing resource allocation result, controls the flight trajectory of the UAV according to the flight trajectory, and changes the coverage range of the MEC server of the UAV. Therefore, the energy efficiency and performance of the multi-UAV assisted MEC system are improved.

[0060] In one embodiment, a robust-based multi-UAV assisted mobile edge computing method specifically includes the following steps:

[0061] Step (1): The UAV equipped with the MEC server flies above the terminal, and the multi-agent system of the terminal on the ground (the multi-agent system includes a terminal task offloading strategy analysis model) determines the best local computing and task offloading strategy, performs local computing on a part of the tasks, and requests to offload another part of the computing tasks to the MEC server of the adjacent UAV through the uplink channel for edge computing.

[0062] Step (2): the multi-agent system of the terminal and the multi-agent system of the UAV (which includes a UAV edge computing strategy analysis model) obtain the changed state parameters and perform strategy analysis with the communication and computing uncertainty as constraint conditions. The communication and computing uncertainty as constraint conditions enables the addition of a robustness design algorithm to the UAV edge computing strategy analysis model and the terminal task offloading strategy analysis model, considers various error and uncertainty factors existing in the actual multi-UAV assisted MEC system, and improves the robustness and reliability of the system optimization algorithm in a complex environment.

[0063] Step (3): the multi-agent system of the UAV determines the CPU computing resource allocation result and the flight trajectory under the premise of ensuring the task delay requirement; the UAV MEC server allocates computing resources, channel bandwidth and the like according to the determined CPU computing resource allocation result, and controls the flight trajectory of the UAV according to the flight trajectory to change the coverage range of the MEC server of the UAV.

[0064] Step (4): the UAV returns the completed computing task to the corresponding terminal through a downlink.

[0065] Further, the environment in step (1) includes the following features: the environment adopts a three-dimensional rectangular coordinate system, all UAVs fly at a fixed height H, and each terminal is located at a height of 0 and is randomly distributed on the ground. The terminal moves on the ground according to a Gaussian-Markov random model. Considering that the interval between each time slot is very short, it can be assumed that the terminal is static within each time slot. In addition, the types of tasks uploaded by the terminal are different, and the computing complexities of different types of tasks are also different.

[0066] Further, the environment parameters obtained by the multi-agent of the multi-UAV in step (2) include the flight speed and position of each UAV, the position of the terminal user, the task sending power, the connection information between each terminal user and the edge server, the CPU computing resource allocated to the terminal by the edge server, and the channel transmission bandwidth. In addition, in the actual edge intelligent network, the factors that need to be considered for robustness design include the following aspects:

[0067] (1) Communication uncertainty: when the terminal communicates with the MEC server of the UAV, due to limited feedback, channel estimation or quantization errors, communication uncertainty is caused; the uncertainty affects data transmission;

[0068] (2) Computational uncertainty: Since the complexity of the computational task cannot usually be obtained in advance, computational uncertainty arises; however, the expected value of the task complexity can be reasonably estimated from the long-term statistical data of a specific task type; this certainty affects data calculations.

[0069] Furthermore, in step (3), the optimal computing and offloading strategy is determined by the intelligent agent system using the MAPPO algorithm and centralized training and decentralized execution (CTDE). The optimization goal is to minimize the weighted energy consumption of the terminal and the UAV while ensuring the mission latency requirements.

[0070] like Figure 3 As shown in Figure 2, each terminal needs to complete different types of computing tasks and submit some tasks to the drone's MEC server for processing. Represents a collection of terminals, using Represents a collection of drones and uses Represents a set of tasks of different types. Without loss of generality, we introduce a time period , which is divided into equal time slots, each time slot is , so the time period is In addition, in the three-dimensional rectangular coordinate system, the drone flies at a fixed height , the terminal height on the ground is 0.

[0071] Among them, task offloading is partial offloading, that is, in each time slot, the terminal can choose to execute the computing task locally or offload it to the MEC server for processing, and both can be performed at the same time.

[0072] Among them, local computing: If the terminal chooses to perform computing tasks locally, the local computing delay of the terminal can be expressed as ,in , Indicates that the task is calculated locally.

[0073] Among them, edge computing: If the terminal chooses to offload the computing task to the drone for processing, the MEC server will calculate it and return the result to the terminal. The computing delay can be expressed as ,in, , Indicates task offload (offload), Indicates that the task is edge-computing on the drone. denotes the time that the terminal takes to offload the task to the UAV, denotes the time that the UAV takes to compute the task. The UAV computation is performed after the task is offloaded to the UAV.

[0074] Further, since the terminal can perform the computation locally and offload to the UAV for edge computation simultaneously, the multi-UAV assisted MEC system can serve the task computation latency of the th terminal in the th time slot is denoted as:

[0075] .

[0076] where, for the robustness optimization of communication and computation uncertainty, considering that in the actual edge intelligent network, there exists computation uncertainty caused by different task complexities and communication uncertainty caused by incomplete CSI, the multi-UAV assisted MEC system is subjected to robustness constraints:

[0077] On the one hand, for the computation task , the exact value of the computation data volume can be obtained by the task analyzer before the computation task is executed, while the task complexity (i.e. computation uncertainty) cannot be obtained. Before the task is processed, its processing time is usually uncertain. However, the expected value of the computation uncertainty can be reasonably estimated from the long-term statistical data of a specific task type. Accordingly, the real task complexity can be denoted as:

[0078] ,

[0079] where, denotes the estimated task complexity, is the estimation error of the task complexity, which is within the error interval of the given task complexity .

[0080] On the other hand, due to limited feedback, channel estimation or quantization error, i.e. communication uncertainty, it is impossible to obtain accurate and error-free CSI in the actual edge intelligent network. For the uncertainty of communication, we adopt the widely used deterministic imperfect CSI model as the robustness constraint, which is denoted as:

[0081]

[0082] where, , is an estimate of the channel gain, is a random error term whose norm is within a given range of random error terms .

[0083] Adding these two robustness constraints can better describe the situation in the actual communication system, and improve the robustness and reliability of the multi-UAV assisted MEC system design and optimization algorithm. Accordingly, the constraint of the overall system delay is expressed as:

[0084] .

[0085] Wherein, the MAPPO algorithm is used to solve the objective function, and the optimization objective of the objective function is to minimize the weighted energy consumption of the terminal set and the UAV set, and the weighted energy consumption can be expressed as:

[0086]

[0087] wherein, is the weighted energy consumption, is the energy consumption generated by the user terminal local calculation in the first time slot, is the energy consumption generated by the task offloading in the communication process and the energy consumption generated by the MEC server calculation in the first time slot, is the energy consumption generated by the task offloading in the communication process and the energy consumption generated by the MEC server calculation in the first time slot, is the energy consumption generated by the UAV flight, and N is the total number of time slots in the time period .

[0088] In this embodiment, there are two types of agents, namely the UAV cluster and the terminal cluster, and each type of agent has its own Actor-Critic policy network. CTDE is used as the training and execution strategy, which uses the global information of the UAV cluster and the terminal cluster for training, and each agent can only observe its own local information during execution.

[0089] The CTDE strategy is specifically:

[0090] (1) During the training process, the Actor-Critic networks of the UAV cluster and the terminal cluster will share a global Critic network for estimating the state value function . The global Critic network is the centralized training part in CTDE.

[0091] (2) When generating actions, each agent's Actor network can only observe its own local information, but cannot observe the information of other agents, which is the decentralized execution part in CTDE. Users in The observation information at time is , the corresponding action is ; For drone swarms, A drone in The observation information at time is , the corresponding action is .

[0092] (3) During training, each agent's Actor network updates its own strategy based on the global Critic network. For the terminal cluster, this is the local computation and task offloading strategy; for the drone cluster, this is the CPU resource allocation results and flight trajectory. This update process is based on information from all terminals and drones.

[0093] (4) During execution, each agent observes its own local information and generates corresponding actions based on its own Actor network.

[0094] Based on the terminal task offload strategy analysis model, local computing and task offload strategies are determined. Furthermore, the drone edge computing strategy analysis model is used to determine CPU computing resource allocation and flight trajectory. The drone's MEC server allocates the corresponding resources, completes the computing tasks, and returns the results to the terminal via a downlink. Throughout this process, communication and computing uncertainties are considered, enhancing the robustness of the model to achieve efficient resource utilization, meet user requirements for low latency and high reliability, and improve the performance of edge intelligent networks.

[0095] It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0096] Any combination of the technical features in the above embodiments can be made, and for the sake of brevity, not all possible combinations are described above, however, as long as the combination of the technical features does not exist in contradiction, it shall be considered within the scope of the present disclosure.

[0097] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it shall not be understood as a limitation on the patent scope of the present application. It shall be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these shall be within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A robustness-based multi-UAV-assisted mobile edge computing method, characterized in that, The method comprises: The terminal acquires task-related information currently required for calculation and inputs the terminal task offloading strategy analysis model, the terminal task offloading strategy analysis model solves a target function by using a MAPPO algorithm according to the task-related information, and outputs an optimal local calculation and task offloading strategy in combination with a constraint condition; The terminal offloads a task required to be calculated on a MEC server of a UAV to the MEC server of the UAV through an uplink channel according to the optimal local calculation and task offloading strategy, and requests the MEC server of the UAV to perform edge calculation; The MEC server of the UAV returns a calculation result to the terminal through a downlink after completing the calculation of the task; The MEC server of the UAV acquires environmental parameters and inputs the environmental parameters into a UAV edge calculation strategy analysis model, the UAV edge calculation strategy analysis model solves the target function by using the MAPPO algorithm according to the environmental parameters, and outputs a CPU calculation resource allocation result and a flight trajectory in combination with the constraint condition; The MEC server of the UAV allocates the CPU calculation resource for edge calculation to terminals in a coverage range according to the CPU calculation resource allocation result, controls a flight trajectory of the UAV according to the flight trajectory, and changes the coverage range of the MEC server of the UAV.

2. The method of claim 1, wherein, The terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model each use an Actor-Critic strategy network, and are trained by using a CTDE strategy, and global information of a terminal cluster and a UAV cluster is used for training during training, and the training mode is as follows: During training, the Actor-Critic strategy networks of the terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model share a global Critic network for estimating a state value function, and the global Critic network is a centralized training part in the CTDE strategy; During action generation, the Actor networks of the terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model each observe local information of itself, that is, a decentralized execution part in the CTDE strategy; For terminal clusters, Users in The observation information at the moment is , the corresponding action is ; For drone swarms, A drone in The observation information at the moment is , the corresponding action is ; The terminal task offloading strategy analysis model and the UAV edge calculation strategy analysis model update their own strategies according to the global Critic network, wherein the strategy updated by the terminal task offloading strategy analysis model is a local calculation and task offloading strategy, and the strategy updated by the UAV edge calculation strategy analysis model is a CPU calculation resource allocation result and a flight trajectory.

3. The method of claim 2, wherein, The target function is: wherein, is the weighted energy consumption, is the energy consumption produced by the user terminal locally calculated in the first is the energy consumption produced by the user terminal locally calculated in the first is the energy consumption produced by the task offloading in the communication process and the energy consumption produced by the MEC server calculation in the first is the energy consumption produced by the task offloading in the communication process and the energy consumption produced by the MEC server calculation in the first is the energy consumption produced by the task offloading in the communication process and the energy consumption produced by the MEC server calculation in the first is the energy consumption produced by the task offloading in the communication process and the energy consumption produced by the MEC server calculation in the first is the energy consumption produced by the UAV flight, N is the time period the total number of slots within the time period.

4. The method of claim 3, wherein, The constraint condition is: , in, For the The task calculation delay of each MU, is the estimation error of task complexity, is the random error term, For a given acceptable range of overall delay, For terminal collection, For time period The set of time slots within is the estimated value of the channel gain, is the range of the random error term, For the actual task complexity, is the estimated task complexity, is the error interval of task complexity, A collection of tasks of different types.

Images