WO2023160012A1 - 一种用于电网线路随机巡检的无人机辅助边缘计算方法 - Google Patents

一种用于电网线路随机巡检的无人机辅助边缘计算方法 Download PDF

Info

Publication number
WO2023160012A1
WO2023160012A1 PCT/CN2022/130532 CN2022130532W WO2023160012A1 WO 2023160012 A1 WO2023160012 A1 WO 2023160012A1 CN 2022130532 W CN2022130532 W CN 2022130532W WO 2023160012 A1 WO2023160012 A1 WO 2023160012A1
Authority
WO
WIPO (PCT)
Prior art keywords
time slot
uav
inspection
enhanced
drone
Prior art date
Application number
PCT/CN2022/130532
Other languages
English (en)
French (fr)
Inventor
谈玲
孙雷
夏景明
Original Assignee
南京信息工程大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京信息工程大学 filed Critical 南京信息工程大学
Publication of WO2023160012A1 publication Critical patent/WO2023160012A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/445—Program loading or initiating
    • G06F9/44594—Unloading
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/502—Proximity
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present application relates to the technical field of mobile edge computing, in particular to a drone-assisted edge computing method for random inspection of power grid lines.
  • the current general method is to use an inspection robot suspended on the transmission line for line inspection, but its moving speed is slow, resulting in a long inspection period and low efficiency.
  • the UAV-assisted random inspection method can realize fast and efficient inspection of power grid lines, which is time-saving and economical.
  • This application uses a digital twin network to build a UAV-assisted random inspection system for power grid lines.
  • This system introduces non-orthogonal multiple access (NOMA) into the power grid line inspection scene for the first time, and solves the problem of mobile UAV clusters in power grid line inspection.
  • NOMA non-orthogonal multiple access
  • the so-called near-far effect means that when the UAV is moving, when the UAV is enhanced to receive signals from two inspection UAVs at different distances at the same time, because the inspection UAV with a closer distance has a stronger signal and the distance is closer.
  • the signal of the distant inspection drone is weak, and the strong signal of the former will cause serious interference to the latter, and the introduction of NOMA can overcome the above interference.
  • the technical problem to be solved in this application is to provide a UAV-assisted edge computing method for random inspection of power grid lines in view of the full coverage of lines in power grid line inspections and the far-near effect of mobile drone clusters.
  • the new model under the condition of completing the power grid line inspection task, realizes the minimization of the energy consumption balance of the UAV, thereby prolonging the working time of the UAV.
  • This application designs a UAV-assisted edge computing method for random inspection of power grid lines, based on a central base station set at a fixed location, the application
  • An unmanned aerial vehicle group including M inspection unmanned aerial vehicles and an enhanced unmanned aerial vehicle conduct inspections on target grid areas including grid facilities and transmission lines; including the following steps:
  • Step S1 Based on the flight mode of each of the inspection drones in the drone group, a drone-assisted random inspection system for power grid lines is constructed, wherein the inspection drones are only responsible for the inspection of the grid in the target grid area.
  • the power grid facilities and transmission lines are used to collect video images, and the obtained video images are processed by the enhanced drone or the central base station, and then enter step S2;
  • Step S2 Based on the random inspection system model of UAV-assisted power grid lines, each inspection UAV in the UAV group performs video image acquisition on the power grid facilities and transmission lines in the target power grid area, and obtains The video image data obtained by the inspection drone corresponding to each time slot collection respectively, and then enter step S3;
  • Step S3 According to the video image data collected by each of the inspection drones corresponding to each time slot, combined with the quality, signal transmission power, and position coordinates of each of the inspection drones, the enhanced wireless The quality of the man-machine, signal transmission power, position coordinates, computing power, the position coordinates of the central base station, and the system communication bandwidth, construct a digital twin network of the UAV-assisted power grid line random inspection system, and use it to fit the The position coordinates of the inspection drone, the enhanced drone, and the resource status of the system, and then enter step S4;
  • Step S4 According to the digital twin network of the UAV-assisted power grid line random inspection system, based on the grid line random inspection system unloading delay and data task processing delay constraints, construct the UAV group corresponding to each The energy consumption model or energy balance model of the time slot, and further construct the energy consumption minimization objective function corresponding to each time slot of the unmanned aerial vehicle group or the energy consumption balance minimization objective function corresponding to each time slot respectively, and then enter step S5;
  • Step S5. Randomly initialize the position coordinates of the enhanced UAV, and based on the position coordinates and video image data of each of the inspection UAVs corresponding to the tth time slot respectively, construct the state of the tth time slot of the system, and then Go to step S6;
  • Step S6 Based on the position coordinates of the enhanced drone and the state of the tth time slot of the system, the DDPG algorithm in deep reinforcement learning is used to minimize the objective function or Minimize the objective function of energy consumption balance, solve the energy consumption model of the UAV group corresponding to each time slot, obtain the state of the tth time slot of the system combined with the position coordinates of the enhanced UAV, and the unmanned inspection
  • the signal transmission power of the tth time slot and the allocated CPU calculation frequency form the action space of the tth time slot of the system, that is, the system corresponding to the state of the tth time slot combined with the position coordinates of the enhanced UAV
  • the tth time slot action space then enter step S7;
  • Step S7 Determine whether the iterative overflow condition is satisfied, and if so, enter step S8, otherwise based on the state of the tth time slot of the system combined with the system resource allocation in the tth time slot action space of the system corresponding to the position coordinates of the enhanced drone, As well as the video image data unloading decision-making scheme, using a genetic algorithm to solve and update the position coordinates of the enhanced drone, and return to step S6;
  • Step S8 According to the position coordinates of the enhanced UAV and the system resource allocation in the action space of the tth time slot of the corresponding system, and the unloading decision-making scheme of the video image data, for each of the patrol inspections in step S2 The man-machine processes the video image data collected corresponding to each time slot, so as to unload the video image data to the enhanced UAV or the central base station for processing.
  • An unmanned aerial vehicle-assisted edge computing method for random inspection of power grid lines described in this application compared with the prior art by adopting the above technical scheme, has the following technical effects:
  • This application designs a UAV-assisted edge computing method for random inspection of power grid lines.
  • the inspection UAV is used to collect video images of the target power grid area, and the inspection UAV is processed with the aid of enhanced UAV.
  • the method of combining the DDPG algorithm and the genetic algorithm in deep reinforcement learning is used to solve the position coordinates, system resource allocation, and task offloading decision of the enhanced UAV.
  • the scheme ensures that the system drones implement power grid inspections on the premise of minimizing energy consumption; among them, considering the harsh environment of the grid inspection area, the design uses drones to collect video images of the target grid area, and uses random
  • the way of inspection reduces the cost of inspection; considering the near-far effect caused by the communication between UAVs moving at high speed, this application introduces NOMA to overcome this shortcoming; and with the goal of optimizing the energy consumption of UAVs in the system,
  • the working time of the UAV is extended under the condition of performance; and the method of combining the
  • Fig. 1 is the implementation flow chart of the UAV-assisted power grid line random inspection method designed in an embodiment of the present application to integrate mobile edge computing;
  • Fig. 2 is a model diagram of the random inspection system of the unmanned aerial vehicle auxiliary grid circuit in the design application embodiment in an embodiment of the present application;
  • Fig. 3 is a schematic diagram of a digital twin network structure of an unmanned aerial vehicle-assisted power grid circuit random inspection in an embodiment of the design application;
  • FIG. 4 is a schematic diagram of DDPG for solving system resource allocation and task offloading decision-making schemes in an embodiment of the design application in an embodiment of the present application;
  • Fig. 5 is a performance diagram of the average system balanced energy consumption corresponding to different algorithm schemes in the design application embodiment in an embodiment of the present application;
  • Fig. 6 is a relationship diagram between the number of inspection drones and the balanced energy consumption of the system corresponding to different algorithm schemes in the design application embodiment in an embodiment of the present application;
  • Fig. 7 is a comparison of the system equilibrium energy consumption of the relative D value under different schemes corresponding to the design application embodiment in an embodiment of the present application.
  • this application proposes a UAV-assisted edge computing method suitable for power grid line inspection.
  • the goal of minimizing the energy consumption of the UAV group is achieved by jointly optimizing computing resources, communication resources, UAV trajectory, and task offloading decisions.
  • this application adopts the genetic algorithm and enhanced A learning-associated algorithm (GA-DDPG) solves the optimization problem for the above objectives.
  • GA-DDPG enhanced A learning-associated algorithm
  • reinforcement learning can quickly give an action strategy, which is suitable for solving problems with time-varying characteristics.
  • the agent in GA-DDPG reinforcement learning needs to obtain comprehensive and accurate system state information.
  • This application embeds the digital twin into the GA-DDPG algorithm, constructs the mapping between physical objects and virtual models, and then achieves the above goals.
  • GA- The genetic algorithm in DDPG is used to reduce the dimension of the decision space in the reinforcement learning algorithm and speed up the overall training speed of the algorithm.
  • Example embodiments will now be described more fully with reference to the accompanying drawings.
  • Example embodiments may, however, be embodied in many forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art.
  • the same reference numerals denote the same or similar parts in the drawings, and thus their repeated descriptions will be omitted.
  • This application designs a UAV-assisted edge computing method for random inspection of power grid lines.
  • an unmanned aerial vehicle UAV
  • UAV unmanned aerial vehicle
  • UAV inspection unmanned aerial vehicles
  • the human-machine group and an enhanced unmanned aerial vehicle SUAV
  • Step S1 Based on the flight mode of each inspection UAV in the UAV group, build a UAV-assisted random inspection system for power grid lines, where the inspection UAV is only responsible for the grid facilities and transmission lines in the target grid area
  • the video image is collected, and the obtained video image is processed by the enhanced drone or the central base station, and then enters step S2.
  • step S1 specifically executes the following steps S11 to S13.
  • Step S11 Based on the motion state of each inspection drone in each time slot remains unchanged, respectively for each inspection drone, according to the following formula:
  • the influence parameter of the vertical movement direction of the t time slot, the preset parameter ⁇ m that obeys the independent Gaussian distribution represents the randomness of the movement speed of the mth inspection drone, and the preset parameter ⁇ m that obeys the independent Gaussian distribution represents the mth
  • the randomness of the horizontal movement direction of an inspection drone, the preset parameters obey the independent Gaussian distribution Indicates the randomness of the vertical movement direction of the mth inspection drone, and then enters step S12.
  • Step S12 According to the length ⁇ of each time slot, for each inspection drone respectively, according to the following formula:
  • x m (t) x m (t-1)+v m (t-1)cos( ⁇ m (t-1)) ⁇
  • y m (t) y m (t-1)+v m (t-1)sin( ⁇ m (t-1)) ⁇
  • h m (t) h m (t-1)+v m (t-1)sin( ⁇ m (t-1)) ⁇
  • x m (t), y m (t), and h m (t) represent the values of the mth inspection drone corresponding to the tth time slot on the x, y, and z coordinate axes
  • x m ( t-1), y m (t-1), h m (t-1) represent the values of the mth inspection UAV corresponding to the first t-1 time slots on the x, y, and z coordinate axes respectively, Then go to step S13.
  • Step S13 According to the moving speed, horizontal moving direction, vertical moving direction, and position coordinates of each inspection UAV corresponding to the tth time slot, a UAV-assisted random inspection system for power grid lines is built, wherein the inspection has no
  • the man-machine is only responsible for collecting video images of the grid facilities and transmission lines in the target grid area, and the acquired video images are processed by the enhanced drone or the central base station, and then enter step S2.
  • Step S2 Based on the random inspection system model of UAV-assisted power grid lines, each inspection UAV in the UAV group collects video images of power grid facilities and transmission lines in the target power grid area, and obtains the inspection UAVs respectively. Collect the obtained video image data corresponding to each time slot, and then enter step S3.
  • Step S3 According to the video image data collected by each inspection UAV corresponding to each time slot, combined with the quality, signal transmission power, and position coordinates of each inspection UAV, the quality and signal transmission power of the UAV are enhanced , location coordinates, computing power, location coordinates of the central base station, and system communication bandwidth, construct a digital twin network of UAV-assisted power grid line random inspection system, as shown in Figure 3, used to fit each inspection UAV , enhance the position coordinates of the drone, and the resource status of the system, and then enter step S4.
  • step S3 specifically executes the following steps S31 to S33.
  • Step S31 According to the random inspection system of the UAV-assisted grid line, combined with the quality of each inspection UAV, and the video image data and signal transmission power collected corresponding to each time slot, the quality of the UAV and the corresponding Each time slot and each inspection UAV is assigned a CPU to calculate the frequency, signal transmission power, and the position coordinates of the central base station to build a real physical entity network, and then enter step S32.
  • Step S32 According to the actual physical entity network, according to the following formula:
  • Construct the digital twin model of each inspection UAV corresponding to each time slot in which, Represents the digital twin model of the m-th inspection drone corresponding to the t-th time slot, Indicates the quality of the mth inspection UAV, Indicates the video image data collected by the mth inspection drone corresponding to the tth time slot, Indicates the signal transmission power of the mth inspection UAV corresponding to the tth time slot, Indicates the position coordinates of the mth inspection drone corresponding to the tth time slot, Indicates the maximum signal transmission power of the mth inspection UAV corresponding to the tth time slot.
  • DT SUAV (t) represents the digital twin model of the enhanced UAV corresponding to the tth time slot
  • W SUAV represents the quality of the enhanced UAV
  • f SUAV ( t) represents the CPU computing frequency assigned to the enhanced UAV corresponding to the tth time slot
  • P SUAV (t) represents the signal transmission power of the enhanced UAV corresponding to the tth time slot
  • L SUAV (t) enhances the UAV Corresponding to the position coordinates of the tth time slot, Indicates the maximum signal transmission power of the enhanced UAV corresponding to the tth time slot
  • C SUAV indicates the number of CPU cycles required for the enhanced UAV to process 1-bit data.
  • Step S33 Based on the digital twin models of each inspection UAV corresponding to each time slot, the digital twin model of the enhanced UAV corresponding to each time slot, and the digital twin model of the central base station, a UAV-assisted random inspection of power grid lines is formed.
  • the digital twin network of the inspection system is used to fit the position coordinates of each inspection UAV, the enhanced UAV, and the resource status of the system, and then enter step S4.
  • Step S4 According to the digital twin network of the UAV-assisted power grid line random inspection system, based on the unloading delay of the power grid line random inspection system and the constraints of data task processing delay, construct the energy of the UAV group corresponding to each time slot. Consumption model or energy balance model, and further construct the energy consumption minimization objective function corresponding to each time slot of the UAV group or the energy balance minimization objective function corresponding to each time slot, and then enter step S5.
  • step S4 specifically executes the following steps S41 to S42.
  • Step S41 According to the digital twin network of the UAV-assisted power grid line random inspection system, construct the video image data collected by each inspection UAV in each time slot corresponding to the total time delay model under each unloading mode, and then enter step S42 .
  • step S41 is further specifically executed as the following steps S411 to S413.
  • Step S411 Based on the fact that each inspection UAV can only choose one between the enhanced UAV and the central base station for video image data unloading in any time slot, according to each inspection UAV and the enhanced UAV.
  • NOMA means of communication, that is, each inspection UAV shares a frequency spectrum to communicate with the enhanced UAV, and the communication between the enhanced UAV and the central base station uses OFDMA, each inspection UAV communicates with the enhanced UAV respectively
  • the data transmission rate corresponding to the tth time slot between UAVs The data transmission rate corresponding to the tth time slot between the inspection UAV m and the enhanced UAV in:
  • B represents the communication channel bandwidth
  • ⁇ 2 represents the additional Gaussian white noise
  • SIC continuous interference cancellation
  • the descending order of channel gain can be expressed as:
  • the kth channel gain in the descending sequence can be expressed as: ⁇ (k) ⁇ M; Indicates the interference of other inspection drones ⁇ k+1,..., ⁇ (M) ⁇ on the data transmission rate when the inspection drone m is uploading data;
  • OFDMA orthogonal frequency division multiple access technology
  • the channel power gain between the enhanced UAV and the central base station is defined as:
  • the video image data collected by the m-th inspection UAV in the t-th time slot is offloaded to the enhanced UAV for processing. Since the data volume of the processing result is small, the processing result can be ignored from the enhanced UAV to the central The transmission delay and transmission energy consumption of the base station; the video image data collected by the m-th inspection drone in the t-th time slot is unloaded to the central base station for processing. Since the central base station uses wired power supply, it can be ignored The calculation energy consumption of the central base station.
  • the inspection UAV m can only choose one of the unloading modes.
  • Construct the communication delay model corresponding to each time slot between each inspection UAV and enhanced UAV Indicates the communication delay corresponding to the t-th time slot between the m-th inspection UAV and the enhanced UAV, Indicates the video image data collected by the mth inspection drone corresponding to the tth time slot.
  • Construct the communication delay model of the video image data collected by each inspection UAV corresponding to each time slot and transmitted between the enhanced UAV and the central base station Indicates the communication delay between the transmission of the video image data collected by the mth inspection drone corresponding to the tth time slot between the enhanced drone and the central base station; then enter step S412.
  • Step S412 Based on definition Correspondingly, the video image data collected by the m-th inspection drone in the t-th time slot is unloaded to the enhanced drone for processing, according to the following formula:
  • C SUAV represents the number of CPU cycles required for the enhanced UAV to process 1-bit data
  • f SUAV (t) represents the CPU computing frequency allocated to the enhanced UAV corresponding to the t-th time slot.
  • the video image data is processed in a non-preemptive manner in descending order of channel power gain, according to the following formula:
  • Construct the queue waiting delay model corresponding to the video image data collected by the mth inspection UAV in the tth time slot before being processed by the enhanced UAV ⁇ (i) represents the serial number of the inspection UAV from which the enhanced UAV sequentially processes the i-th video image data
  • k represents the enhancement of the video image data collected by the m-th inspection UAV in the t-th time slot. The sequence number of the drone waiting to be processed.
  • the video image data collected by the m-th inspection drone in the t-th time slot is unloaded to the central base station for processing, according to the following formula:
  • Step S42 According to the video image data collected in each time slot of each inspection drone corresponding to the total delay model under each unloading mode, based on the grid line random inspection system unloading delay and data task processing delay constraints, Construct the energy balance model of the UAV swarm corresponding to each time slot, and further construct the energy balance minimization objective function of the UAV swarm corresponding to each time slot, and then enter step S5.
  • step S42 is further designed to execute the following steps S421 to S422.
  • the step S42 includes step S421 to step S422;
  • Step S421. Based on the central base station using wired power supply, according to the following formula:
  • Step S422. According to the energy consumption model E all (t) of the unmanned aerial vehicle group corresponding to the tth time slot, according to the following formula:
  • C5-C7 represent the preset constraints to enhance the range of activities of the UAV
  • C8 represents the conditional requirement for enhancing the full-duplex communication of the UAV
  • C9 represents the video collected by the m-th inspection UAV in the t-th time slot image data The offload processing needs to be completed within this time slot.
  • step S42 is further designed to also execute the following steps S421' to S422'.
  • Step S421' Based on the central base station using wired power supply, according to the following formula:
  • ⁇ represents the energy balance coefficient, Indicates the flight energy consumption of the mth inspection UAV in the tth time slot; flyE SUAV (t) represents the flight energy consumption of the enhanced UAV in the tth time slot; Indicates the video image data collected by the mth inspection drone in the tth time slot Unloading to the energy consumed by the enhanced UAV processing, ⁇ SUAV represents the effective switched capacitance corresponding to the CPU of the enhanced UAV; Indicates the video image data collected by the mth inspection drone in the tth time slot Transmitting energy consumption to and from the augmented UAV; represent data After enhancing the transmission energy consumption between the UAV and the central base station, then enter step S422.
  • Step S422' According to the energy balance model corresponding to the tth time slot of the UAV group According to the following formula:
  • C5-C7 represent the preset constraints to enhance the range of activities of the UAV
  • C8 represents the conditional requirement for enhancing the full-duplex communication of the UAV
  • C9 represents the video collected by the m-th inspection UAV in the t-th time slot image data The offload processing needs to be completed within this time slot.
  • Step S5. Randomly initially enhance the position coordinates of the UAV, and based on the position coordinates and video image data of each inspection UAV corresponding to the t-th time slot, construct the state of the system in the t-th time slot, and then enter step S6.
  • Step S6 Based on the position coordinates of the enhanced UAV and the state of the tth time slot of the system, the DDPG algorithm in deep reinforcement learning is used to minimize the objective function according to the energy balance of the UAV group corresponding to each time slot, and solve the problem of no
  • the human-machine group corresponds to the energy consumption model of each time slot, and obtains the state of the system in the tth time slot combined with the position coordinates of the enhanced UAV
  • each inspection UAV corresponds to the signal transmission power of the tth time slot
  • Each inspection UAV corresponds to the tth time slot respectively about the unloading method of the enhanced UAV or the central base station, the signal transmission power of the enhanced UAV corresponding to the tth time slot and the allocated CPU calculation frequency, so
  • the tth time slot action space of the system is formed, that is, the tth time slot action space of the system corresponding to the tth time slot state of the system combined with the position coordinates of the enhanced UAV, and then enters step S7
  • the Actor network group includes two deep neural networks with identical parameters: the Actor strategy network, with all parameters denoted as ⁇ ⁇ , and the Actor target network, with all parameters denoted as is ⁇ ⁇ ′
  • the critic network group includes two deep neural networks with exactly the same parameters: Critic policy network, all parameters are denoted as ⁇ Q
  • critic target network all parameters are denoted as ⁇ Q′ .
  • the input of the Actor strategy network is the current state of the system s t
  • the action space that can be selected based on the current system state s t is:
  • the obtained reward r t is defined as:
  • -1000 in the reward function represents a penalty item. If the conditional requirements for enhancing the full-duplex communication of the UAV are not met or the data collected by the inspection UAV m in the time slot t is not unloaded and processed in the time slot After completion, the preset penalty value -1000 will be given accordingly.
  • step S6 regarding the DDPG algorithm in deep reinforcement learning, as shown in FIG. 4 , the specific execution is as follows.
  • Step S7 Determine whether the iterative overflow condition is satisfied, and if so, enter step S8; otherwise, based on the system resource allocation in the tth time slot action space of the system corresponding to the position coordinates of the enhanced UAV, and the video
  • the image data unloading decision-making scheme uses the genetic algorithm to solve and update the position coordinates of the enhanced UAV, and returns to step S6.
  • the iteration overflow condition is the maximum preset number of iterations, or from the current iteration to the historical iteration direction, the variance of the energy consumption of the unmanned aerial vehicle group corresponding to the t-th time slot in each iteration within the preset iteration times is less than the preset energy consumption fluctuation range .
  • step S7 if the iteration overflow condition is not satisfied, the following steps S71 to S73 are performed.
  • Step S71 Randomly initialize the tth time slot population Among them, 1 ⁇ i ⁇ I, I represents the number of individuals in the tth time slot population K(t), Indicate the i-th position coordinate of the enhanced UAV in the population K(t) of the t-th time slot, and then enter step S72.
  • the phenotype of the enhanced UAV position coordinates is further converted into a genotype in the form of binary coding.
  • the binary coding method is as follows:
  • the range of x(t) is [x min , x max ], and the parameter is represented by a binary code symbol with length ⁇ , that is, the interval is divided into 2 ⁇ -1 parts, and similarly [y min ,y max ], [h min ,h max ] is also divided into 2 ⁇ -1 parts, the genotype corresponding to x(t) represents the data in the interval [0,x max -x min ], y(t), h(t) Similarly, the genotype of an individual can be expressed as:
  • Step S72 For each individual in the tth time slot population K(t), respectively, based on the tth time slot state of the system combined with the system resource allocation in the tth time slot action space corresponding to the position coordinates of the enhanced UAV , and the video image data unloading decision-making scheme, according to the following formula:
  • Step S73 Determine whether the fitness corresponding to each individual in the t-th time slot population K(t) satisfies the preset fitness threshold, and if so, select the individual corresponding to the highest fitness, that is, obtain the enhancement corresponding to the individual UAV position coordinates, that is, update the position coordinates of the enhanced UAV, and return to step S6; otherwise, based on the fitness of each individual in the tth time slot population K(t), the tth time slot population K(t) The data in (t) is selected, crossed, and mutated, and each individual in the population K(t) of the tth time slot is updated, and returns to step S72; among them, binary code conversion is adopted between steps S71 and S72 Corresponding to the operation, here is the decoding (y(t), h(t) are the same):
  • b i represents the binary number of the ith bit.
  • the preset fitness threshold here is the preset fitness lower limit, if the preset fitness threshold is the preset fitness lower limit, then it is judged that each individual in the tth time slot population K(t) is Whether the corresponding fitness is greater than the preset lower limit of fitness.
  • Step S8 According to the position coordinates of the enhanced UAV and the system resource allocation in the action space of the tth time slot of the corresponding system, and the video image data unloading decision-making scheme, each inspection UAV corresponds to each time slot in step S2
  • the acquired video image data is collected and processed, and the video image data is offloaded to the enhanced UAV or central base station for processing, and the enhanced UAV or central base station executes the power grid system for the video image data unloaded from the inspection UAV Defect identification, and power grid system defect location.
  • the Actor-Critic algorithm cannot reach the convergence state with the increase of training times, because the Actor-Critic algorithm needs to update the Actor network and the Critic network synchronously during the training process, and the choice of the Actor network action decision depends on the Critic network.
  • the Actor-Critic algorithm is more difficult to converge in some scenarios.
  • DQN and GA-DDPG cut off the correlation between the target Q value and the evaluation Q value during the training process, and promote the convergence of the critic network.
  • the DQN algorithm converges faster, the convergence effect is poor. Because the DQN algorithm used in this paper discretizes the continuous action space and narrows the breadth of the available action space, it cannot continuously and accurately find the optimal action decision, and the system equilibrium energy consumption fluctuates during the algorithm convergence stage.
  • the balanced energy consumption of the system increases with the increase of the number of inspection drones, and with the increase of the number of inspection drones, the gap between the balanced energy consumption of the system optimized by the GA-DDPG algorithm and the DQN algorithm gradually widens. This is because the number of variables in the action space increases with the number of inspection drones, more variables lead to an increase in the probability of the DQN algorithm skipping better actions, and then the optimization effect of the DQN algorithm gradually becomes worse. Finally, in the case of adopting the scheme of all unloading to enhanced drones, when the number of inspection drones is small, the effect difference between this scheme and DQN and GA-DDPG is not obvious.
  • the blue curve represents the scheme proposed in this paper
  • the purple curve represents the transmission power (PP) of the inspection UAV that is not optimized on the basis of the proposed scheme
  • the green curve represents the unoptimized PP and enhanced power of the UAV on the basis of the proposed scheme.
  • the red curve represents the unoptimized PP

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请涉及一种用于电网线路随机巡检的无人机辅助边缘计算方法，采用随机分布的巡检无人机对目标电网区域进行视频图像采集工作，可降低巡检的资金及时间成本；借助增强无人机辅助处理巡检无人机所采集的视频图像数据，以最小化系统无人机能耗为目标，在相同载能条件下延长无人机的工作时间；引入NOMA克服了移动无人机间通信产生的远近效应；而且采用深度强化学习中的DDPG算法与遗传算法相结合的方法求解增强无人机的位置坐标、系统资源分配、以及任务卸载决策方案，具有迭代速度快、时间复杂度低、系统实时性强的优点，能够确保系统无人机在能耗最小化的前提下实施电网线路巡检工作。

Description

一种用于电网线路随机巡检的无人机辅助边缘计算方法
相关申请
本申请要求于2022年2月25日提交中国专利局、申请号为2022101744973、申请名称为“一种用于电网巡检的无人机辅助边缘计算方法”的中国专利申请的优先权其全部内容通过引用结合在本申请中。
技术领域
本申请涉及移动边缘计算技术领域，具体涉及一种用于电网线路随机巡检的无人机辅助边缘计算方法。
背景技术
电力是国计民生的重要基础保障，电网的可靠性与安全性必须得到保障。在热电厂、变电站附近区域，输电线路的分布往往非常密集且错综复杂，其线路巡检显得尤为重要。针对恶劣部署环境下的电网线路，很难依靠传统的人工方式实施电网线路巡检。鉴于无人机良好的灵敏性、较低的风险性且易于部署，可以由其充当无线传感网络中的感知节点，负责数据的采集工作。另一方面，配置在无人机上的高速图像采集以及基于红外、紫外的传感器成像技术发展较为成熟，能够快速完成电网区域的视频图像采集工作，因此无人机辅助电网线路巡检方式可作为电力公司的一个高性价比的选择，其发展前景可观。
电网线路区域存在高压电辐射风险，不宜实施人工巡检，目前的通用办法是采用悬挂于输电线路的巡检机器人进行线路巡检，但其移动速度较慢，导致巡检周期长且巡检效率低。基于无人机辅助的随机巡检方式可以针对电网线路实现快速、高效巡检，省时且经济。本申请采用数字孪生网络构建无人机辅助电网线路随机巡检系统，该系统首次将非正交多址(NOMA)引入至电网线路巡检场景中，解决了电网线路巡检中移动无人机集群的通信远近效应问题。所谓远近效应是指无人机在移动过程中，增强无人机同时接收两个不同距离巡检无人机发来的信号时，由于距离较近的巡检无人机信号较强，距离较远的巡检无人机信号较弱，前者的强信号将会对后者产生严重的干扰，而引入NOMA可克服上述干扰。
发明内容
本申请所要解决的技术问题是：针对电网线路巡检中线路全覆盖及移动无人机集群的通信远近效应问题，提供一种用于电网线路随机巡检的无人机辅助边缘计算方法，采用全新模型，在完成电网线路巡检任务的条件下，实现了无人机能耗均衡的最小化，进而延长了无人机的工作时间。
本申请的各示例性的实施例为了解决上述技术问题采用以下技术方案：本申请设计了一种用于电网线路随机巡检的无人机辅助边缘计算方法，基于固定位置设置的中央基站，应用包含M个巡检无人机的无人机群以及一个增强无人机，针对包括电网设施、输电线路的目标电网区域进行巡检；包括如下步骤：
步骤S1.基于所述无人机群中各所述巡检无人机的飞行模式，构建无人机辅助电网线路随机巡检系统，其中，所述巡检无人机只负责对目标电网区中电网设施、输电线路进行视频图像采集，所获视频图像由所述增强无人机或所述中央基站进行数据处理，然后进入步骤S2；
步骤S2.基于无人机辅助电网线路随机巡检系统模型，由所述无人机群中各所述巡检无人机对目标电网区域中的电网设施、输电线路进行视频图像采集，获得各所述巡检无人机分别对应各时隙采集所获得的视频图像数据，然后进入步骤S3；
步骤S3.根据各所述巡检无人机分别对应各时隙采集所获得的所述视频图像数据，结合各所述巡检无人机的质量、信号发射功率、位置坐标，所述增强无人机的质量、信号发射功率、位置坐标、计算能力，所述中央基站的位置坐标、以及系统通信带宽，构建无人机辅助电网线路随机巡检系统的数字孪生网络，用于拟合各所述巡检无人机、所述增强无人机的位置坐标、以及所述系统的资源状态，然后进入步骤S4；
步骤S4.根据无人机辅助电网线路随机巡检系统的所述数字孪生网络，基于电网线路随机巡检系统卸载时延、以及数据任务处理时延约束条件，构建所述无人机群分别对应各时隙的能耗模型或能耗均衡模型，并进一步构建无人机群分别对应各时隙的能耗最小化目标函数或分别对应各时隙的能耗均衡最小化目标函数，然后进入步骤S5；
步骤S5.随机初始所述增强无人机的位置坐标，并基于各所述巡检无人机分别对应第t个时隙的位置坐标、视频图像数据，构建系统第t个时隙状态，然后进入步骤S6；
步骤S6.基于所述增强无人机的位置坐标，以及系统第t个时隙状态，采用深度强化学习中的DDPG算法，根据无人机群分别对应各时隙的能耗均衡最小化目标函数或能耗均衡最小化目标函数，求解所述无人机群分别对应各时隙的能耗模型，获得系统第t个时隙状态结合增强无人机的位置坐标所对应的，由各巡检无人机分别对应第t个时隙的信号发射功率、由各巡检无人机分别对应第t个时隙关于所述增强无人机或所述中央基站的卸载方式、所述增强无人机对应第t个时隙的信号发射功率和被分配的CPU计算频率，所组成的系统第t个时隙动作空间，即系统第t个时隙状态结合所述增强无人机位置坐标所对应的系统第t个时隙动作空间，然后进入步骤S7；
步骤S7.判断是否满足迭代溢出条件，是则进入步骤S8，否则基于系统第t个时隙状态结合所述增强无人机位置坐标所对应系统第t个时隙动作空间中的系统资源分配、以及所述视频图像数据卸载决策方案，采用遗传算法求解更新所述增强无人机的位置坐标，并返回步骤S6；
步骤S8.根据所述增强无人机的位置坐标以及相应系统第t个时隙动作空间中的系统资源分配、以及所述视频图像数据的卸载决策方案，针对步骤S2中各所述巡检无人机分别对应各时隙采集所获得的所述视频图像数据进行处理，实现将所述视频图像数据卸载至所述增强无人机或所述中央基站进行处理。
本申请所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，采用以上技术方案与现有技术相比，具有以下技术效果：
本申请所设计一种用于电网线路随机巡检的无人机辅助边缘计算方法，采用巡检无人机对目标电网区域进行视频图像采集工作，借助增强无人机辅助处理巡检无人机所采集的视频图像数据，以最小化系统无人机能耗为目标，采用深度强化学习中的DDPG算法与遗传算法相结合的方法求解增强无人机的位置坐标、系统资源分配、以及任务卸载决策方案，确保系统无人机在能耗最小化的前提下实施电网巡检；其中，考虑到电网巡检区域的环境较为恶劣，设计采用无人机进行目标电网区的视频图像采集工作，采用随机巡检的方式，降低了巡检成本；考虑到高速移动的无人机之间通信产生的远近效应，本申请引入NOMA克服了该缺点；并以优化系统无人机能耗为目标，在同载能的条件下延长了无人机的工作时间；以及采用DDPG算法与遗传算法相结合的方法，求解增强无人机的位置坐标、系统资源分配、以及任务卸载决策方案，该方法迭代速度较快，时间复杂度不高，可以提高系统的实时性；最后，采用随机巡检的方式，进一步节约了巡检成本。
附图说明
图1为本申请一实施例中所设计融合移动边缘计算的无人机辅助电网线路随机巡检方法实施流程图；
图2为本申请一实施例中设计应用实施例中无人机辅助电网线路随机巡检系统模型图；
图3为本申请一实施例中设计应用实施例的无人机辅助电网线路随机巡检的数字孪生网络结构示意图；
图4为本申请一实施例中设计应用实施例用于求解系统资源分配、以及任务卸载决策方案的DDPG示意图；
图5为本申请一实施例中设计应用实施例对应不同算法方案下平均系统均衡能耗表现图；
图6为本申请一实施例中设计应用实施例对应不同算法方案下巡检无人机数量与系统均衡能耗之间的关系图；
图7为本申请一实施例中设计应用实施例对应不同方案下相对D值的系统均衡能耗比较。
具体实施方式
为进一步降低巡检成本，本申请提出了一种基于适用于电网线路巡检的无人机辅助边缘计算方法，考虑到无人机载能的有限性，在借助无人机辅助电网线路巡检的同时，要尽可能降低无人机的能耗，从而在相同能耗条件下延长无人机的工作时间，使得无人机的连续工作能力进一步增强，提高巡检效益。具体而言，基于数字孪生网络所提供的信息，通过联合优化计算资源、通信资源、无人机轨迹以及任务卸载决策实现无人机群能耗均衡最小化的目标。考虑到巡检场景中的时延要求敏感，变量间耦合性较高，且数字孪生网络具有时变特征(系无人机在不同时隙的位置不同导致)，因此本申请采用遗传算法与强化学习相结合的算法(GA-DDPG)解决上述目标的优化问题。基于训练好的策略，强化学习可快速给出行动策略，适合具有时变特征问题的求解。在GA-DDPG强化学习中的智能体需要获取全面和准确的系统状态信息，本申请将数字孪生嵌入GA-DDPG算法中，构建物理对象与虚拟模型之间的映射，进而实现上述目标，GA-DDPG中的遗传算法用于降低强化学习算法中决策空间的维度，加快算法整体的训练速度。
现在将参考附图更全面地描述示例实施例。然而，示例实施例能够以多种形式实施，且不应被理解为限于在此阐述的实施例。相反，提供这些实施例使得本申请将全面和完整，并将示例实施例的构思全面地传达给本领域的技术人员。在图中相同的附图标记表示相同或类似的部分，因而将省略对它们的重复描述。
所描述的特征、结构或特性可以以任何合适的方式结合在一个或更多实施例中。在下面的描述中，提供许多具体细节从而给出对本公开的实施例的充分理解。然而，本领域技术人员将意识到，可以实践本公开的技术方案而没有这些特定细节中的一个或更多，或者可以采用其它的方式、组元、材料、装置或操作等。在这些情况下，将不详细示出或描述公知结构、方法、装置、实现、材料或者操作。
附图中所示的流程图仅是示例性说明，不是必须包括所有的内容和操作/步骤，也不是必须按所描述的顺序执行。例如，有的操作/步骤还可以分解，而有的操作/步骤可以合并或部分合并，因此实际执行的顺序有可能根据实际情况改变。
下面结合说明书附图对本申请的具体实施方式作进一步详细的说明。
本申请所设计一种用于电网线路随机巡检的无人机辅助边缘计算方法，如图2所示，基于固定位置设置的中央基站，应用包含M个巡检无人机(UAV)的无人机群以及一个增强无人机(SUAV)，针对包括电网设施、输电线路的目标电网区域进行巡检，其中，各巡检无人机分别均配备高速图像捕获模块；在一实施例中，如图1所示，具体执行如下步骤S1至步骤S8。
步骤S1.基于无人机群中各巡检无人机的飞行模式，构建无人机辅助电网线路随机巡检系统，其中，巡检无人机只负责对目标电网区中电网设施、输电线路进行视频图像采集，所获视频图像由增强无人机或中央基站进行数据处理，然后进入步骤S2。
在一实施例中，上述步骤S1具体执行如下步骤S11至步骤S13。
步骤S11.基于各时隙内各巡检无人机的运动状态保持不变，分别针对各巡检无人机，按如下公式：
Figure PCTCN2022130532-appb-000001
Figure PCTCN2022130532-appb-000002
Figure PCTCN2022130532-appb-000003
获得第m个巡检无人机对应第t个时隙的移动速度v m(t)、水平移动方向α m(t)、竖直移动方向β m(t)，其中，1≤m≤M，
Figure PCTCN2022130532-appb-000004
表示所有巡检无人机的平均移动速度，
Figure PCTCN2022130532-appb-000005
表示第m个巡检无人机对应前t-1个时隙内的平均水平方向移动角度，
Figure PCTCN2022130532-appb-000006
表示第m个巡检无人机对应前t-1个时隙内的平均竖直方向移动角度；v m(t-1)、α m(t-1)、β m(t-1)依次表示第m个巡检无人机对应前t-1个时隙的移动速度、水平移动方向、竖直移动方向，0＜λ 1＜1，λ 1表示预设用于调整巡检无人机对应前t-1个时隙移动速度对第t个时隙移动速度的影响参数，0＜λ 2＜1，λ 2表示预设用于调整巡检无人机对应前t-1个时隙水平移动方向对第t个时隙水平移动方向的影响参数，0＜λ 3＜1，λ 3表示预设用于调整巡检无人机对应前t-1个时隙竖直移动方向对第t个时隙竖直移动方向的影响参数，预设服从独立高斯分布的参数φ m表示第m个巡检无人机移动速度的随机性，预设服从独立高斯分布的参数ψ m表示第m个巡检无人机水平移动方向的随机性，预设服从独立高斯分布的参数
Figure PCTCN2022130532-appb-000007
表示第m个巡检无人机竖直移动方向的随机性，然后进入步骤S12。
步骤S12.根据各时隙长度τ，分别针对各巡检无人机，按如下公式：
x m(t)＝x m(t-1)+v m(t-1)cos(α m(t-1))τ
y m(t)＝y m(t-1)+v m(t-1)sin(α m(t-1))τ
h m(t)＝h m(t-1)+v m(t-1)sin(β m(t-1))τ
获得第m个巡检无人机对应第t个时隙的位置坐标
Figure PCTCN2022130532-appb-000008
其中，x m(t)、y m(t)、h m(t)表示第m个巡检无人机对应第t个时隙分别在x、y、z坐标轴上的值，x m(t-1)、y m(t-1)、h m(t-1)表示第m个巡检无人机对应前t-1个时隙分别在x、y、z坐标轴上的值，然后进入步骤S13。
步骤S13.根据各巡检无人机分别对应第t个时隙的移动速度、水平移动方向、竖直移动方向、位置坐标，搭建无人机辅助电网线路随机巡检系统，其中，巡检无人机只负责对目标电网区中电网设施、输电线路进行视频图像采集，所获视频图像由增强无人机或中央基站进行数据处理，然后进入步骤S2。
步骤S2.基于无人机辅助电网线路随机巡检系统模型，由无人机群中各巡检无人机对目标电网区中电网设施、输电线路进行视频图像采集，获得各巡检无人机分别对应各时隙采集所获得的视频图像数据，然后进入步骤S3。
步骤S3.根据各巡检无人机分别对应各时隙采集所获得的视频图像数据，结合各巡检无人机的质量、信号发射功率、位置坐标，增强无人机的质量、信号发射功率、位置坐标、计算能力，中央基站的位置坐标、以及系统通信带宽，构建无人机辅助电网线路随机巡检系统的数字孪生网络，如图3所示，用于拟合各巡检无人机、增强无人机的位置坐标、以及系统的资源状态，然后进入步骤S4。
在一实施例中，上述步骤S3具体执行如下步骤S31至步骤S33。
步骤S31.根据无人机辅助电网线路随机巡检系统，结合各巡检无人机的质量、以及分别对应各时隙所采集的视频图像数据、信号发射功率，增强无人机的质量、对应各时隙各巡检无人机被分配的CPU计算频率、信号发射功率，以及中央基站的位置坐标，构建现实物理实体网络，然后进入步骤S32。
步骤S32.根据现实物理实体网络，按如下公式：
Figure PCTCN2022130532-appb-000009
构建各巡检无人机分别对应各时隙的数字孪生模型，其中，
Figure PCTCN2022130532-appb-000010
表示第m个巡检无人机对应第t个时隙的数字孪生模型，
Figure PCTCN2022130532-appb-000011
表示第m个巡检无人机的质量，
Figure PCTCN2022130532-appb-000012
表示第m个巡检无人机对应第t个时隙所采集的视频图像数据，
Figure PCTCN2022130532-appb-000013
表示第m个巡检无人机对应第t个时隙的信号发射功率，
Figure PCTCN2022130532-appb-000014
表示第m个巡检无人机对应第t个时隙的位置坐标，
Figure PCTCN2022130532-appb-000015
表示第m个巡检无人机对应第t个时隙的最大信号发射功率。
同时按如下公式：
Figure PCTCN2022130532-appb-000016
构建增强无人机对应各时隙的数字孪生模型，其中，DT SUAV(t)表示增强无人机对应第t个时隙的数字孪生模型，W SUAV表示增强无人机的质量，f SUAV(t)表示增强无人机对应第t个时隙被分配的CPU计算频率，P SUAV(t)表示增强无人机对应第t个时隙的信号发射功率，L SUAV(t)增强无人机对应第t个时隙的位置坐标，
Figure PCTCN2022130532-appb-000017
表示增强无人机对应第t个时隙的最大信号发射功率，
Figure PCTCN2022130532-appb-000018
表示增强无人机的最大CPU计算频率，C SUAV表示增强无人机处理1bit的数据所需要CPU的周期数。
以及按如下公式：
DT BS＝{L BS}
构建中央基站的数字孪生模型DT BS，其中，L BS表示中央基站的位置坐标，然后进入步骤S33。
步骤S33.基于各巡检无人机分别对应各时隙的数字孪生模型、以及增强无人机对应各时隙的数字孪生模型、中央基站的数字孪生模型，构成无人机辅助电网线路随机巡检系统的数字孪生网络，用于拟合各巡检无人机、增强无人机的位置坐标、以及系统的资源状态，然后进入步骤S4。
步骤S4.根据无人机辅助电网线路随机巡检系统的数字孪生网络，基于电网线路随机巡检系统卸载时延、以及数据任务处理时延约束条件，构建无人机群分别对应各时隙的能耗模型或能耗均衡模型，并进一步构建无人机群分别对应各时隙的能耗最小化目标函数或分别对应各时隙的能耗均衡最小化目标函数，然后进入步骤S5。
在一实施例中，上述步骤S4具体执行如下步骤S41至步骤S42。
步骤S41.根据无人机辅助电网线路随机巡检系统的数字孪生网络，构建各巡检无人机各时隙所采集视频图像数据分别对应各卸载方式下的总时延模型，然后进入步骤S42。
这里上述步骤S41进一步具体执行如下步骤S411至步骤S413。
步骤S411.基于各巡检无人机在任一时隙内仅能在增强无人机与中央基站之间择一进行视频图像数据卸载，根据各巡检无人机分别与增强无人机之间采用NOMA的方式进行通信，即各巡检无人机共用一个频谱与增强无人机进行通信，以及增强无人机与中央基站之间采用OFDMA的方式进行通信，各巡检无人机分别与增强无人机之间对应第t个时隙的数据传输速率
Figure PCTCN2022130532-appb-000019
巡检无人机m与增强无人机之间对应第t个时隙的数据传输速率
Figure PCTCN2022130532-appb-000020
其中：
Figure PCTCN2022130532-appb-000021
其中，B表示通信信道带宽，σ 2表示附加的高斯白噪声；
Figure PCTCN2022130532-appb-000022
表示在时隙t内，巡检无人机m与增强无人机之间的信道功率增益，其被定义为：
Figure PCTCN2022130532-appb-000023
g 0表示单位距离的路径损耗；增强无人机接收端采用连续干扰消除的方式(SIC)对M个巡检无人机发射的叠加信号进行解码，其解码顺序按照信道增益的降序执行，在第t个时隙内，信道增益的降序可表示为：
Figure PCTCN2022130532-appb-000024
降序序列中的第k个信道增益可表示为：ρ(k)∈M；
Figure PCTCN2022130532-appb-000025
表示巡检无人机m在上传数据时，其余巡检无人机{k+1,…,ρ(M)}对数据传输速率的干扰；
在任意时隙内，增强无人机与中央基站之间采用OFDMA(正交频分多址接入技术)的方式进行通信，根据香农公式，增强无人机与中央基站之间的数据传输速率为：
Figure PCTCN2022130532-appb-000026
其中，
Figure PCTCN2022130532-appb-000027
表示在第t个时隙内，增强无人机与中央基站之间的信道功率增益，其被定义为：
Figure PCTCN2022130532-appb-000028
对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理，由于处理结果的数据量较小，因此可以忽略处理结果从增强无人机至中央基站的传输时延及传输能耗；对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到中央基站进行处理，由于中央基站采用有线的方式进行供电，因此可忽略中央基站的计算能耗， 此外，在任一时隙内，巡检无人机m仅能选择其中一种卸载模式。
进一步按如下公式：
Figure PCTCN2022130532-appb-000029
构建各巡检无人机与增强无人机之间分别对应各时隙的通信时延模型
Figure PCTCN2022130532-appb-000030
Figure PCTCN2022130532-appb-000031
表示第m个巡检无人机与增强无人机之间对应第t个时隙的通信时延，
Figure PCTCN2022130532-appb-000032
表示第m个巡检无人机对应第t个时隙所采集的视频图像数据。
以及按如下公式：
Figure PCTCN2022130532-appb-000033
构建各巡检无人机分别对应各时隙所采集视频图像数据在增强无人机与中央基站之间传输的通信时延模型
Figure PCTCN2022130532-appb-000034
表示第m个巡检无人机对应第t个时隙所采集视频图像数据在增强无人机与中央基站之间传输的通信时延；然后进入步骤S412。
步骤S412.基于定义
Figure PCTCN2022130532-appb-000035
对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理，则按如下公式：
Figure PCTCN2022130532-appb-000036
构建增强无人机接收端针对第m个巡检无人机在第t个时隙所采集视频图像数据的数据处理时延模型
Figure PCTCN2022130532-appb-000037
其中，C SUAV表示增强无人机处理1bit的数据所需要CPU的周期数，f SUAV(t)表示增强无人机对应第t个时隙被分配的CPU计算频率。
并基于增强无人机采用非抢占的方式针对视频图像数据按照信道功率增益降序的方式进行处理，则按如下公式：
Figure PCTCN2022130532-appb-000038
构建第m个巡检无人机在第t个时隙所采集视频图像数据在被增强无人机处理之前所对应的队列等待时延模型
Figure PCTCN2022130532-appb-000039
ρ(i)表示增强无人机依次处理第i个视频图像数据所来自巡检无人机的序号，k表示第m个巡检无人机在第t个时隙所采集视频图像数据在 增强无人机等待处理的排序序号。
进而按如下公式：
Figure PCTCN2022130532-appb-000040
构建第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理所对应的总时延模型T m,0(t)，然后进入步骤S413。
步骤S413.基于定义
Figure PCTCN2022130532-appb-000041
对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到中央基站进行处理，则按如下公式：
Figure PCTCN2022130532-appb-000042
构建第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理所对应的总时延模型T m,1(t)，然后进入步骤S42。
步骤S42.根据各巡检无人机各时隙所采集视频图像数据分别对应各卸载方式下的总时延模型，基于电网线路随机巡检系统卸载时延、以及数据任务处理时延约束条件，构建无人机群分别对应各时隙的能耗均衡模型，并进一步构建无人机群分别对应各时隙的能耗均衡最小化目标函数，然后进入步骤S5。
在一实施例中，上述步骤S42进一步设计执行如下步骤S421至步骤S422。
所述步骤S42包括步骤S421至步骤S422；
步骤S421.基于中央基站采用有线方式进行供电，按如下公式：
Figure PCTCN2022130532-appb-000043
构建无人机群对应第t个时隙的能耗模型E all(t)，其中，
Figure PCTCN2022130532-appb-000044
表示第m个巡检无人机在第t个时隙内的飞行能耗；
Figure PCTCN2022130532-appb-000045
flyE SUAV(t)表示增强无人机在第t个时隙内的飞行能耗；
Figure PCTCN2022130532-appb-000046
表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000047
卸载至增强无人机处理所消耗的能量，κ SUAV表示增强无人机CPU对应的有效开关电容；
Figure PCTCN2022130532-appb-000048
表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000049
在与增强无人机之间的传输能耗；
Figure PCTCN2022130532-appb-000050
表示数据
Figure PCTCN2022130532-appb-000051
在增强无人机与中央基站之间的传输能耗，然后进入步骤S422；
步骤S422.根据无人机群对应第t个时隙的能耗模型E all(t)，按如下公式：
Figure PCTCN2022130532-appb-000052
进一步构建无人机群分别对应各时隙的能耗最小化目标函数
Figure PCTCN2022130532-appb-000053
其中，C5-C7表示预设约束增强无人机的活动范围，C8表示增强无人机全双工通信的条件需求，C9表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000054
需要在该时隙内卸载处理完毕。
在一实施例中，上述步骤S42进一步设计还可以执行如下步骤S421’至步骤S422’。
步骤S421’.基于中央基站采用有线方式进行供电，按如下公式：
Figure PCTCN2022130532-appb-000055
构建无人机群对应第t个时隙的能耗均衡模型
Figure PCTCN2022130532-appb-000056
其中，χ表示能耗均衡系数，
Figure PCTCN2022130532-appb-000057
表示第m个巡检无人机在第t个时隙 内的飞行能耗；
Figure PCTCN2022130532-appb-000058
flyE SUAV(t)表示增强无人机在第t个时隙内的飞行能耗；
Figure PCTCN2022130532-appb-000059
表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000060
卸载至增强无人机处理所消耗的能量，κ SUAV表示增强无人机CPU对应的有效开关电容；
Figure PCTCN2022130532-appb-000061
表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000062
在与增强无人机之间的传输能耗；
Figure PCTCN2022130532-appb-000063
表示数据
Figure PCTCN2022130532-appb-000064
在增强无人机与中央基站之间的传输能耗，然后进入步骤S422。
步骤S422’.根据无人机群对应第t个时隙的能耗均衡模型
Figure PCTCN2022130532-appb-000065
按如下公式：
Figure PCTCN2022130532-appb-000066
进一步构建无人机群分别对应各时隙的能耗均衡最小化目标函数
Figure PCTCN2022130532-appb-000067
其中，C5-C7表示预设约束增强无人机的活动范围，C8表示增强无人机全双工通信的条件需求，C9表示第m个巡检无人机在第t个时隙所采集视频图像数据
Figure PCTCN2022130532-appb-000068
需要在该时隙内卸载处理完毕。
步骤S5.随机初始增强无人机的位置坐标，并基于各巡检无人机分别对应第t个时隙的位置坐标、视频图像数据，构建系统第t个时隙状态，然后进入步骤S6。
步骤S6.基于增强无人机的位置坐标，以及系统第t个时隙状态，采用深度强化学习 中的DDPG算法，根据无人机群分别对应各时隙的能耗均衡最小化目标函数，求解无人机群分别对应各时隙的能耗模型，获得系统第t个时隙状态结合增强无人机的位置坐标所对应的，由各巡检无人机分别对应第t个时隙的信号发射功率、由各巡检无人机分别对应第t个时隙关于增强无人机或中央基站的卸载方式、增强无人机对应第t个时隙的信号发射功率和被分配的CPU计算频率，所组成的系统第t个时隙动作空间，即系统第t个时隙状态结合增强无人机位置坐标所对应的系统第t个时隙动作空间，然后进入步骤S7。
关于上述步骤S6，具体执行如下操作：
首先构建两组神经网络，分别为Actor网络组以及Critic网络组，其中Actor网络组包括两个参数完全相同的深度神经网络：Actor策略网络，所有参数记为θ μ、Actor目标网络，所有参数记为θ μ′，Critic网络组包括两个参数完全相同的深度神经网络：Critic策略网络，所有参数记为θ Q、Critic目标网络，所有参数记为θ Q′。
然后基于增强无人机的位置坐标，在第t个时隙内，Actor策略网络输入的是系统当前状态s t，输出行动μ(s t)附加随机噪声N t形成行动决策a t与环境进行交互，即a t＝μ(s t|θ μ)+N t，进而得到奖励r t并进入系统下一个时隙状态s t+1，同时将这一记录{s t,a t,r t,s t+1}存至经验回放池中；
其中，系统当前状态s t、动作空间a t以及奖励函数r t分别如下表示：
Figure PCTCN2022130532-appb-000069
基于当前系统状态s t可选择的动作空间为：
Figure PCTCN2022130532-appb-000070
基于当前系统状态s t以及该状态上的行动决策a t，所获得的奖励r t定义为：
Figure PCTCN2022130532-appb-000071
其中，奖励函数中的-1000代表惩罚项，若增强无人机全双工通信的条件需求未得到满足或t时隙内巡检无人机m所采集的数据未在该时隙内卸载处理完毕，则会相应给出预设惩罚数值-1000。
上述关于步骤S6的具体执行操作，在一实施例中，关于深度强化学习中的DDPG算法，如图4所示，具体如下执行。
S61，从第1个时隙开始，重复执行上述操作，直至经验回放池被填满；
S62，从经验回放池中随机抽取N个样本，并记其中的一个样本为{s i,a i,r i,s i+1}；
S63，将状态s i+1、行动决策μ′(s i+1|θ μ′)输入至Critic目标网络，输出基于当前状态以及行动决策所获得的Q值：Q′(s i+1,μ′(s i+1|θ μ′)|θ Q′)，其中行动决策μ′(s i+1|θ μ′)基于s i+1由Actor目标网络提供，并记：y i＝r i+γQ′(s i+1,μ′(s i+1|θ μ′)|θ Q′)；
S64，将状态s i、行动决策a i输入至Critic策略网络，输出基于当前状态以及行动决策所获得的Q值：Q(s i,a i|θ Q)；
S65，采用如下损失函数更新Critic策略网络的参数θ Q：
Figure PCTCN2022130532-appb-000072
S66，采用策略梯度上升方法更新Actor策略网络的参数θ μ，实现策略目标函数J(θ μ)的最大化：
Figure PCTCN2022130532-appb-000073
其中，
Figure PCTCN2022130532-appb-000074
为Actor策略网络基于状态s i所获取的行动决策，
Figure PCTCN2022130532-appb-000075
为Critic策略网络基于状态s i以及行动决策
Figure PCTCN2022130532-appb-000076
所获取的Q值；
S67，采用软更新的方式定时更新Actor目标网络的参数θ μ′以及Critic目标网络的参数θ Q′：
″
θ μ＝υθ μ+(1-υ)θ μ
θ Q′＝υθ Q+(1-υ)θ Q′
步骤S7.判断是否满足迭代溢出条件，是则进入步骤S8，否则基于系统第t个时隙状态结合增强无人机位置坐标所对应系统第t个时隙动作空间中的系统资源分配、以及视频图像数据卸载决策方案，采用遗传算法求解更新增强无人机的位置坐标，并返回步骤S6。
其中，迭代溢出条件为最大预设迭代次数，或者以当前迭代起向历史迭代方向、预设 迭代次数内各迭代下无人机群对应第t个时隙能耗的方差小于预设能耗波动范围。
在一实施例中，上述步骤S7中，若不满足迭代溢出条件，则执行如下步骤S71至步骤S73。
步骤S71.随机初始化第t个时隙种群
Figure PCTCN2022130532-appb-000077
其中，1≤i≤I，I表示第t个时隙种群K(t)中个体的个数，
Figure PCTCN2022130532-appb-000078
表示第t个时隙种群K(t)中增强无人机第i个位置坐标，然后进入步骤S72。
实际应用中，进一步将增强无人机位置坐标的表现型采用二进制编码的形式转换为基因型，二进制编码方式具体如下：
x(t)的范围为[x min,x max]，用长度为ε的二进制编码符号来表示该参数，即该区间被划分为2 ε-1份，同理[y min,y max]、[h min,h max]也被划分为2 ε-1份，x(t)所对应的基因型表示区间[0,x max-x min]中的数据，y(t)、h(t)同理，因此某一个体的基因型可表示为：
Figure PCTCN2022130532-appb-000079
步骤S72.分别针对第t个时隙种群K(t)中的各个个体，基于系统第t个时隙状态结合增强无人机位置坐标所对应系统第t个时隙动作空间中的系统资源分配、以及视频图像数据卸载决策方案，按如下公式：
Figure PCTCN2022130532-appb-000080
获得第t个时隙种群K(t)中各个个体分别对应的适应度，然后进入步骤S73。
步骤S73.判断第t个时隙种群K(t)中各个个体分别所对应适应度是否均满足预设适应度阈值，是则挑选最高适应度所对应的个体，即获得该个体所对应的增强无人机位置坐标，即更新增强无人机的位置坐标，并返回步骤S6；否则基于第t个时隙种群K(t)中各个个体分别对应的适应度，对第t个时隙种群K(t)中的数据进行选择、交叉、变异操作，更新第t个时隙种群K(t)中的各个个体，并返回步骤S72；其中，与步骤S71、步骤S72之间采用二进制编码形式转换操作相对应的，这里解码(y(t)、h(t)同理)：
Figure PCTCN2022130532-appb-000081
其中，b i表示第i位的二进制数。
在一实施例中，这里的预设适应度阈值为预设适应度下限，若预设适应度阈值为预设适应度下限时，则判断第t个时隙种群K(t)中各个个体分别所对应适应度是否均大于预设适应度下限。
步骤S8.根据增强无人机的位置坐标以及相应系统第t个时隙动作空间中的系统资源分配、以及视频图像数据卸载决策方案，针对步骤S2中各巡检无人机分别对应各时隙采集所获得的视频图像数据进行处理，实现视频图像数据卸载至增强无人机或中央基站进行处理，由增强无人机或中央基站针对来自巡检无人机所卸载的视频图像数据执行电网系统缺陷识别、以及电网系统缺陷定位。
将本申请所设计融合移动边缘计算的无人机辅助电网线路随机巡检方法，应用于实际当中，图5展示了在M＝3情况下不同算法方案之间的性能比较。其中Actor-Critic算法无法随训练次数的增加达到收敛状态，这是因为Actor-Critic算法在训练的过程中需要同步更新Actor网络以及Critic网络，而Actor网络行动决策的选择依赖于Critic网络给与的价值评估，考虑到Critic网络自身难以收敛，因此Actor-Critic算法在一些场景下更加难以收敛。相比之下，得益于Critic评估网络和Critic目标网络的双重网络结构，DQN以及GA-DDPG在训练过程中切断了目标Q值与评估Q值之间的相关性，促使Critic网络收敛。此外，由图可知DQN算法约在Episode＝90处收敛，GA-DDPG算法约在Episode＝200处收敛，相对GA-DDPG算法而言，DQN算法虽然收敛速度较快但收敛效果较差，这是因为本文所采用的DQN算法将连续动作空间离散化，缩小了可用动作空间的广度，导致其无法持续准确找到最佳动作决策，进而系统均衡能耗在算法收敛阶段出现波动现象。
使用算法收敛后获得的均衡能耗结果，我们比较了不同巡检无人机(PUAV)数量设置下的三种算法方案，具体包括GA-DDPG、DQN以及全部计算任务卸载至增强无人机三种方案，结果如图6所示。我们可以观察到，对于相同的巡检无人机数量，GA-DDPG算法优化的系统均衡能耗相对DQN较低。这是因为GA-DDPG算法探索了一个连续的动作空间，并采取了精确的动作，最终获得了最优策略，显著减少了系统均衡能耗，而DQN算法中的动作离散化可能会导致其跳过更好的动作。此外，系统均衡能耗随着巡检无人机数量的增加而增加且随着巡检无人机数量的增加，GA-DDPG算法优化的系统均衡能耗与DQN算法的差距逐渐拉大，这是因为动作空间中变量的个数随着巡检无人机数量的增多而增多，较多的变量导致DQN算法跳过更好动作的概率增加，进而DQN算法的优化效 果逐渐变差。最后，在采用全部卸载至增强无人机方案的情况下，当巡检无人机的数量较小时，该方案与DQN以及GA-DDPG效果差距不明显，随着巡检无人机数量的增多，该方案的劣势逐渐凸显，这是因为嵌入至增强无人机端的MEC服务器无法满足更多的计算需求，此时将个别巡检无人机所采集的计算任务卸载至中央基站更为合理。
图7比较了当M＝3时不同方案下相对D值的系统均衡能耗(我们认定在任意时隙巡检无人机所采集的数据量服从均值为D的高斯分布)。其中，蓝色曲线代表本文所提议的方案，紫色曲线代表在所提议方案基础上未优化巡检无人机的发射功率(PP)，绿色曲线代表在所提议方案基础上未优化PP与增强无人机的发射功率(SP)，红色曲线代表在所提议方案基础上未优化PP、SP以及增强无人机的计算资源(SC)。由图可知以下几点，首先，随着D值的增大，上述四种方案的系统均衡能耗分别增加，这是因为在一般情况下，D值的增大意味着在不同时隙内各巡检无人机所采集的任务量增多，导致更多的计算资源以及通信资源被消耗。其次，通过联合优化PP、SP以及SC，本文所提议方案的性能得到明显的提高并且优于其余三种方案。最后，我们可以观察到，蓝色曲线与紫色曲线之间的性能差距相对明显，这是因为巡检无人机的数量不为一，因此优化PP相当于优化多个变量，而多变量的同步优化进一步提高了蓝色曲线的性能。
上面结合附图对本申请的实施方式作了详细说明，但是本申请并不限于上述实施方式，在本领域普通技术人员所具备的知识范围内，还可以在不脱离本申请宗旨的前提下做出各种变化。

Claims (11)

  1. 一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：基于固定位置设置的中央基站，应用包含M个巡检无人机的无人机群以及一个增强无人机，针对包括电网设施、输电线路的目标电网区域进行巡检；包括如下步骤：
    步骤S1.基于所述无人机群中各所述巡检无人机的飞行模式，构建无人机辅助电网线路随机巡检系统，其中，所述巡检无人机只负责对目标电网区中电网设施、输电线路进行视频图像采集，所获视频图像由所述增强无人机或所述中央基站进行数据处理，然后进入步骤S2；
    步骤S2.基于无人机辅助电网线路随机巡检系统模型，由所述无人机群中各所述巡检无人机对目标电网区域中的电网设施、输电线路进行视频图像采集，获得各所述巡检无人机分别对应各时隙采集所获得的视频图像数据，然后进入步骤S3；
    步骤S3.根据各所述巡检无人机分别对应各时隙采集所获得的所述视频图像数据，结合各所述巡检无人机的质量、信号发射功率、位置坐标，所述增强无人机的质量、信号发射功率、位置坐标、计算能力，所述中央基站的位置坐标、以及系统通信带宽，构建无人机辅助电网线路随机巡检系统的数字孪生网络，用于拟合各所述巡检无人机、所述增强无人机的位置坐标、以及所述系统的资源状态，然后进入步骤S4；
    步骤S4.根据无人机辅助电网线路随机巡检系统的所述数字孪生网络，基于电网线路随机巡检系统卸载时延、以及数据任务处理时延约束条件，构建所述无人机群分别对应各时隙的能耗模型或能耗均衡模型，并进一步构建无人机群分别对应各时隙的能耗最小化目标函数或分别对应各时隙的能耗均衡最小化目标函数，然后进入步骤S5；
    步骤S5.随机初始所述增强无人机的位置坐标，并基于各所述巡检无人机分别对应第t个时隙的位置坐标、视频图像数据，构建系统第t个时隙状态，然后进入步骤S6；
    步骤S6.基于所述增强无人机的位置坐标，以及系统第t个时隙状态，采用深度强化学习中的DDPG算法，根据无人机群分别对应各时隙的能耗均衡最小化目标函数或能耗均衡最小化目标函数，求解所述无人机群分别对应各时隙的能耗模型，获得系统第t个时隙状态结合增强无人机的位置坐标所对应的，由各巡检无人机分别对应第t个时隙的信号发射功率、由各巡检无人机分别对应第t个时隙关于所述增强无人机或所述中央基站的卸载方式、所述增强无人机对应第t个时隙的信号发射功率和被分配的CPU计算频率，所组成的系统第t个时隙动作空间，即系统第t个时隙状态结合所述增强无人机位置坐标所对应的系统第t个时隙动作空间，然后进入步骤S7；
    步骤S7.判断是否满足迭代溢出条件，是则进入步骤S8，否则基于系统第t个时隙状态结 合所述增强无人机位置坐标所对应系统第t个时隙动作空间中的系统资源分配、以及所述视频图像数据卸载决策方案，采用遗传算法求解更新所述增强无人机的位置坐标，并返回步骤S6；
    步骤S8.根据所述增强无人机的位置坐标以及相应系统第t个时隙动作空间中的系统资源分配、以及所述视频图像数据的卸载决策方案，针对步骤S2中各所述巡检无人机分别对应各时隙采集所获得的所述视频图像数据进行处理，实现将所述视频图像数据卸载至所述增强无人机或所述中央基站进行处理。
  2. 根据权利要求1所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S1包括如下步骤S11至步骤S13；
    步骤S11.基于各时隙内各所述巡检无人机的运动状态保持不变，分别针对各所述巡检无人机，按如下公式：
    Figure PCTCN2022130532-appb-100001
    Figure PCTCN2022130532-appb-100002
    Figure PCTCN2022130532-appb-100003
    获得第m个巡检无人机对应第t个时隙的移动速度v m(t)、水平移动方向α m(t)、竖直移动方向β m(t)，其中，1≤m≤M，
    Figure PCTCN2022130532-appb-100004
    表示所有巡检无人机的平均移动速度，
    Figure PCTCN2022130532-appb-100005
    表示第m个巡检无人机对应前t-1个时隙内的平均水平方向移动角度，
    Figure PCTCN2022130532-appb-100006
    表示第m个巡检无人机对应前t-1个时隙内的平均竖直方向移动角度；v m(t-1)、α m(t-1)、β m(t-1)依次表示第m个巡检无人机对应前t-1个时隙的移动速度、水平移动方向、竖直移动方向，0＜λ 1＜1，λ 1表示预设用于调整巡检无人机对应前t-1个时隙移动速度对第t个时隙移动速度的影响参数，0＜λ 2＜1，λ 2表示预设用于调整巡检无人机对应前t-1个时隙水平移动方向对第t个时隙水平移动方向的影响参数，0＜λ 3＜1，λ 3表示预设用于调整巡检无人机对应前t-1个时隙竖直移动方向对第t个时隙竖直移动方向的影响参数，预设服从独立高斯分布的参数φ m表示第m个巡检无人机移动速度的随机性，预设服从独立高斯分布的参数ψ m表示第m个巡检无人机水平移动方向的随机性，预设服从独立高斯分布的参数
    Figure PCTCN2022130532-appb-100007
    表示第m个巡检无人机竖直移动方向的随机性，然后进入步骤S12；
    步骤S12.根据各时隙长度τ，分别针对各所述巡检无人机，按如下公式：
    x m(t)＝x m(t-1)+v m(t-1)cos(α m(t-1))τ
    y m(t)＝y m(t-1)+v m(t-1)sin(α m(t-1))τ
    h m(t)＝h m(t-1)+v m(t-1)sin(β m(t-1))τ
    获得第m个巡检无人机对应第t个时隙的位置坐标
    Figure PCTCN2022130532-appb-100008
    其中，x m(t)、y m(t)、h m(t)表示第m个巡检无人机对应第t个时隙分别在x、y、z坐标轴上的值，x m(t-1)、y m(t-1)、h m(t-1)表示第m个巡检无人机对应前t-1个时隙分别在x、y、z坐标轴上的值，然后进入步骤S13；
    步骤S13.根据各巡检无人机分别对应第t个时隙的移动速度、水平移动方向、竖直移动方向、位置坐标，搭建所述无人机辅助电网线路随机巡检系统，其中，所述巡检无人机只负责对所述目标电网区域中的所述电网设施、所述输电线路进行视频图像采集，所获视频图像由所述增强无人机或所述中央基站进行数据处理，然后进入步骤S2。
  3. 根据权利要求1所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S3包括如下步骤S31至步骤S33；
    步骤S31.根据所述无人机辅助电网线路随机巡检系统，结合各所述巡检无人机的质量、以及分别对应各时隙所采集的所述视频图像数据、信号发射功率，所述增强无人机的质量、对应各时隙各巡检无人机被分配的CPU计算频率、信号发射功率，以及所述中央基站的位置坐标，构建现实物理实体网络，然后进入步骤S32；
    步骤S32.根据现实物理实体网络，按如下公式：
    Figure PCTCN2022130532-appb-100009
    构建各所述巡检无人机分别对应各时隙的数字孪生模型，其中，
    Figure PCTCN2022130532-appb-100010
    表示第m个巡检无人机对应第t个时隙的数字孪生模型，
    Figure PCTCN2022130532-appb-100011
    表示第m个巡检无人机的质量，
    Figure PCTCN2022130532-appb-100012
    表示第m个巡检无人机对应第t个时隙所采集的视频图像数据，
    Figure PCTCN2022130532-appb-100013
    表示第m个巡检无人机对应第t个时隙的信号发射功率，
    Figure PCTCN2022130532-appb-100014
    表示第m个巡检无人机对应第t个时隙的位置坐标，
    Figure PCTCN2022130532-appb-100015
    表示第m个巡检无人机对应第t个时隙的最大信号发射功率；
    同时按如下公式：
    Figure PCTCN2022130532-appb-100016
    构建增强无人机对应各时隙的所述数字孪生模型，其中，DT SUAV(t)表示增强无人机对应第t个时隙的数字孪生模型，W SUAV表示增强无人机的质量，f SUAV(t)表示增强无人机对应第t个时隙被分配的CPU计算频率，P SUAV(t)表示增强无人机对应第t个时隙的信号发射功率，L SUAV(t)增强无人机对应第t个时隙的位置坐标，
    Figure PCTCN2022130532-appb-100017
    表示增强无人机对应第t个时隙的最大信号发射功率，
    Figure PCTCN2022130532-appb-100018
    表示增强无人机的最大CPU计算频率，C SUAV表示增强无人机处理1bit的数据所需要CPU的周期数；
    以及按如下公式：
    DT BS＝{L BS}
    构建所述中央基站的数字孪生模型DT BS，其中，L BS表示所述中央基站的位置坐标，然后进入步骤S33；以及
    步骤S33.基于各所述巡检无人机分别对应各时隙的所述数字孪生模型、以及所述增强无人机对应各时隙的数字孪生模型、所述中央基站的数字孪生模型，构成无人机辅助电网线路随机巡检系统的数字孪生网络，用于拟合各所述巡检无人机、所述增强无人机的位置坐标、以及所述系统的资源状态，然后进入步骤S4。
  4. 根据权利要求3所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S4包括如下步骤S41至步骤S42；
    步骤S41.根据无人机辅助电网线路随机巡检系统的数字孪生网络，构建各所述巡检无人机各时隙所采集视频图像数据分别对应各卸载方式下的总时延模型，然后进入步骤S42；以及
    步骤S42.根据各巡检无人机各时隙所采集视频图像数据分别对应各卸载方式下的总时延模型，基于电网线路随机巡检系统卸载时延、以及数据任务处理时延约束条件，构建无人机群分别对应各时隙的能耗模型或能耗均衡模型，并进一步构建无人机群分别对应各时隙的能耗均衡最小化目标函数，然后进入步骤S5。
  5. 根据权利要求4所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S41包括如下步骤S411至步骤S413；
    步骤S411.基于巡检无人机在任一时隙内仅能在所述增强无人机与所述中央基站之间择一进行视频图像数据卸载，根据各所述巡检无人机共用一个频谱与增强无人机进行通信，各所述巡检无人机分别与所述增强无人机之间对应第t个时隙的数据传输速率
    Figure PCTCN2022130532-appb-100019
    所 述增强无人机与所述中央基站之间对应第t个时隙的数据传输速率R SUAV(t)，按如下公式：
    Figure PCTCN2022130532-appb-100020
    构建各所述巡检无人机与所述增强无人机之间分别对应各时隙的通信时延模型
    Figure PCTCN2022130532-appb-100021
    表示第m个巡检无人机与增强无人机之间对应第t个时隙的通信时延，
    Figure PCTCN2022130532-appb-100022
    表示第m个巡检无人机对应第t个时隙所采集的视频图像数据；
    以及按如下公式：
    Figure PCTCN2022130532-appb-100023
    构建各所述巡检无人机分别对应各时隙所采集视频图像数据在所述增强无人机与所述中央基站之间传输的通信时延模型
    Figure PCTCN2022130532-appb-100024
    表示第m个巡检无人机对应第t个时隙所采集视频图像数据在所述增强无人机与所述中央基站之间传输的通信时延；然后进入步骤S412；
    步骤S412.基于定义
    Figure PCTCN2022130532-appb-100025
    对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到所述增强无人机进行处理，则按如下公式：
    Figure PCTCN2022130532-appb-100026
    构建所述增强无人机接收端针对第m个巡检无人机在第t个时隙所采集视频图像数据的数据处理时延模型
    Figure PCTCN2022130532-appb-100027
    其中，C SUAV表示所述增强无人机处理1bit的数据所需要CPU的周期数，f SUAV(t)表示所述增强无人机对应第t个时隙被分配的CPU计算频率；并基于所述增强无人机采用非抢占的方式针对视频图像数据按照信道功率增益降序的方式进行处理，则按如下公式：
    Figure PCTCN2022130532-appb-100028
    构建第m个巡检无人机在第t个时隙所采集视频图像数据在被增强无人机处理之前所对应的队列等待时延模型
    Figure PCTCN2022130532-appb-100029
    ρ(i)表示增强无人机依次处理第i个视频图像数据所来自巡检无人机的序号，k表示第m个巡检无人机在第t个时隙所采集视频图像数据在增强无人机等待处理的排序序号；
    进而按如下公式：
    Figure PCTCN2022130532-appb-100030
    构建第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理所对应的总时延模型T m,0(t)，然后进入步骤S413；以及
    步骤S413.基于定义
    Figure PCTCN2022130532-appb-100031
    对应第m个巡检无人机在第t个时隙所采集视频图像数据卸载到中央基站进行处理，则按如下公式：
    Figure PCTCN2022130532-appb-100032
    构建第m个巡检无人机在第t个时隙所采集视频图像数据卸载到增强无人机进行处理所对应的总时延模型T m,1(t)，然后进入步骤S42。
  6. 根据权利要求5所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述各所述巡检无人机分别与所述增强无人机之间采用NOMA的方式进行通信，所述增强无人机与所述中央基站之间采用OFDMA的方式进行通信。
  7. 根据权利要求5所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S42包括步骤S421至步骤S422；
    步骤S421.基于中央基站采用有线方式进行供电，按如下公式：
    Figure PCTCN2022130532-appb-100033
    构建无人机群对应第t个时隙的能耗模型E all(t)，其中，
    Figure PCTCN2022130532-appb-100034
    表示第m个巡检无人机在第t个时隙内的飞行能耗；
    Figure PCTCN2022130532-appb-100035
    flyE SUAV(t)表示增强无人机在第t个时隙内的飞行能耗；
    Figure PCTCN2022130532-appb-100036
    表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100037
    卸载至增强无人机处理所消耗的能量，κ SUAV表示增强无人机CPU对应的有效开关电容；
    Figure PCTCN2022130532-appb-100038
    表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100039
    在与增强无人机之间的传输能耗；
    Figure PCTCN2022130532-appb-100040
    表示数据
    Figure PCTCN2022130532-appb-100041
    在增强无人机与中央基站之间的传输能耗，然后进入步骤S422；以及
    步骤S422.根据无人机群对应第t个时隙的能耗模型E all(t)，按如下公式：
    Figure PCTCN2022130532-appb-100042
    s.t.C1:
    Figure PCTCN2022130532-appb-100043
    C2:
    Figure PCTCN2022130532-appb-100044
    C3:
    Figure PCTCN2022130532-appb-100045
    C4:
    Figure PCTCN2022130532-appb-100046
    C5:x min≤x(t)＜x max
    C6:y min≤y(t)＜y max
    C7:h min≤h(t)＜h max
    C8:
    Figure PCTCN2022130532-appb-100047
    C9:
    Figure PCTCN2022130532-appb-100048
    进一步构建无人机群分别对应各时隙的能耗最小化目标函数
    Figure PCTCN2022130532-appb-100049
    其中，C5-C7表示预设约束增强无人机的活动范围，C8表示增强无人机全双工通信的条件需求，C9表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100050
    需要在该时隙内卸载处理完毕。
  8. 根据权利要求5所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S42包括步骤S421’至步骤S422’；
    步骤S421’.基于中央基站采用有线方式进行供电，按如下公式：
    Figure PCTCN2022130532-appb-100051
    构建无人机群对应第t个时隙的能耗均衡模型
    Figure PCTCN2022130532-appb-100052
    其中，χ表示能耗均衡系数，
    Figure PCTCN2022130532-appb-100053
    表示第m个巡检无人机在第t个时隙内的飞行能耗；
    Figure PCTCN2022130532-appb-100054
    flyE SUAV(t)表示增强无人机 在第t个时隙内的飞行能耗；
    Figure PCTCN2022130532-appb-100055
    表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100056
    卸载至增强无人机处理所消耗的能量，κ SUAV表示增强无人机CPU对应的有效开关电容；
    Figure PCTCN2022130532-appb-100057
    表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100058
    在与增强无人机之间的传输能耗；
    Figure PCTCN2022130532-appb-100059
    表示数据
    Figure PCTCN2022130532-appb-100060
    在增强无人机与中央基站之间的传输能耗，然后进入步骤S422；以及
    步骤S422’.根据无人机群对应第t个时隙的能耗均衡模型
    Figure PCTCN2022130532-appb-100061
    按如下公式：
    Figure PCTCN2022130532-appb-100062
    s.t.C1:
    Figure PCTCN2022130532-appb-100063
    C2:
    Figure PCTCN2022130532-appb-100064
    C3:
    Figure PCTCN2022130532-appb-100065
    C4:
    Figure PCTCN2022130532-appb-100066
    C5:x min≤x(t)＜x max
    C6:y min≤y(t)＜y max
    C7:h min≤h(t)＜h max
    C8:
    Figure PCTCN2022130532-appb-100067
    C9:
    Figure PCTCN2022130532-appb-100068
    进一步构建无人机群分别对应各时隙的能耗均衡最小化目标函数
    Figure PCTCN2022130532-appb-100069
    其中，C5-C7表示预设约束增强无人机的活动范围，C8表示增强无人机全双工通信的条件需求，C9表示第m个巡检无人机在第t个时隙所采集视频图像数据
    Figure PCTCN2022130532-appb-100070
    需要在该时隙内卸载处理完毕。
  9. 根据权利要求7或8所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S7中，若不满足迭代溢出条件，则执行如下步骤S71至步骤S73；
    步骤S71.随机初始化第t个时隙种群
    Figure PCTCN2022130532-appb-100071
    其中，1≤i≤I，I表示第t个时隙种群K(t)中个体的个数，
    Figure PCTCN2022130532-appb-100072
    表示第t个时隙种群K(t)中增强无人机第i个位置坐标，然后进入步骤S72；
    步骤S72.分别针对第t个时隙种群K(t)中的各个个体，基于系统第t个时隙状态结合增强无人机位置坐标所对应系统第t个时隙动作空间中的系统资源分配、以及视频图像数据卸载决策方案，按如下公式：
    Figure PCTCN2022130532-appb-100073
    获得第t个时隙种群K(t)中各个个体分别对应的适应度，然后进入步骤S73；以及
    步骤S73.判断第t个时隙种群K(t)中各个个体分别所对应适应度是否均满足预设适应度阈值，是则挑选最高适应度所对应的个体，即获得该个体所对应的增强无人机位置坐标，即更新增强无人机的位置坐标，并返回步骤S6；否则基于第t个时隙种群K(t)中各个个体分别对应的适应度，对第t个时隙种群K(t)中的数据进行选择、交叉、变异操作，更新第t个时隙种群K(t)中的各个个体，并返回步骤S72。
  10. 根据权利要求9所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S73中，预设适应度阈值为预设适应度下限，若预设适应度阈值为预设适应度下限时，则判断第t个时隙种群K(t)中各个个体分别所对应适应度是否均大于预设适应度下限。
  11. 根据权利要求1所述一种用于电网线路随机巡检的无人机辅助边缘计算方法，其中：所述步骤S7中的迭代溢出条件为最大预设迭代次数，或者以当前迭代起向历史迭代方向、预设迭代次数内各迭代下无人机群对应第t个时隙能耗的方差小于预设能耗波动范围。
PCT/CN2022/130532 2022-02-25 2022-11-08 一种用于电网线路随机巡检的无人机辅助边缘计算方法 WO2023160012A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210174497.3A CN114237917B (zh) 2022-02-25 2022-02-25 一种用于电网巡检的无人机辅助边缘计算方法
CN202210174497.3 2022-02-25

Publications (1)

Publication Number Publication Date
WO2023160012A1 true WO2023160012A1 (zh) 2023-08-31

Family

ID=80748173

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/130532 WO2023160012A1 (zh) 2022-02-25 2022-11-08 一种用于电网线路随机巡检的无人机辅助边缘计算方法

Country Status (2)

Country Link
CN (1) CN114237917B (zh)
WO (1) WO2023160012A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116993183A (zh) * 2023-09-27 2023-11-03 电子科技大学中山学院 无人机辅助边缘计算中概率性计算卸载的服务优化方法
CN117022715A (zh) * 2023-10-09 2023-11-10 国网山东省电力公司沂水县供电公司 一种巡检无人机电能补给方法
CN117192288A (zh) * 2023-09-22 2023-12-08 河南蓝犀牛工业装备技术有限公司 一种智慧配电网故障定位方法及其系统
CN117354759A (zh) * 2023-12-06 2024-01-05 吉林大学 一种多无人机辅助mec的任务卸载与充电调度联合优化方法
CN117371761A (zh) * 2023-12-04 2024-01-09 集美大学 一种智慧海洋物联网任务调度方法、装置、设备及介质
CN117748747A (zh) * 2024-02-21 2024-03-22 青岛哈尔滨工程大学创新发展中心 一种auv集群能源在线监测及管理系统以及方法
CN117765451A (zh) * 2024-02-22 2024-03-26 江苏征途技术股份有限公司 基于ai智能辅控系统设备的联控分析方法及系统
CN117891613A (zh) * 2024-02-29 2024-04-16 湘江实验室 一种基于车联网的计算卸载和资源分配方法及系统
CN117915405A (zh) * 2024-03-18 2024-04-19 华东交通大学 一种分布式多无人机协同任务卸载方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114237917B (zh) * 2022-02-25 2022-06-10 南京信息工程大学 一种用于电网巡检的无人机辅助边缘计算方法
CN115866559B (zh) * 2022-11-25 2024-04-30 西安电子科技大学 一种非正交多址接入辅助的车联网低能耗安全卸载方法
CN116820778A (zh) * 2023-07-13 2023-09-29 中国电信股份有限公司技术创新中心 边缘设备资源分配方法、系统、装置、设备及介质
CN117541025B (zh) * 2024-01-05 2024-03-19 南京信息工程大学 一种用于密集输电线路巡检的边缘计算方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180129881A1 (en) * 2016-11-08 2018-05-10 Dedrone Holdings, Inc. Systems, Methods, Apparatuses, and Devices for Identifying, Tracking, and Managing Unmanned Aerial Vehicles
CN113296963A (zh) * 2021-07-27 2021-08-24 南京信息工程大学 一种考虑用户移动性的无人机辅助的边缘计算方法
CN114065963A (zh) * 2021-11-04 2022-02-18 湖北工业大学 一种电力物联网中基于深度强化学习的计算任务卸载方法
CN114237917A (zh) * 2022-02-25 2022-03-25 南京信息工程大学 一种用于电网巡检的无人机辅助边缘计算方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111864618B (zh) * 2019-04-24 2021-09-10 广州煜煊信息科技有限公司 一种用于电力系统的无人机巡检方法及系统
US20220361011A1 (en) * 2019-07-01 2022-11-10 Nokia Solutions And Networks Oy Mismatch detection in digital twin data by comparing to thresholds the reliability and latency constraint communication related values received from sources with geolocation information
CN111651649A (zh) * 2020-04-10 2020-09-11 安徽继远软件有限公司 一种输电线路及杆塔的虚拟围栏构建方法及系统
CN111524034B (zh) * 2020-05-12 2023-11-03 华北电力大学 高可靠低时延低能耗的电力巡检系统及巡检方法
CN112381963B (zh) * 2020-11-12 2022-02-22 广东电网有限责任公司 一种基于数字孪生技术的电力物联网智能巡检方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180129881A1 (en) * 2016-11-08 2018-05-10 Dedrone Holdings, Inc. Systems, Methods, Apparatuses, and Devices for Identifying, Tracking, and Managing Unmanned Aerial Vehicles
CN113296963A (zh) * 2021-07-27 2021-08-24 南京信息工程大学 一种考虑用户移动性的无人机辅助的边缘计算方法
CN114065963A (zh) * 2021-11-04 2022-02-18 湖北工业大学 一种电力物联网中基于深度强化学习的计算任务卸载方法
CN114237917A (zh) * 2022-02-25 2022-03-25 南京信息工程大学 一种用于电网巡检的无人机辅助边缘计算方法

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117192288A (zh) * 2023-09-22 2023-12-08 河南蓝犀牛工业装备技术有限公司 一种智慧配电网故障定位方法及其系统
CN116993183A (zh) * 2023-09-27 2023-11-03 电子科技大学中山学院 无人机辅助边缘计算中概率性计算卸载的服务优化方法
CN116993183B (zh) * 2023-09-27 2023-12-29 电子科技大学中山学院 无人机辅助边缘计算中概率性计算卸载的服务优化方法
CN117022715A (zh) * 2023-10-09 2023-11-10 国网山东省电力公司沂水县供电公司 一种巡检无人机电能补给方法
CN117022715B (zh) * 2023-10-09 2024-02-06 国网山东省电力公司沂水县供电公司 一种巡检无人机电能补给方法
CN117371761A (zh) * 2023-12-04 2024-01-09 集美大学 一种智慧海洋物联网任务调度方法、装置、设备及介质
CN117354759A (zh) * 2023-12-06 2024-01-05 吉林大学 一种多无人机辅助mec的任务卸载与充电调度联合优化方法
CN117354759B (zh) * 2023-12-06 2024-03-19 吉林大学 一种多无人机辅助mec的任务卸载与充电调度联合优化方法
CN117748747A (zh) * 2024-02-21 2024-03-22 青岛哈尔滨工程大学创新发展中心 一种auv集群能源在线监测及管理系统以及方法
CN117748747B (zh) * 2024-02-21 2024-05-17 青岛哈尔滨工程大学创新发展中心 一种auv集群能源在线监测及管理系统以及方法
CN117765451A (zh) * 2024-02-22 2024-03-26 江苏征途技术股份有限公司 基于ai智能辅控系统设备的联控分析方法及系统
CN117765451B (zh) * 2024-02-22 2024-04-30 江苏征途技术股份有限公司 基于ai智能辅控系统设备的联控分析方法及系统
CN117891613A (zh) * 2024-02-29 2024-04-16 湘江实验室 一种基于车联网的计算卸载和资源分配方法及系统
CN117891613B (zh) * 2024-02-29 2024-05-31 湘江实验室 一种基于车联网的计算卸载和资源分配方法及系统
CN117915405A (zh) * 2024-03-18 2024-04-19 华东交通大学 一种分布式多无人机协同任务卸载方法
CN117915405B (zh) * 2024-03-18 2024-05-31 华东交通大学 一种分布式多无人机协同任务卸载方法

Also Published As

Publication number Publication date
CN114237917B (zh) 2022-06-10
CN114237917A (zh) 2022-03-25

Similar Documents

Publication Publication Date Title
WO2023160012A1 (zh) 一种用于电网线路随机巡检的无人机辅助边缘计算方法
Yu et al. Multi-objective optimization for UAV-assisted wireless powered IoT networks based on extended DDPG algorithm
Huang et al. Deep reinforcement learning for UAV navigation through massive MIMO technique
WO2021017227A1 (zh) 无人机轨迹优化方法、装置及存储介质
CN106966298B (zh) 基于机器视觉的装配式建筑智能吊装方法与系统
Han et al. Artificial intelligence-based handoff management for dense WLANs: A deep reinforcement learning approach
Huang et al. An overview of intelligent wireless communications using deep reinforcement learning
Li et al. Deep reinforcement learning-based mobility-aware robust proactive resource allocation in heterogeneous networks
Dai et al. Mobile crowdsensing for data freshness: A deep reinforcement learning approach
CN112965507B (zh) 一种基于智能优化的集群无人机协同工作系统及方法
WO2022242468A1 (zh) 任务卸载方法、调度优化方法和装置、电子设备及存储介质
Cui et al. Multiagent reinforcement learning-based cooperative multitype task offloading strategy for internet of vehicles in B5G/6G network
CN117580105B (zh) 一种面向电网巡检的无人机任务卸载优化方法
CN113676357B (zh) 面向电力物联网中边缘数据处理的决策方法及其应用
Liu et al. A decoupled learning strategy for MEC-enabled wireless virtual reality (VR) network
Jiang et al. MARS: A DRL-based Multi-task Resource Scheduling Framework for UAV with IRS-assisted Mobile Edge Computing System
Lu et al. A drl-based decentralized computation offloading method: An example of an intelligent manufacturing scenario
CN116704823B (zh) 基于强化学习的无人机智能轨迹规划和通感资源分配方法
CN116939866A (zh) 一种基于协同计算和资源分配联合优化的无线联邦学习效率提升方法
Zeng et al. Periodic Collaboration and Real-Time Dispatch Using an Actor–Critic Framework for UAV Movement in Mobile Edge Computing
CN115361453B (zh) 一种面向边缘服务网络的负载公平卸载与迁移方法
Martini et al. Challenges and perspectives in neuromorphic-based visual IoT systems and networks
Si et al. UAV-assisted Semantic Communication with Hybrid Action Reinforcement Learning
Xu et al. Joint optimization task offloading and trajectory control for unmanned-aerial-vehicle-assisted mobile edge computing
CN117541025B (zh) 一种用于密集输电线路巡检的边缘计算方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22928274

Country of ref document: EP

Kind code of ref document: A1