Power inspection unmanned aerial vehicle edge computing task offloading method and device

Abstract

The application provides a power inspection unmanned aerial vehicle edge computing task offloading method and device, and relates to the technical field of mobile edge computing. The method comprises the following steps: acquiring the position of a target time slot power inspection unmanned aerial vehicle, the position of each power equipment and the task amount of the edge computing task of each power equipment; inputting the position of the target time slot power inspection unmanned aerial vehicle, the position of each power equipment and the task amount of the edge computing task of each power equipment into a first model corresponding to the power inspection unmanned aerial vehicle, acquiring the offloading instruction of each power equipment in the target time slot and the flight control parameter of the target time slot power inspection unmanned aerial vehicle; and based on the offloading instruction, performing edge computing task offloading on a target power equipment in each power equipment in the target time slot. The power inspection unmanned aerial vehicle edge computing task offloading method and device provided by the application can more effectively meet the edge computing task offloading demand of the ground power equipment, improve the service quality and reduce the task delay.

Description

Technical Field

[0001] This invention relates to the field of mobile edge computing (MEC) technology, and in particular to a method and apparatus for offloading edge computing tasks from a power line inspection drone. Background Technology

[0002] To cope with the rapid growth of business data and traffic generated by power equipment, computing and storage resources can be deployed at the network edge close to the power equipment. Based on edge computing technology, computing tasks can be offloaded to MEC servers, effectively handling the business of power equipment, reducing the load on the power system, and reducing the transmission overhead of task requests from power equipment. However, under dynamic, real-time task requests, ground-based fixed MEC servers may not be able to meet the service needs of some power equipment. Power inspection drones (hereinafter referred to as "drones") equipped with MEC servers can improve the service quality of power equipment based on real-time business requests, but the maneuverability of drones will bring additional energy consumption, and the allocation of offloading tasks is also a significant challenge when power equipment issues massive requests and multiple drones provide services together.

[0003] Existing reinforcement learning-based methods for offloading edge computing tasks from power plant drones typically involve training a drone agent offline using reinforcement learning algorithms to obtain the optimal strategy for task offloading. However, the optimal strategy obtained by these methods performs poorly in real-world deployment environments and is difficult to effectively meet the edge computing task offloading requirements of ground-based power equipment. Summary of the Invention

[0004] This invention provides a method and apparatus for offloading edge computing tasks from a power inspection drone, which addresses the shortcomings of existing technologies that cannot effectively meet the offloading requirements of edge computing tasks from ground-based power equipment, and achieves effective offloading of edge computing tasks from ground-based power equipment.

[0005] This invention provides a method for offloading edge computing tasks from a power line inspection drone, used in power line inspection drones, the method comprising:

[0006] The location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task for each power device are obtained.

[0007] The location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device are input into the first model corresponding to the power inspection drone to obtain the unloading instruction for each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot.

[0008] Based on the unloading instruction, the edge computing task is unloaded for the target power equipment in each of the power equipment in the target time slot, and the flight of the power inspection drone is controlled based on the flight control parameters.

[0009] The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot; the unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0010] According to the present invention, a method for offloading edge computing tasks from a power line inspection drone, before inputting the location of the power line inspection drone, the location of each power device, and the workload of the edge computing tasks of each power device in the target time slot into the first model corresponding to the power line inspection drone, and obtaining the offloading instruction for each power device in the target time slot and the flight control parameters of the power line inspection drone in the target time slot, the method further includes:

[0011] Based on the MATD3 algorithm and a preset optimization objective, the generated sample data is trained offline.

[0012] According to the present invention, a method for offloading edge computing tasks from a power line inspection drone, before performing offline training on the generated sample data based on the MATD3 algorithm and a preset optimization objective, further includes:

[0013] Acquire historical operational data of various power inspection drones and power equipment in a real deployment environment;

[0014] Based on the historical operational data, the generated sample data is obtained according to the generative adversarial network algorithm.

[0015] According to the present invention, a method for offloading edge computing tasks from a power inspection drone is provided, wherein the optimization objective includes minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of processing the offloaded edge computing tasks.

[0016] This invention also provides a method for offloading edge computing tasks from a power inspection drone, for use with power equipment, the method comprising:

[0017] Obtain the workload and idle computing resources of the edge computing tasks of the power equipment in the target time slot;

[0018] The task volume and idle computing resources of the edge computing task of the power equipment in the target time slot are input into the second model corresponding to the power equipment to obtain the offloading rate of the edge computing task of the power equipment in the target time slot, so that the target power inspection drone can offload the edge computing task of the power equipment in the target time slot based on the offloading rate.

[0019] The unloading instruction from the target power inspection drone to the power equipment in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot. The unloading instruction is determined based on the first model corresponding to the target power inspection drone, the location of the target power inspection drone in the target time slot, the location of each power equipment, and the workload of the edge computing tasks of each power equipment. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0020] The present invention also provides an edge computing task offloading device for power inspection drones, comprising:

[0021] The first acquisition module is used to acquire the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device.

[0022] The second acquisition module is used for the location and the workload of edge computing tasks of each of the power devices. It inputs the first model corresponding to the power inspection drone, and acquires the unloading instruction of each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot.

[0023] The task offloading module is used to perform edge computing task offloading on the target power equipment in each of the power equipment in the target time slot based on the offloading instruction, and to control the flight of the power inspection drone based on the flight control parameters.

[0024] The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot; the unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0025] The present invention also provides an edge computing task offloading device for power inspection drones, comprising:

[0026] The third acquisition module is used to acquire the task volume and idle computing resources of the edge computing tasks of the target time slot power equipment;

[0027] The fourth acquisition module is used to input the task volume and idle computing resources of the edge computing task of the power equipment in the target time slot into the second model corresponding to the power equipment, and obtain the offloading rate of the edge computing task of the power equipment in the target time slot, so that the target power inspection drone can offload the edge computing task of the power equipment in the target time slot based on the offloading rate.

[0028] The unloading instruction from the target power inspection drone to the power equipment in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot. The unloading instruction is determined based on the first model corresponding to the target power inspection drone, the location of the target power inspection drone in the target time slot, the location of each power equipment, and the workload of the edge computing tasks of each power equipment. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0029] The present invention also provides an edge computing task offloading system for power inspection drones, comprising: a plurality of edge computing task offloading devices for power inspection drones as described above and a plurality of edge computing task offloading devices for power inspection drones as described above.

[0030] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the edge computing task offloading method for power inspection drones as described above.

[0031] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the edge computing task offloading method for power inspection drones as described above.

[0032] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the edge computing task offloading method for power inspection drones as described above.

[0033] The present invention provides a method and apparatus for offloading edge computing tasks from a power inspection drone. This method utilizes a flight control model for the power inspection drone and an offloading model for edge computing tasks from the power equipment, based on the collaboration between the power inspection drone and the power equipment—two heterogeneous intelligent agents. It makes decisions based on the workload of the power equipment's edge computing tasks, determining the offloading rate and the power inspection drone's offloading instruction to the power equipment. This allows a portion of the tasks to be computed locally on the power equipment to fully utilize its computing resources, while the remaining portion is offloaded to the power inspection drone according to the offloading rate. This approach more effectively meets the offloading needs of edge computing tasks from ground-based power equipment, improves service quality, and reduces task latency and energy consumption of the power inspection drone. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0035] Figure 1 This is one of the flowcharts illustrating the edge computing task offloading method for power inspection drones provided by the present invention;

[0036] Figure 2 This is a schematic diagram of the acquisition process of each first model and each second model in the edge computing task offloading method of the power inspection drone provided by the present invention;

[0037] Figure 3 This is the second flowchart illustrating the edge computing task offloading method for power inspection drones provided by this invention.

[0038] Figure 4 This is one of the structural schematic diagrams of the edge computing task unloading device for power inspection drones provided by the present invention;

[0039] Figure 5 This is the second schematic diagram of the edge computing task unloading device for power inspection drones provided by the present invention;

[0040] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0042] The following is combined Figures 1 to 6 This invention describes a method and apparatus for offloading edge computing tasks from a power line inspection drone.

[0043] To facilitate understanding of the following embodiments of the present invention, the relevant knowledge involved in the embodiments of the present invention will be described and explained below.

[0044] The edge computing task offloading method for power line inspection drones provided in various embodiments of the present invention is applicable to a power line inspection drone edge computing task offloading system. This system may include multiple power line inspection drones and multiple power devices. The power line inspection drones can be used to inspect power devices. Each power line inspection drone is equipped with an MEC server, allowing it to offload edge computing tasks from the power devices via its onboard MEC server, thereby improving service quality and reducing task latency.

[0045] The edge computing task offloading method for power inspection drones provided in various embodiments of the present invention is a joint optimization mechanism for the collaborative offloading of edge computing tasks between power inspection drones and power equipment. In the method, each power inspection drone can handle edge computing tasks of multiple power equipment and maneuver according to real-time task requests to improve service quality; the offloading rate of edge computing tasks of power equipment is determined by its own idle computing resources and task size to further reduce task latency.

[0046] The edge computing task offloading method for power inspection drones provided in various embodiments of the present invention allows power equipment to leave some edge computing tasks for local computing to make full use of its own computing resources, and to offload some edge computing tasks that the power equipment cannot handle to the power inspection drone equipped with an MEC server. Furthermore, the power inspection drone maneuvers according to real-time dynamic task requests to improve service quality and reduce task latency. This method can solve the problem that traditional ground-based fixed MEC servers cannot fully meet the dynamic real-time task offloading needs of power equipment.

[0047] The power line inspection drone edge computing task offloading system can include multiple power line inspection drones {1,2,…,N} and multiple power equipment {1,2,…,M}. Here, N and M are both positive integers.

[0048] The offloading instruction a for the edge computing task of a power inspection drone n (1≤n≤N, n is a positive integer) on power equipment m (1≤m≤M, m is a positive integer) in time slot t. n，m (t)∈{0,1}, used to indicate the allocation of edge computing tasks to power equipment m by the power inspection drone n. n，m (t) = 1, indicating that time slot t provides services to power equipment m for power inspection drone n, that is, time slot t unloads the edge computing tasks of power equipment m from power equipment m; a n，m (t) = 0, indicating that the power inspection drone n in time slot t does not provide services to the power equipment m, that is, the power inspection drone n in time slot t does not unload the edge computing tasks of the power equipment m.

[0049] For any unloading power device m, once the assignment of the unloading edge computing task is determined, the task can only be performed by the assigned power inspection drone.

[0050]

[0051] The above formula shows that for any unloading power device m, the edge computing task of that power device can be unloaded by at most one power inspection drone. And any power inspection drone can unload the edge computing tasks of zero, one or more power devices.

[0052] Optionally, the positions of the power inspection drone n and the power equipment m can be represented by three-dimensional coordinates:

[0053] P n =[x n y n , z n ], P m =[x m y n ,0]

[0054] Among them, P n P represents the location of the power line inspection drone n. m This represents the location of the power equipment m. Alternatively, it can be assumed that the power inspection drone flies at a fixed altitude, i.e., z. n It is a constant greater than 0, and the height of all electrical equipment is 0.

[0055] Since the drone flies at a fixed altitude, the motion of the power line inspection drone n can be determined by the azimuth angle φ. n ∈[0, 2π) and velocity v n ∈[0, v max The description is as follows. Where, v max This indicates the maximum speed of the power line inspection drone.

[0056] Assuming the power line inspection drone flies at a constant speed in each time slot, the flight distance d of the power line inspection drone n within one time slot (time slot t) is... n (t) can be represented as follows:

[0057] d n (t)=v n (t)t0

[0058] Where t0 represents the duration of the time slot. Due to the finite flight speed of the power line inspection drone n, the maximum flight distance of the power line inspection drone n within one time slot is d. max .

[0059] The coordinate transformation of the power line inspection drone n from time slot t to time slot (t+1) can be represented as follows:

[0060] x n (t+1)=x n (t)+d n (t)cosφ n (t)

[0061] y n (t+1)=y n (t)+d n (t)sinφ n (t)

[0062] Channel gain h between power inspection drone n and power equipment m n，m This can be given by the free space path loss model: h n，m =g0(d n，m ) -α .

[0063] Where g0 represents the gain at a reference distance of 1m; α represents the path loss exponent; d n，m This represents the distance between the power line inspection drone n and the power equipment m.

[0064] Signal-to-interference plus noise ratio (SINR) between the power inspection drone n and the power equipment m in time slot t. n，m (t) can be represented as:

[0065]

[0066] Among them, P m σ represents the transmitting power of electrical equipment m; 2 This represents the power of Gaussian white noise.

[0067] In time slot t, power equipment m transmits some edge computing tasks to power inspection drone n for processing. The task offloading transmission rate of the above-mentioned tasks from power equipment m to power inspection drone n is r. n，m (t) is

[0068] r n，m (t)=B log2(1+SINR n，m (t))

[0069] Where B represents the system bandwidth.

[0070] The edge computing task generated by power device m in time slot t can be represented as: in, This indicates the size of the task, i.e., the amount of work required. This represents the proportion of the task unloaded by power equipment m, i.e., the unloading rate. Therefore, the amount of task unloaded by power equipment m from the power inspection drone in time slot t can be expressed as: The workload of power equipment m processed locally can be expressed as:

[0071] Transmission delay of edge computing tasks with power equipment unloading in time slot t and m for

[0072] The computational processing latency of the edge computing task for unloading power equipment m in time slot t at the power inspection drone is:

[0073] Where ω represents the number of CPU cycles required to compute each bit of the task; f n，m (t) represents the computing resources allocated to the edge computing task of the power inspection drone n for unloading power equipment m.

[0074] The latency of edge computing tasks for power equipment m in time slot t to be processed locally is

[0075]

[0076] Among them, f m (t) represents the idle computing resources of power equipment m in time slot t.

[0077] The total latency of the edge computing task for time slot t power equipment m can be expressed as:

[0078] Finally, the average latency of the power line inspection drone n in processing the edge computing task unloading in time slot t can be expressed as:

[0079]

[0080] Alternatively, to simplify the model, it is assumed that the energy consumption of the power line inspection drone is only related to its flight speed, and the energy consumption of the power line inspection drone during flight time slot t is... It can be represented as

[0081]

[0082] Among them, M n This represents the mass of the power line inspection drone n.

[0083] Computational energy consumption of power line inspection drone n for all edge computing tasks unloaded in time slot t It can be represented as

[0084]

[0085] Where k is a CPU-related constant.

[0086] Figure 1 This is one of the flowcharts illustrating the edge computing task offloading method for power line inspection drones provided by this invention. For example... Figure 1 As shown, the execution subject of the power inspection drone edge computing task offloading method provided in this embodiment of the invention can be a power inspection drone edge computing task offloading device, which is used for power inspection drones. The method includes: step 101, step 102 and step 103.

[0087] Step 101: Obtain the location of the target time slot power inspection drone, the location of each power device, and the workload of the edge computing task for each power device.

[0088] Specifically, the edge computing task offloading method for power inspection drones provided in this embodiment of the invention can be used for any power inspection drone in the aforementioned power inspection drone edge computing task offloading system. The power inspection drone edge computing task offloading device, as the executing entity, can be the MEC server carried by the power inspection drone.

[0089] The target time slot can be any time slot. The following description uses time slot t as an example. It should be understood that the target time slot is not limited to time slot t.

[0090] The location [x′] of the power line inspection drone carrying the MEC server in time slot t can be obtained based on any positioning method. j (t),y′ j (t),z′ j (t)].

[0091] Alternatively, the location of the electrical equipment can be fixed, thus allowing the location of each electrical device to be obtained based on a pre-obtained location method.

[0092] Alternatively, the power equipment can also obtain its own location based on any positioning method, and then send its own location to the edge computing task offloading device of the power inspection drone, so that the edge computing task offloading device of the power inspection drone can obtain the location of each power equipment in the target time slot.

[0093] The embodiments of the present invention do not specifically limit the positioning method used as described above. For example, positioning methods based on any Global Navigation Satellite System (GNSS), or the time difference of arrival method, can be used.

[0094] Optionally, before the target time slot (time slot t), each power device can obtain the task volume of its own edge computing task in time slot t and send the task volume to each power inspection drone in the power inspection drone edge computing task offloading system, so that the power inspection drone edge computing task offloading device can obtain the task volume of each power device's edge computing task in the target time slot.

[0095] Step 102: Input the location of the target time slot power inspection drone, the location of each power device, and the workload of the edge computing task of each power device into the first model corresponding to the power inspection drone, and obtain the unloading instruction of the target time slot for each power device and the flight control parameters of the target time slot power inspection drone.

[0096] Step 103: Based on the unloading instruction, perform edge computing task unloading on the target power equipment in each power equipment in the target time slot, and control the flight of the power inspection drone based on the flight control parameters.

[0097] The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot; the unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on the generated sample data.

[0098] Specifically, embodiments of the present invention can employ a heterogeneous multi-agent reinforcement learning algorithm to solve for the optimal offloading strategy. This optimization problem includes the joint optimization of the flight trajectory, task offloading allocation, and task offloading rate of the power line inspection drone. The power line inspection drone and power equipment can be treated as two heterogeneous agents to collaboratively optimize the flight trajectory, task offloading allocation, and task offloading rate of the power line inspection drone.

[0099] Optionally, the flight control parameters of the power line inspection drone can be used to indicate the drone's flight trajectory. These flight control parameters may include flight distance and azimuth, or they may include flight speed and azimuth.

[0100] Optionally, the unloading instruction from the power inspection drone to the power equipment can be used to instruct the power inspection drone to unload the allocation of edge computing tasks to the power equipment.

[0101] The task offloading rate refers to the proportion of edge computing tasks of the target time slot power equipment that are offloaded to a certain power inspection drone.

[0102] This invention takes into account the large randomness of the real deployment environment caused by the computing resources of power equipment, the size (i.e., workload) of edge computing tasks, and the dynamic changes of edge computing tasks in the power drone inspection system. A new algorithm is adopted to ensure the performance of the computing offloading strategy in the real deployment environment.

[0103] This invention constructs a flight control (i.e., maneuvering) model for power inspection drones and an offloading model for edge computing tasks of power equipment, which are two heterogeneous intelligent agents that cooperate. The model is trained based on information such as real-time task requests, task attributes, the distribution of power inspection drones and power equipment, and computing power. A reinforcement learning algorithm is used to make decisions on the flight control of power inspection drones and the offloading of edge computing tasks of power equipment, so as to achieve a preset optimization goal, thereby obtaining a first model corresponding to each power inspection drone and a second model corresponding to each power equipment.

[0104] The generated sample data may include information such as real-time task requests, task attributes, the distribution of power line inspection drones and power equipment, and computing power. Optionally, the generated sample data may include the location of each power line inspection drone in historical time slots in the real deployment environment, the location of each power equipment, the workload of edge computing tasks for each power equipment, and the idle computing resources of each power equipment. Historical time slots may include multiple time slots preceding the target time slot.

[0105] Optionally, the reinforcement learning algorithm may be the MADDPG (Multi-Agent Deep Deterministic Policy Gradient) algorithm, or any improved version of the MADDPG algorithm. This embodiment of the invention does not limit the specific reinforcement learning algorithm used.

[0106] Optionally, the optimization objective can be determined based on at least one of the energy consumption of each power line inspection drone and the latency of each power line inspection drone in processing the unloading edge computing task.

[0107] It should be noted that during the training process, any intelligent agent (power inspection drone or power equipment) can be used as a central node. The central node trains the first model corresponding to each power inspection drone and the second model corresponding to each power equipment based on generated sample data, resulting in trained first and second models. Each first model is then distributed to the MEC server mounted on the corresponding power inspection drone, and each second model is distributed to the corresponding power equipment. After obtaining the first model distributed by the central node, the MEC server mounted on the corresponding power inspection drone can make independent decisions based on that first model, without requiring overall control from the central node. It can then obtain the unloading instruction for each power equipment and the flight control parameters for that power inspection drone in the target time slot, based on the location of the power inspection drone in the target time slot, the location of each power equipment, and the workload of the edge computing tasks of each power equipment.

[0108] After obtaining the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device, the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device can be input into the first model corresponding to the power inspection drone equipped with the edge computing task offloading device of the power inspection drone to obtain the offloading instruction of the power inspection drone for each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot.

[0109] Optionally, after determining the target time slot and the unloading instruction for each power device by the power inspection drone, the power device corresponding to the value of 1 in the unloading instruction can be identified as the target power device. The number of target power devices can be one or more.

[0110] After identifying the target power equipment, the power inspection drone can offload edge computing tasks for each target power equipment in the target time slot. For any target power equipment, the amount of edge computing tasks offloaded by the power inspection drone in the target time slot can be equal to the product of the amount of edge computing tasks on the target power equipment in the target time slot and the offloading rate of edge computing tasks on the target power equipment in the target time slot.

[0111] After obtaining the flight control parameters of the power line inspection drone in the target time slot, the flight of the power line inspection drone can be controlled based on the flight control parameters of the power line inspection drone in the target time slot.

[0112] This invention, through a flight control model for power inspection drones and an offloading model for edge computing tasks of power equipment based on the collaboration of two heterogeneous intelligent agents—power inspection drones and power equipment—makes decisions based on the workload of edge computing tasks of the power equipment, determines the offloading rate, and provides offloading instructions from the power inspection drone to the power equipment. This allows a portion of the tasks to be computed locally on the power equipment to fully utilize its own computing resources, while the remaining portion is offloaded to the power inspection drone based on the offloading rate. This approach more effectively meets the offloading needs of edge computing tasks of ground-based power equipment, improves service quality, and reduces task latency and energy consumption of the power inspection drone.

[0113] Based on any of the above embodiments, before inputting the location of the target time slot power inspection drone, the location of each power device, and the workload of the edge computing task of each power device into the first model corresponding to the power inspection drone, and obtaining the unloading instruction of the target time slot for each power device and the flight control parameters of the target time slot power inspection drone, the method further includes: performing offline training on the generated sample data based on the MATD3 algorithm and a preset optimization target.

[0114] Specifically, in this embodiment of the invention, to achieve the optimal offloading strategy for edge computing tasks of power equipment in the power drone inspection system, it involves making reasonable decisions regarding task offloading allocation, the proportion of task offloading, and the maneuvering of the power inspection drone. For both the power inspection drone and the power equipment, the common goal is to improve service quality and reduce task latency. Furthermore, the power inspection drone also needs to minimize its own energy consumption; therefore, the preset optimization objectives can include both task latency and the power inspection drone's energy consumption.

[0115] Power line inspection drones and equipment are considered as two heterogeneous intelligent agents that collaborate to optimize the drone's maneuvers, task allocation, and task unloading ratio. Specifically, the drone is responsible for making decisions regarding maneuvers and task allocation, while the equipment is responsible for determining the task unloading ratio.

[0116] Using the edge computing task offloading device of the power inspection drone as the central node, and based on the preset optimization objective, the generated sample data can be trained according to the MATD3 algorithm to obtain the first model corresponding to each power inspection drone and the second model corresponding to each power equipment.

[0117] The TD3 (Twin Delayed Deep Deterministic Policy Gradient) algorithm can address the problem of excessively large median function estimation in reinforcement learning. The MATD3 (Multi-Agent Twin Delayed Deep Deterministic Policy Gradient) algorithm is a multi-agent extension of TD3 and an improved version of the MADDPG algorithm, featuring centralized training and distributed execution. Each agent possesses a policy model that can derive corresponding action policy outputs based on its partial observation inputs. During the policy model training phase, a central node acquires all agents' observations, actions, rewards, and the next observation to form experience and adds it to an experience pool. Each time the network parameters are updated, a portion of the experience pool is selected for updating.

[0118] In the MATD3 algorithm, for ease of representation, let k = M + N, meaning there are k agents in the system. The joint observation of all agents can be represented as o = {o1, o2, ..., o}. k A batch of size M is randomly selected from the experience pool. b experience {o j ,o j ′ ,a j ,r j} contains the joint state, actions, and rewards of all agents. The network update for agent x is as follows, where the policy Actor network of agent x can be represented as: The two evaluation Critic networks can be represented as and

[0119] The parameters of the policy network are updated using the policy gradient method:

[0120]

[0121] The algorithm uses two Critic networks. The network with the smallest estimated Q-value is selected to obtain the target value y. j .

[0122]

[0123] Then, it can be based on strategy The two Critic networks can be updated by minimizing the loss function:

[0124]

[0125] To solve the joint optimization problem of the maneuvering of the power inspection drone and the offloading of edge computing tasks of the power equipment under the collaborative architecture of power inspection drone and power equipment, this embodiment of the invention designs observation, action strategies and rewards for two types of agents based on the MATD3 algorithm.

[0126] Intelligent agents can be categorized into two types: power line inspection drones and power equipment.

[0127] Observation: Each agent acquires partial observations from the environment based on its own needs.

[0128] Time-slot t power line inspection drone j intelligent agent observation This includes the location of time slot t itself, the location of each power device, and the workload of edge computing tasks for each power device.

[0129]

[0130] Where, x j ′ (t),y j ′ (t),z j ′ (t) represents the position of the power line inspection drone itself.

[0131] Observation of time-slot t power equipment i intelligent agent This includes the workload of edge computing tasks within time slot t itself, as well as its own idle computing resources.

[0132]

[0133] Action policy: After each agent acquires its own observations, it inputs them into the Actor network to obtain the action policy.

[0134] Action strategy of time-slot t power inspection drone j intelligent agent It can include time slot t for unloading instructions for the power equipment, as well as its own azimuth and flight distance.

[0135]

[0136] Action strategy of time slot t power equipment i intelligent agent This can include the offloading rate of the task itself in time slot t.

[0137]

[0138] Rewards: During training, the agent will receive a reward for performing the action strategy in the current state after interacting with the environment.

[0139] Rewards for the Time-Slotted Power Inspection Drone J Intelligent Agent It is related to the latency of the edge computing tasks on the power equipment being processed and its own energy consumption.

[0140]

[0141] in, and represents the weights of energy consumption of the power inspection drone j agent and latency of edge computing tasks of power equipment, respectively; p represents the penalty for violating constraints.

[0142] Rewards for the intelligent agent in time slot t power equipment i It is only related to the latency of its own edge computing tasks.

[0143]

[0144] in, The weight represents the latency of the edge computing task of power device i.

[0145] The reinforcement learning algorithm uses the MATD3 algorithm. Once the policy model is trained, each agent can make action decisions based on its own local observations and locally deployed policy model, without the need for a central node to coordinate and control.

[0146] This invention constructs a model for the maneuvering of a power inspection drone and the offloading of edge computing tasks from power equipment, two heterogeneous intelligent agents, through collaboration between the drone and the equipment. The model is trained based on information such as real-time task requests, task attributes, the distribution of the drone and the equipment, and computing power. The MATD3 reinforcement learning algorithm is used to make decisions on the flight control of the drone and the offloading of edge computing tasks from the equipment, thereby achieving the preset optimization objectives.

[0147] Based on the content of any of the above embodiments, according to the MATD3 algorithm, before offline training on the generated sample data based on a preset optimization objective, the method further includes: acquiring historical operating data of each power inspection drone and each power equipment in the real deployment environment.

[0148] Specifically, the historical operational data of each power line inspection drone and each power equipment can include the location of each power line inspection drone, the location of each power equipment, the workload of edge computing tasks for each power equipment, and the idle computing resources of each power equipment in historical time slots. Historical time slots can include multiple time slots preceding the target time slot.

[0149] Based on historical operational data from real deployment environments, sample data is generated using generative adversarial network algorithms.

[0150] Specifically, the generated sample data can include not only historical running data, but also the first data generated based on historical running data using the Generative Adversarial Network (GAN) algorithm.

[0151] GANs contain two sub-networks: a generator (G) and a discriminator (D). The generator aims to produce "fake" samples that can fool the discriminator, which is essentially a binary classifier designed to distinguish between generated "fake" samples and real samples. Through the game between the generator and the discriminator, the generator eventually produces samples that approximate the distribution of real samples. Therefore, this embodiment of the invention employs the MATD3-assisted training algorithm based on GANs, using GANs to fit finite environmental states sampled from real deployment environments. The goal is to obtain a generator that can produce approximate real environmental states for offline training of the agent, thereby addressing the problems of high online training costs and the inability of traditional offline training to guarantee the effectiveness of agent strategies in real-world deployments.

[0152] The objective function of GAN can be expressed as follows:

[0153]

[0154] Here, V(G,D) represents the degree of difference between the real sample and the generated sample. The purpose of G is to minimize V(G,D), that is, to minimize the degree of difference between the real sample and the generated sample, while the purpose of D is to maximize the degree of difference between the real sample and the generated sample.

[0155] Discriminator D updates its parameters via gradient ascent.

[0156]

[0157] The generator G updates its parameters via gradient descent.

[0158]

[0159] Through a minimax game between the generator and the discriminator, a Nash equilibrium will eventually be reached, and the generator can generate samples that are similar to the distribution of the real environment.

[0160] This invention uses GAN to simulate the distribution of equipment environment state attributes in a power inspection system. By using the generated network environment state during offline training, the agent's experience is enhanced, enabling the agent to make action strategy decisions in real time and efficiently based on partial observations in an environment where resource and service requests are dynamically changing.

[0161] Using GANs, environmental state samples obtained from real deployment environments can be fitted to learn the attribute distribution of the real environment. This yields a generator that can approximate the task requests and network states in a power drone inspection system. During offline reinforcement learning training, the GAN generator is used as the environmental state generator to generate state attributes that approximate the real deployment environment. This enhances the agent's ability to cope with real-world environments, enabling the agent's strategies trained offline to be efficiently applied to real power drone inspection scenarios. Compared to online training in a real environment, this approach offers higher feasibility and significantly reduces training costs. It also allows for more effective strategies to ensure optimal flight control of power drones and offloading of edge computing tasks from power equipment within the power drone inspection system, achieving the preset optimization goals.

[0162] The embodiments of the present invention employ an offline training assistance method for reinforcement learning agents based on generative adversarial networks, which can reduce the cost of sampling from real environments and ensure that the agents have good performance in real deployment environments.

[0163] This invention utilizes GAN to learn the distribution of real-world environmental states, i.e., historical operational data. During training, the agent is trained based on the generated environmental states, which ensures the effectiveness of the agent's strategy in the real deployment environment. Compared with online training in the real environment, this significantly reduces training costs.

[0164] Based on the content of any of the above embodiments, the optimization objective includes minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task for processing unloading.

[0165] Specifically, in order to optimize the maneuvering, unloading task allocation, and task unloading ratio of power line inspection drones, the optimization objective can be to minimize the total energy consumption of each power line inspection drone in a time slot and the average latency of the edge computing task for processing unloading.

[0166] Alternatively, the optimization objective can be expressed by the following formula:

[0167]

[0168] C1:

[0169] C2:

[0170] C3:

[0171] C4:

[0172] C5:

[0173] C6:

[0174] Among them, constraints C1-C3 are used to constrain the range of variable values; constraint C4 represents the minimum distance between power inspection drones to avoid collisions; constraint C5 represents that an edge computing task of power equipment can only be offloaded and processed by one power inspection drone; constraint C6 represents that the sum of computing resources allocated to each task by the MEC server carried by the power inspection drone cannot exceed the maximum computing resources of the MEC server itself.

[0175] Where, α e and α j These represent the weights of the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task being processed and unloaded, respectively.

[0176] Alternatively, the optimization objective can be to minimize the sum of the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task that handles the unloading.

[0177] To facilitate understanding of the above embodiments of the present invention, the acquisition process of each first model and each second model is described below.

[0178] Figure 2 This is a schematic diagram illustrating the acquisition process of each first model and each second model in the edge computing task offloading method for power line inspection drones provided by this invention. For example... Figure 2 As shown, the process of obtaining each first model and each second model may include the following steps:

[0179] Step 201: Establish a collaborative architecture between power inspection drones and power equipment.

[0180] By treating power inspection drones and power equipment as two heterogeneous intelligent agents, the drones and power equipment collaborate to optimize their flight trajectories, task offloading allocation, and task offloading rate, thus forming a collaborative architecture between power inspection drones and power equipment.

[0181] Step 202: Construct optimization objectives based on the inspection needs of power line inspection drones.

[0182] The aforementioned preset optimization objectives can be constructed based on the inspection needs of power line inspection drones.

[0183] Step 203: Solve for the optimal flight control strategy for the power line inspection drone and the offloading strategy for edge computing tasks of power equipment.

[0184] The optimal flight control strategy for power line inspection drones and the offloading strategy for edge computing tasks of power equipment are solved based on the GAN-MATD3 algorithm, thereby obtaining the first model and the second model.

[0185] The embodiments of the present invention aim to minimize the total energy consumption of each power inspection drone in a time slot and the average latency of the offloading edge computing tasks. This can effectively guarantee the optimal flight control strategy for power inspection drones and the offloading strategy for edge computing tasks of power equipment, and achieve the optimization goal of minimizing the energy consumption of power inspection drones and the latency of edge computing tasks of power equipment.

[0186] Figure 3 This is the second flowchart illustrating the edge computing task offloading method for power line inspection drones provided by this invention. Figure 3 As shown, the execution subject of the power inspection drone edge computing task offloading method provided in this embodiment of the invention can be a power inspection drone edge computing task offloading device, which is used for power equipment. The method includes: step 301 and step 302.

[0187] Step 301: Obtain the task load and available computing resources of the edge computing task for the target time slot power equipment.

[0188] Specifically, the edge computing task offloading method for power line inspection drones provided in this embodiment of the invention can be used for any power device in the aforementioned power line inspection drone edge computing task offloading system. The following description uses power device m as an example. It is understood that the power device is not limited to power device m.

[0189] The target time slot can be any time slot. The following description uses time slot t as an example. It should be understood that the target time slot is not limited to time slot t.

[0190] It is understandable that before the target time slot (time slot t), the power equipment can obtain the workload of the edge computing tasks of time slot t itself and the idle computing resources of time slot t itself.

[0191] Step 302: Input the task load and idle computing resources of the edge computing task of the target time slot power equipment into the second model corresponding to the power equipment, and obtain the offloading rate of the edge computing task of the target time slot power equipment, so that the target power inspection drone can offload the edge computing task of the target time slot power equipment based on the offloading rate in the target time slot.

[0192] Among them, the unloading instruction for the target power inspection drone in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot; the unloading instruction is determined based on the first model corresponding to the target power inspection drone, as well as the position of the target power inspection drone in the target time slot, the position of each power equipment, and the workload of the edge computing tasks of each power equipment; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on the generated sample data.

[0193] It should be noted that after obtaining the second model issued by the central node, the power equipment corresponding to the second model can make independent decisions based on the second model without the central node's overall control. The offloading rate of the edge computing tasks of the power equipment in the target time slot can be obtained based on the task volume and idle computing resources of the edge computing tasks of the power equipment in the target time slot.

[0194] Optionally, the workload and idle computing resources of the edge computing tasks of the target time slot power equipment are input into the second model corresponding to the power equipment to obtain the offloading rate of the edge computing tasks of the target time slot power equipment. Before the target power inspection drone offloads the edge computing tasks of the target time slot power equipment based on the offloading rate in the target time slot, the process may further include: offline training of the generated sample data according to the MATD3 algorithm based on a preset optimization objective. That is, using the edge computing task offloading device of the power inspection drone as the central node, the generated sample data can be trained according to the MATD3 algorithm based on a preset optimization objective to obtain the first model corresponding to each power inspection drone and the second model corresponding to each power equipment.

[0195] Optionally, according to the MATD3 algorithm, before offline training on the generated sample data based on a preset optimization objective, the process may also include: obtaining historical operating data of each power inspection drone and each power equipment in a real deployment environment; and obtaining generated sample data based on the historical operating data and a generative adversarial network algorithm.

[0196] Optionally, the optimization objective may include minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task for processing offloading.

[0197] This invention, through a flight control model for power inspection drones and an offloading model for edge computing tasks of power equipment based on the collaboration of two heterogeneous intelligent agents—power inspection drones and power equipment—makes decisions based on the workload of edge computing tasks of the power equipment, determines the offloading rate, and provides offloading instructions from the power inspection drone to the power equipment. This allows a portion of the tasks to be computed locally on the power equipment to fully utilize its own computing resources, while the remaining portion is offloaded to the power inspection drone based on the offloading rate. This approach more effectively meets the offloading needs of edge computing tasks of ground-based power equipment, improves service quality, and reduces task latency and energy consumption of the power inspection drone.

[0198] The edge computing task offloading device for power inspection drones provided by the present invention is described below. The edge computing task offloading device for power inspection drones described below can be referred to in correspondence with the edge computing task offloading method for power inspection drones described above.

[0199] Figure 4 This is one of the structural schematic diagrams of the edge computing task offloading device for power line inspection drones provided by the present invention. Based on the content of any of the above embodiments, as... Figure 4 As shown, the edge computing task offloading device for power line inspection drones includes: a first acquisition module 401, a second acquisition module 402, and a task offloading module 403, wherein:

[0200] The first acquisition module 401 is used to acquire the location of the target time slot power inspection drone, the location of each power device, and the workload of the edge computing task of each power device.

[0201] The second acquisition module 402 is used for the location and edge computing task of each power equipment. It inputs the first model corresponding to the power inspection drone and acquires the unloading instruction of each power equipment in the target time slot and the flight control parameters of the power inspection drone in the target time slot.

[0202] The task unloading module 403 is used to perform edge computing task unloading on the target power equipment in each power equipment in the target time slot based on the unloading instruction, and to control the flight of the power inspection drone based on the flight control parameters.

[0203] The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot; the unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on the generated sample data.

[0204] Specifically, the first acquisition module 401, the second acquisition module 402, and the task unloading module 403 can be electrically connected in sequence.

[0205] Optionally, the edge computing task offloading device for the power line inspection drone may also include:

[0206] The first training module is used to perform offline training on the generated sample data based on the MATD3 algorithm and a preset optimization objective.

[0207] Optionally, the edge computing task offloading device for the power inspection drone may further include: a first generation module, used to acquire historical operating data of each power inspection drone and each power equipment in the real deployment environment; and based on the historical operating data, to acquire generated sample data according to the generative adversarial network algorithm.

[0208] Optionally, the optimization objective may include minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task for processing offloading.

[0209] The edge computing task offloading device for power inspection drones provided in this embodiment of the invention is used to execute the edge computing task offloading method for power inspection drones described above. Its implementation method is consistent with the implementation method of the edge computing task offloading method for power inspection drones provided by this invention, and can achieve the same beneficial effects, so it will not be described again here.

[0210] This power line inspection drone edge computing task offloading device is used in the power line inspection drone edge computing task offloading methods of the foregoing embodiments. Therefore, the descriptions and definitions in the power line inspection drone edge computing task offloading methods of the foregoing embodiments can be used to understand the execution modules in the embodiments of the present invention.

[0211] This invention, through a flight control model for power inspection drones and an offloading model for edge computing tasks of power equipment based on the collaboration of two heterogeneous intelligent agents—power inspection drones and power equipment—makes decisions based on the workload of edge computing tasks of the power equipment, determines the offloading rate, and provides offloading instructions from the power inspection drone to the power equipment. This allows a portion of the tasks to be computed locally on the power equipment to fully utilize its own computing resources, while the remaining portion is offloaded to the power inspection drone based on the offloading rate. This approach more effectively meets the offloading needs of edge computing tasks of ground-based power equipment, improves service quality, and reduces task latency and energy consumption of the power inspection drone.

[0212] Figure 5 This is a second structural schematic diagram of the edge computing task offloading device for power line inspection drones provided by the present invention. Based on the content of any of the above embodiments, as... Figure 5As shown, the edge computing task offloading device for power line inspection drones includes: a third acquisition module 501 and a fourth acquisition module 502, wherein:

[0213] The third acquisition module 501 is used to acquire the task volume and idle computing resources of the edge computing task of the target time slot power equipment;

[0214] The fourth acquisition module 502 is used to input the task volume and idle computing resources of the edge computing task of the target time slot power equipment into the second model corresponding to the power equipment, and obtain the offloading rate of the edge computing task of the target time slot power equipment, so that the target power inspection drone can offload the edge computing task of the target time slot power equipment based on the offloading rate in the target time slot.

[0215] Among them, the unloading instruction for the target power inspection drone in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot; the unloading instruction is determined based on the first model corresponding to the target power inspection drone, as well as the position of the target power inspection drone in the target time slot, the position of each power equipment, and the workload of the edge computing tasks of each power equipment; the first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on the generated sample data.

[0216] Specifically, the third acquisition module 501 and the fourth acquisition module 502 can be electrically connected.

[0217] Optionally, the edge computing task offloading device for the power line inspection drone may also include:

[0218] The second training module is used to perform offline training on the generated sample data based on the MATD3 algorithm and a preset optimization objective.

[0219] Optionally, the edge computing task offloading device for the power inspection drone may further include: a second generation module, used to acquire historical operating data of each power inspection drone and each power equipment in the real deployment environment; and based on the historical operating data, to acquire generated sample data according to the generative adversarial network algorithm.

[0220] Optionally, the optimization objective may include minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task for processing offloading.

[0221] The edge computing task offloading device for power inspection drones provided in this embodiment of the invention is used to execute the edge computing task offloading method for power inspection drones described above. Its implementation method is consistent with the implementation method of the edge computing task offloading method for power inspection drones provided by this invention, and can achieve the same beneficial effects, so it will not be described again here.

[0222] This power line inspection drone edge computing task offloading device is used in the power line inspection drone edge computing task offloading methods of the foregoing embodiments. Therefore, the descriptions and definitions in the power line inspection drone edge computing task offloading methods of the foregoing embodiments can be used to understand the execution modules in the embodiments of the present invention.

[0223] This invention, through a flight control model for power inspection drones and an offloading model for edge computing tasks of power equipment based on the collaboration of two heterogeneous intelligent agents—power inspection drones and power equipment—makes decisions based on the workload of edge computing tasks of the power equipment, determines the offloading rate, and provides offloading instructions from the power inspection drone to the power equipment. This allows a portion of the tasks to be computed locally on the power equipment to fully utilize its own computing resources, while the remaining portion is offloaded to the power inspection drone based on the offloading rate. This approach more effectively meets the offloading needs of edge computing tasks of ground-based power equipment, improves service quality, and reduces task latency and energy consumption of the power inspection drone.

[0224] Based on any of the above embodiments, a power inspection drone edge computing task offloading system includes: multiple first power inspection drone edge computing task offloading devices and multiple second power inspection drone edge computing task offloading devices.

[0225] Specifically, the first power line inspection drone edge computing task offloading device can be the MEC server carried by the power line inspection drone. The second power line inspection drone edge computing task offloading device can be power equipment.

[0226] The specific process of multiple first power inspection drone edge computing task offloading devices and multiple second power inspection drone edge computing task offloading devices executing the power inspection drone edge computing task offloading method can be found in the aforementioned embodiments, and will not be repeated here.

[0227] This invention, through a flight control model for power inspection drones and an offloading model for edge computing tasks of power equipment based on the collaboration of two heterogeneous intelligent agents—power inspection drones and power equipment—makes decisions based on the workload of edge computing tasks of the power equipment, determines the offloading rate and the offloading instruction from the power inspection drone to the power equipment, leaves a portion of the tasks to be computed locally on the power equipment to make full use of the power equipment's own computing resources, and offloads another portion to the power inspection drone according to the offloading rate. This can more effectively meet the offloading needs of edge computing tasks of ground-based power equipment, improve service quality and reduce task latency.

[0228] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a method for offloading edge computing tasks from a power inspection drone. This method includes: obtaining the location of the power inspection drone in the target time slot, the location of each power device, and the workload of edge computing tasks for each power device; inputting the location of the power inspection drone in the target time slot, the location of each power device, and the workload of edge computing tasks for each power device into a first model corresponding to the power inspection drone; obtaining an offloading instruction for each power device in the target time slot and flight control parameters of the power inspection drone in the target time slot; based on the offloading instruction, offloading edge computing tasks for the target power devices in the target time slot, and controlling the flight of the power inspection drone based on the flight control parameters; wherein, the offloading workload is determined based on the workload and offloading rate of the edge computing tasks of the target power devices in the target time slot; the offloading rate of the edge computing tasks of the target power devices in the target time slot is determined based on a second model corresponding to the target power devices, as well as the workload and available computing resources of the edge computing tasks of the target power devices in the target time slot; the first model corresponding to each power inspection drone and the second model corresponding to each power device are obtained after training based on generated sample data.

[0229] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0230] The processor 610 in the electronic device provided in this embodiment of the invention can call the logical instructions in the memory 630. Its implementation method is consistent with the implementation method of the power inspection drone edge computing task offloading method provided in this invention, and can achieve the same beneficial effect. It will not be described again here.

[0231] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the power inspection drone edge computing task offloading method provided by the above methods, the method comprising: obtaining the location of the power inspection drone in the target time slot, the location of each power device, and the task load of the edge computing task of each power device; inputting the location of the power inspection drone in the target time slot, the location of each power device, and the task load of the edge computing task of each power device into a first model corresponding to the power inspection drone, and obtaining the offloading of the target time slot for each power device. The system specifies the flight control parameters for the power inspection drone in the target time slot. Based on the unloading instruction, edge computing tasks are unloaded from the target power equipment in each target time slot, and the flight of the power inspection drone is controlled based on the flight control parameters. The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot. The unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0232] When the computer program product provided in this embodiment of the invention is executed, it implements the above-mentioned method for offloading edge computing tasks of power inspection drones. Its specific implementation method is consistent with the implementation method described in the aforementioned method embodiment, and can achieve the same beneficial effects, which will not be repeated here.

[0233] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the aforementioned edge computing task offloading methods for power inspection drones. The method includes: obtaining the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task for each power device; inputting the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task for each power device into a first model corresponding to the power inspection drone; and obtaining the offloading instruction for each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot. The system performs edge computing task offloading on target power equipment in each power equipment slot based on offloading instructions, and controls the flight of the power inspection drone based on flight control parameters. The offloading task volume is determined based on the task volume and offloading rate of the edge computing task of the target power equipment in the target time slot. The offloading rate of the edge computing task of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the task volume and idle computing resources of the edge computing task of the target power equipment in the target time slot. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

[0234] When the computer program stored on the non-transitory computer-readable storage medium provided in this embodiment of the invention is executed, it implements the above-mentioned power inspection drone edge computing task offloading method. Its specific implementation method is consistent with the implementation method described in the aforementioned method embodiment and can achieve the same beneficial effect, so it will not be repeated here.

[0235] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the embodiments of the present invention according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0236] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0237] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for offloading edge computing tasks from a power line inspection drone, characterized in that, The method includes: The location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task for each power device are obtained. The location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device are input into the first model corresponding to the power inspection drone to obtain the unloading instruction for each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot. Based on the unloading instruction, the edge computing task is unloaded for the target power equipment in each of the power equipment in the target time slot, and the flight of the power inspection drone is controlled based on the flight control parameters. The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot. The unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data. The optimal flight control strategy for the power inspection drone and the unloading strategy for the edge computing tasks of the power equipment are solved based on the GAN-MATD3 algorithm to obtain each first model and each second model. The positions of the power inspection drones and each power equipment in the target time slot are... Before inputting the edge computing task workload of each of the power equipment into the first model corresponding to the power inspection drone, and obtaining the unloading instruction for each power equipment in the target time slot and the flight control parameters of the power inspection drone in the target time slot, the method further includes: offline training of the generated sample data based on the MATD3 algorithm and a preset optimization objective; once the policy model is trained, each agent makes action decisions based on its local observations and the locally deployed policy model; before offline training of the generated sample data based on the MATD3 algorithm and a preset optimization objective, the method further includes: obtaining historical operating data of each power inspection drone and each power equipment in the real deployment environment; and obtaining the generated sample data based on the historical operating data and a generative adversarial network algorithm.

2. The method for offloading edge computing tasks from a power line inspection drone according to claim 1, characterized in that, The optimization objectives include minimizing the total energy consumption of each power inspection drone in a time slot and the average latency of the edge computing task for processing unloading.

3. The method for offloading edge computing tasks from a power line inspection drone according to any one of claims 1-2, characterized in that, The method further includes: Obtain the workload and idle computing resources of the edge computing tasks of the power equipment in the target time slot; The task volume and idle computing resources of the edge computing task of the power equipment in the target time slot are input into the second model corresponding to the power equipment to obtain the offloading rate of the edge computing task of the power equipment in the target time slot, so that the target power inspection drone can offload the edge computing task of the power equipment in the target time slot based on the offloading rate. The unloading instruction from the target power inspection drone to the power equipment in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot. The unloading instruction is determined based on the first model corresponding to the target power inspection drone, the location of the target power inspection drone in the target time slot, the location of each power equipment, and the workload of the edge computing tasks of each power equipment. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

4. A power line inspection drone edge computing task offloading device, characterized in that, include: The first acquisition module is used to acquire the location of the power inspection drone in the target time slot, the location of each power device, and the workload of the edge computing task of each power device. The second acquisition module is used to input the location and the task volume of edge computing of each power device into the first model corresponding to the power inspection drone, and to acquire the unloading instruction of each power device in the target time slot and the flight control parameters of the power inspection drone in the target time slot. The task offloading module is used to perform edge computing task offloading on the target power equipment in each of the power equipment in the target time slot based on the offloading instruction, and to control the flight of the power inspection drone based on the flight control parameters. The unloading workload is determined based on the workload and unloading rate of the edge computing tasks of the target power equipment in the target time slot. The unloading rate of the edge computing tasks of the target power equipment in the target time slot is determined based on the second model corresponding to the target power equipment, as well as the workload and idle computing resources of the edge computing tasks of the target power equipment in the target time slot. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data. The optimal flight control strategy for the power inspection drone and the unloading strategy for the edge computing tasks of the power equipment are solved based on the GAN-MATD3 algorithm to obtain each first model and each second model. The positions of the power inspection drones and each power equipment in the target time slot are... Before inputting the edge computing task workload of each of the power equipment into the first model corresponding to the power inspection drone, and obtaining the unloading instruction for each power equipment in the target time slot and the flight control parameters of the power inspection drone in the target time slot, the method further includes: offline training of the generated sample data based on the MATD3 algorithm and a preset optimization objective; once the policy model is trained, each agent makes action decisions based on its local observations and the locally deployed policy model; before offline training of the generated sample data based on the MATD3 algorithm and a preset optimization objective, the method further includes: obtaining historical operating data of each power inspection drone and each power equipment in the real deployment environment; and obtaining the generated sample data based on the historical operating data and a generative adversarial network algorithm.

5. The edge computing task offloading device for power line inspection UAVs according to claim 4, characterized in that, Also includes: The third acquisition module is used to acquire the task volume and idle computing resources of the edge computing tasks of the target time slot power equipment; The fourth acquisition module is used to input the task volume and idle computing resources of the edge computing task of the power equipment in the target time slot into the second model corresponding to the power equipment, and obtain the offloading rate of the edge computing task of the power equipment in the target time slot, so that the target power inspection drone can offload the edge computing task of the power equipment in the target time slot based on the offloading rate. The unloading instruction from the target power inspection drone to the power equipment in the target time slot is used to instruct the target power inspection drone to unload the edge computing tasks of the power equipment in the target time slot. The unloading instruction is determined based on the first model corresponding to the target power inspection drone, the location of the target power inspection drone in the target time slot, the location of each power equipment, and the workload of the edge computing tasks of each power equipment. The first model corresponding to each power inspection drone and the second model corresponding to each power equipment are obtained after training based on generated sample data.

6. A power line inspection drone edge computing task offloading system, characterized in that, include: The edge computing task offloading device for power inspection drones as described in any one of claims 4 to 5.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the edge computing task offloading method for power inspection drones as described in any one of claims 1 to 3.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the edge computing task offloading method for power inspection drones as described in any one of claims 1 to 3.

Images