A power grid inspection unmanned aerial vehicle high-time-efficiency unloading and trajectory planning method

Abstract

A method for efficient unloading and trajectory planning of a power grid inspection drone includes the following steps: Step S1: Establish a communication transmission model between the drone and the high-altitude platform, an energy consumption and replenishment model for the drone, and a task processing time model for the drone; Step S2: Define the optimization objective as minimizing the total inspection completion time, and then determine the resource allocation strategy, inspection trajectory decision, and task unloading decision as variables to be optimized. Combined with the model in Step S1, construct a joint optimization model with the objective of minimizing the total inspection completion time, and impose constraints such as task delay, resource capacity, energy feasibility, and inspection sequence; Step S3: Decompose and solve the joint optimization problem. First, use a task-aware genetic algorithm to determine the optimal drone inspection path sequence; then, based on this path, use the TD3 deep reinforcement learning algorithm to dynamically decide the task unloading ratio and computational resource allocation strategy for each inspection point.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to a collaborative optimization method for drone trajectory planning and task unloading in smart grid inspection. Background Technology

[0002] With the continuous advancement of smart grid construction, using drones for power facility inspection has become an important means of ensuring the safe operation of the power grid. Traditional drone inspections mainly rely on manual remote control or pre-set simple waypoints. The high-definition images or video data acquired during inspections usually need to be brought back to the ground station for offline processing after the mission. However, with the increasing requirements for inspection accuracy, the amount of data generated by airborne sensors has exploded, and this traditional mode can no longer meet the real-time and intelligent needs of power inspection.

[0003] Existing drone inspection technologies mainly face challenges related to limited onboard computing resources and energy consumption. Drones are typically battery-powered, and the processing power of onboard computing platforms is limited. Performing complex image recognition and defect detection tasks locally on the drone not only consumes a large amount of computing power and shortens the drone's flight time, but may also lead to excessively high task processing latency, failing to meet the needs of real-time early warning. While mobile edge computing technology offers a solution to these problems by offloading some computational tasks to edge servers to alleviate the computational burden on drones, existing offloading strategies are often quite simplistic. For example, the translated paper "Deep Reinforcement Learning-Based Computation Offloading for Mobile Edge Computing in 6G" describes a complete offloading strategy that requires ground equipment to offload all computational tasks to the drone for processing. The drone must complete all computations independently and cannot rely on the computing power of external edge servers for collaborative processing. Another example is the translated paper "Hierarchical Aerial Computing for Internet of Things via Cooperation of HAPs and UAVs," which describes a binary offloading strategy based on high-altitude platform and drone collaboration. Although this strategy introduces HAPs as external computing nodes, its task model is still a complete offloading. Each computational task from an IoT device can only be decided as a whole: either all computation is performed on the drone or all offloaded to the high-altitude platform, without the ability to perform fine-grained partial offloading of individual tasks. Offloading all tasks is limited by communication bandwidth and transmission energy consumption; processing only locally is limited by computational performance. Finding the optimal balance between computational latency, transmission energy consumption, and flight energy consumption is a problem that current technologies have not yet fully solved.

[0004] Furthermore, existing intelligent inspection solutions, such as the translation of "Dynamic Energy-Efficient Computation Offloading in NOMA-Enabled Air–Ground-Integrated Edge Computing," describe a dynamic task offloading and resource allocation method in a three-layer architecture of ground-UAV-high-altitude platform. This method effectively reduces the total system energy consumption through stochastic optimization techniques, but it only treats task offloading and resource allocation as decision variables, completely ignoring the UAV's inspection trajectory planning and failing to achieve coordinated optimization of flight path and computation offloading. Another example is the translation of "Intelligent Ubiquitous Network Accessibility for Wireless-Powered MEC in UAV-Assisted B5G," which describes an intelligent charging-offloading scheme. This scheme effectively reduces system service latency by jointly optimizing the task offloading decision, connection scheduling, and UAV charging and computation resource allocation of ground equipment; however, it only considers the UAV's hovering position as a preset fixed point and does not treat the UAV's active flight trajectory or access sequence as optimization variables, thus failing to achieve joint optimization of energy consumption and latency for a specific inspection path. The flight trajectory planning and computational task offloading strategies for UAVs are typically designed independently—that is, the path is planned first, followed by computation, or vice versa. However, the UAV's flight position directly affects the quality of the communication channel between it and the edge server, thus impacting the task offloading rate and energy consumption; conversely, task offloading decisions also affect the UAV's hovering time and energy consumption, thereby constraining trajectory selection. The lack of joint and coordinated optimization of trajectory, computation, and communication makes it difficult to achieve optimal overall system inspection efficiency.

[0005] Meanwhile, the power line inspection environment is complex and ever-changing, with varying urgency and data volume across different inspection tasks. Existing static or semi-static scheduling algorithms struggle to dynamically adapt to environmental changes and efficiently utilize limited onboard energy and computing resources while meeting task deadline constraints. At the mathematical modeling level, the joint optimization problem of UAV inspection is typically modeled as a mixed-integer nonlinear programming problem. These problems are characterized by high dimensionality and non-convexity, resulting in extremely high solution complexity. Traditional convex optimization algorithms or heuristic algorithms often face limitations such as high computational cost, slow convergence speed, and difficulty adapting to dynamic, time-varying environments when dealing with such strongly coupled dynamic optimization problems. This often leads to optimization results getting trapped in local optima, failing to achieve globally optimal resource scheduling in complex power line inspection scenarios.

[0006] Especially for UAV inspection scenarios assisted by high-altitude platforms, while these platforms offer wide coverage, the channel gain of UAVs varies significantly across different flight altitudes and orientations. Simply relying on fixed strategies cannot fully utilize moments with favorable channel conditions for data transmission. Furthermore, when performing inspection tasks, UAVs must consider not only the processing time of the current task but also whether the remaining battery power can support the completion of all subsequent tasks. This contradiction between long-term resource allocation and short-term task scheduling further increases the difficulty of optimization. Therefore, there is an urgent need for an intelligent decision-making mechanism that can comprehensively consider dynamic channel state information, energy consumption models, and task priorities to achieve deep coupling optimization of UAV trajectory and computational offloading. Summary of the Invention

[0007] The purpose of this invention is to design an efficient and reasonable collaborative optimization mechanism for UAV inspection trajectory planning and task unloading, in order to solve the technical problems in the existing technology where UAV inspection trajectory planning and computational task unloading strategies are designed independently and lack joint collaborative optimization, making it difficult to find the optimal balance between computational latency, transmission energy consumption and flight energy consumption, and existing static or semi-static scheduling algorithms are unable to dynamically adapt to environmental changes and efficiently utilize limited airborne energy and computing resources while meeting task deadline constraints. Therefore, this invention proposes a high-time-efficiency unloading and trajectory planning method for power grid inspection UAVs.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for efficient unloading and trajectory planning of a power grid inspection drone includes the following steps: Step S1: Establish a communication transmission model between the UAV and the high-altitude platform, a model of the UAV's energy consumption and replenishment, and a model of the UAV's mission processing time; Step S2: Define the optimization objective as minimizing the total inspection completion time. Then, determine the resource allocation strategy, inspection trajectory decision, and task unloading decision as variables to be optimized. Combine the model from Step S1 to construct a joint optimization model with the objective of minimizing the total inspection completion time, and impose constraints such as task latency, resource capacity, energy feasibility, and inspection order. Step S3: Decompose and solve the joint optimization problem. First, use a task-aware genetic algorithm to determine the optimal UAV inspection path sequence. Then, based on the path, dynamically decide the task unloading ratio and computing resource allocation strategy for each inspection point.

[0009] Step S1 includes the following steps: Step 1-1: Describe the system architecture of edge-cloud collaborative drone inspection, and determine the set of inspection points and the set of charging stations and their spatial relationship; Steps 1-2: Establish a communication transmission model between the UAV and the high-altitude platform; Steps 1-3: Establish the UAV's own task processing and energy consumption and replenishment model, as well as the task processing time model.

[0010] In step 1-1, the attributes of the drone, high-altitude platform, charging base station, and inspection point in the system are represented as follows: 1-1-1) Define the system as including a drone. A high-altitude platform containing a cloud monitoring center And a set of charging stations deployed around the inspection area to recharge the drones. Meanwhile, a set of inspection points is set up within the inspection area. Wherein, the set of inspection points is represented as ,in This represents the k-th inspection point, where k is the total number of inspection points; the set of charging stations is represented as... ,in Let m represent the m-th charging station, where m is the total number of charging stations; define the set of time slots for task execution as... ,in This refers to the nth discrete time slot, where N refers to the total number of time slots for task execution. Define inspection points The attribute is ,in Here are the coordinates of the inspection point. This represents the size of the task data generated by the drone at this inspection point in the nth time slot. Assign time constraints to the task; and define the inspection access sequence as follows: ,in Indicates the first k The indexes of the accessed inspection, and , representing the start and end points of the drone, respectively; Define drones The attribute is ,in For the drone's coordinates, For the drone's remaining energy, Provides onboard computing resources for the drone; the drone starts from a preset starting point within the mission cycle. Take off and ascend to the preset operating altitude Perform inspection tasks and return to the preset endpoint after completing all tasks. It also stipulates that all inspection tasks must be completed no later than [date missing]. Finish; 1-1-2) The channel gain between the UAV and the high-altitude platform is calculated; where This represents the probability of a line-of-sight link, used to characterize the probability of line-of-sight propagation. This refers to the additional attenuation coefficient of non-line-of-sight links; This refers to the reference channel gain at a distance of 1 meter; This refers to the Euclidean distance between the drone and the high-altitude platform; 1-1-3) indicates that the drone's charging decision is a binary variable. ,Right now Each drone will choose the nearest charging station to recharge its energy via wireless charging. Based on the attributes of the drone, high-altitude platform, charging base station, and inspection point described in step 1-1, a drone inspection system process is established, in which the drone performs inspections in sequence; then, when stopping at each inspection point, tasks are assigned, and some tasks are offloaded to the high-altitude platform for calculation; finally, the drone determines whether its own battery needs to be replenished. This framework effectively alleviates the computational burden during drone inspections.

[0011] In steps 1-2, the communication transmission model between the UAV and the high-altitude platform is as follows: 1-2-1) Using the calculated The communication rate between the UAV and the high-altitude platform at inspection point k is obtained. ,in Channel bandwidth allocated to high-altitude platforms For the drone's transmission power, For noise power, For the channel gain between the UAV and the high-altitude platform; calculate The time required to transmit the drone mission to the high-altitude platform; This indicates the proportion of drone unloading tasks; 1-2-2) Define C as the CPU cycles required to process a unit of data size, and calculate... Obtain the number of CPU cycles required for the high-altitude platform to execute the task; calculate The computation time of the high-altitude platform is obtained, and Substitute The total mission transmission time between the UAV and the high-altitude platform was obtained; among which... This refers to the computing resources of the high-altitude platform; 1-2-3) Calculation The total transmission energy consumption of the drone is obtained, where the definition is... This refers to the drone's launch power.

[0012] In steps 1-3, the drone needs to handle its own tasks and replenish its energy. The drone's energy consumption, energy replenishment, and task processing time model are as follows: Step 1-3-1) The energy consumption of the UAV for processing tasks consists of hovering energy consumption and computing energy consumption. The hovering energy consumption of the drone was obtained, among which This refers to the hovering power of the drone. This refers to the hovering time of the drone; calculation The energy consumption of the UAV computing task is obtained, among which This refers to an effective capacitor switch. This refers to the computing resources allocated to the drone for this mission. This refers to the time consumed by the drone; Step 1-3-2) Define the charging threshold for the drone as follows: Therefore, when the drone's battery level falls below this threshold, the drone will trigger a charging event at the current inspection point, and the calculation will be performed. The energy replenished for the drone at the current inspection point, of which The receiving power of the drone already includes the transmitting power and energy conversion efficiency of the charging station. Charging time for the drone; Step 1-3-3) Calculation The energy consumption of the drone flying from the previous inspection point to the current inspection point is obtained, among which... It is a power function that gives the total power required to maintain flight at a speed of V. For the flight time of the drone; (Steps 1-3-4) Therefore, it is possible to calculate the total energy consumption of the drone in the current time interval. ,in Energy consumption for drone hovering. This represents the total transmission energy consumption of the drone; (Steps 1-3-5) Simultaneously, the remaining energy of the drone before it departs for the next inspection point can be calculated. ,in This is the maximum battery capacity of the drone. The remaining energy of the drone when it leaves the previous inspection point. This represents the total energy consumption of the drone during the current time interval. Steps 1-3-6) Calculation The total time the drone spends processing inspection tasks within that time slot is obtained, including the drone's flight time. The time required for drones to process tasks And drone charging time ,in , This refers to the total mission transmission time between the drone and the high-altitude platform.

[0013] In step S2, the objective function of the constructed joint optimization model is: In the formula, The set of computing resource decisions allocated to drones for handling tasks. For the inspection trajectory decision set of drones, The decision set of the HAP is used to offload part of the mission for the drone.

[0014] The constraints on the objective function in the constructed joint optimization model are as follows: 1) The processing time for each task must not exceed the maximum tolerable time inherent to the task itself; 2) The computing resources allocated to a drone for a mission cannot exceed the computing resources it possesses; 3) The remaining battery power of the drone must not exceed the maximum battery capacity, nor fall below the battery power threshold; 4) The inspection sequence must be followed; 5) The remaining battery power of the drone before it departs from any inspection point must be sufficient to support the energy consumption requirements of subsequent missions.

[0015] Step S3 includes the following sub-steps: In step 3-1, the overall optimization objective is divided into two sub-problems: first, optimizing the inspection sequence based on global information; and second, determining the inspection sequence and then having the UAV inspect in order, followed by calculating the unloading ratio and resource allocation at each point. In step 3-2, a task-aware genetic algorithm is used to solve the inspection path optimization sub-problem to generate a globally optimized UAV inspection sequence. In step 3-3, after the inspection sequence is determined, a deep reinforcement learning algorithm based on TD3 is used to dynamically decide the task unloading ratio and the computing resource allocation strategy of the edge server at each inspection point for the task unloading and computing resource allocation sub-problem.

[0016] The specific implementation of the genetic algorithm and the TD3 algorithm includes the following steps: 3-4-1) First, initialize the power grid inspection area information, task point coordinates, and task urgency weights; during the genetic algorithm optimization process, construct a fitness function that includes flight distance cost and task urgency, and iteratively update the particle state through selection, crossover, and mutation operations to search for and obtain the optimal UAV inspection path sequence. This is used to determine the order in which the drone visits each mission point and its flight path; 3-4-2) Determining the inspection route Then, the hovering process of the UAV at each task point is modeled as a Markov decision process; the UAV, as an intelligent agent, outputs the optimal action strategy based on the current system state using the TD3 algorithm.

[0017] The action strategy specifically includes: task unloading ratio and the drone's local computing frequency The TD3 algorithm evaluates action value through a dual-commenter network and a target policy smoothing mechanism. It is trained with the objective function of minimizing the total inspection completion time of the UAV. That is, under the premise of meeting the task deadline constraint, the algorithm minimizes the total time cost for the UAV to complete all inspection tasks by dynamically adjusting the unloading decision and computing resources.

[0018] Compared with the prior art, the present invention has the following technical effects: 1) Existing technologies typically design UAV flight trajectory planning and computational task offloading strategies independently, i.e., planning the path first and then considering computation, or vice versa. This lack of in-depth exploration of the coupling relationship between the two leads to difficulties in achieving optimal overall system inspection efficiency. This invention proposes a high-efficiency offloading and trajectory planning method for power grid inspection UAVs. By constructing a two-stage TGTD optimization framework, trajectory planning and resource scheduling are deeply coupled, effectively resolving the contradiction between flight position affecting communication channel quality and task decision-making constraining hovering time, significantly reducing the overall system task processing time.

[0019] 2) This invention fully considers the time-varying characteristics of the channel environment and the heterogeneity of task data in smart grid inspection scenarios, and adopts a reinforcement learning algorithm based on TD3 for dynamic resource scheduling. This mechanism can adaptively adjust the task offloading ratio and UAV computing resource allocation while meeting strict task deadline constraints, avoiding the poor adaptability of traditional static strategies in complex electromagnetic environments or sudden tasks. It achieves an optimal balance between computing energy consumption, transmission energy consumption, and flight energy consumption, extending the effective operating endurance of the UAV.

[0020] 3) Addressing the high-complexity characteristics of the UAV inspection joint optimization problem—characterized by its high dimensionality, non-convexity, and mixed-integer nonlinear programming (MINLP) nature—this invention decomposes the complex original problem into two sub-problems: trajectory planning and resource allocation. Task-aware genetic algorithms (TGP2) and TD3 are designed to solve these sub-problems respectively. Experimental results show that this phased solution strategy, compared to traditional convex optimization algorithms or single heuristic algorithms, significantly reduces computational complexity, accelerates convergence, and more effectively utilizes channel gain in wide-area coverage scenarios assisted by high-altitude platforms, demonstrating strong global optimization capabilities and practical engineering value. Attached Figure Description

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram of the model of the present invention; Figure 3 This is the convergence graph of the algorithm in this invention; Figure 4 This is a line graph showing the total inspection completion time of the UAV at different inspection points in this invention. Figure 5 This is a line graph showing the total inspection completion time of the UAV at different UAV speeds in this invention. Figure 6 This is a line graph showing the total inspection completion time of drones under different maximum battery capacities in this invention. Figure 7 This is a line graph showing the total inspection completion time of the UAV under different inspection point task sizes in this invention. Detailed Implementation

[0022] like Figure 1 As shown, a method for high-efficiency unloading and trajectory planning of a power grid inspection drone includes the following steps: Step S1: Establish a communication transmission model between the UAV and the high-altitude platform, a model of the UAV's energy consumption and replenishment, and a model of the UAV's mission processing time; Step S2: Define the optimization objective as minimizing the total inspection completion time, then determine the resource allocation strategy, inspection trajectory decision and task unloading decision as variables to be optimized. Combine the model in Step S1 to construct a joint optimization model with the objective of minimizing the total inspection completion time, and impose constraints such as task delay, resource capacity, energy feasibility and inspection order. Step S3: Decompose and solve the joint optimization problem. First, use a task-aware genetic algorithm to determine the optimal UAV inspection path sequence. Then, based on the path, use the TD3 deep reinforcement learning algorithm to dynamically decide the task unloading ratio and computing resource allocation strategy for each inspection point.

[0023] Step S1 includes the following steps: Step 1-1: Describe the system architecture of edge-cloud collaborative drone inspection, and determine the set of inspection points and the set of charging stations and their spatial relationship; Steps 1-2: Establish a communication transmission model between the UAV and the high-altitude platform; Steps 1-3: Establish the UAV's own task processing and energy consumption and replenishment model, as well as the task processing time model.

[0024] In step 1-1, the attributes of the drone, high-altitude platform, charging base station, and inspection point in the system are represented as follows: 1-1-1) Define the system as including a drone. A high-altitude platform containing a cloud monitoring center And a set of charging stations deployed around the inspection area to recharge the drones. Meanwhile, a set of inspection points is set up within the inspection area. Wherein, the set of inspection points is represented as ,in This represents the k-th inspection point, where k is the total number of inspection points; the set of charging stations is represented as... ,in Let m represent the m-th charging station, where m is the total number of charging stations; define the set of time slots for task execution as... ,in It refers to the nth discrete time slot, and N refers to the total number of time slots for task execution.

[0025] Define inspection points The attribute is ,in Here are the coordinates of the inspection point. This represents the size of the task data generated by the drone at this inspection point in the nth time slot. This is a task time constraint. The inspection access sequence is also defined as follows: ,in Indicates the first k The indexes of the accessed inspection, and , representing the start and end points of the drone, respectively.

[0026] Define drones The attribute is ,in For the drone's coordinates, For the drone's remaining energy, Provides onboard computing resources for the drone; the drone starts from a preset starting point within the mission cycle. Take off and ascend to the preset operating altitude Perform inspection tasks and return to the preset endpoint after completing all tasks. It also stipulates that all inspection tasks must be completed no later than [date missing]. Finish; 1-1-2) The channel gain between the UAV and the high-altitude platform is calculated; where This represents the probability of a line-of-sight link, used to characterize the probability of line-of-sight propagation. This refers to the additional attenuation coefficient of non-line-of-sight links; This refers to the reference channel gain at a distance of 1 meter; This refers to the Euclidean distance between the drone and the high-altitude platform; 1-1-3) indicates that the drone's charging decision is a binary variable. ,Right now Each drone will choose the nearest charging station to recharge its energy via wireless charging. Based on the attributes of the drone, high-altitude platform, charging base station, and inspection point described in step 1-1, a drone inspection system process is established, in which the drone performs inspections in sequence; then, when stopping at each inspection point, tasks are assigned, and some tasks are offloaded to the high-altitude platform for calculation; finally, the drone determines whether its own battery needs to be replenished. This framework effectively alleviates the computational burden during drone inspections.

[0027] In steps 1-2, the communication transmission model between the UAV and the high-altitude platform is as follows: 1-2-1) Using the calculated The communication rate between the UAV and the high-altitude platform at inspection point k can be obtained. ,in Channel bandwidth allocated to high-altitude platforms For the drone's transmission power, For noise power, For the channel gain between the UAV and the high-altitude platform; calculate The time required to transmit the drone mission to the high-altitude platform; This indicates the proportion of drone unloading tasks; 1-2-2) Define C as the CPU cycles required to process a unit of data size, and calculate... Obtain the number of CPU cycles required for the high-altitude platform to execute the task; calculate The computation time of the high-altitude platform is obtained, and Substitute The total mission transmission time between the UAV and the high-altitude platform was obtained; among which... This refers to the computing resources of the high-altitude platform; 1-2-3) Calculation The total transmission energy consumption of the drone is obtained, where the definition is... This refers to the drone's launch power.

[0028] Drones need to handle their own tasks and recharge their energy. The energy consumption, energy replenishment, and task processing time model for drones are as follows: Step 1-3-1) The energy consumption of the UAV for processing tasks consists of hovering energy consumption and computing energy consumption. The hovering energy consumption of the drone was obtained, among which This refers to the hovering power of the drone. This refers to the hovering time of the drone; calculation The energy consumption of the UAV computing task is obtained, among which This refers to an effective capacitor switch. This refers to the computing resources allocated to the drone for this mission. This refers to the time consumed by the drone; Step 1-3-2) Define the charging threshold for the drone as follows: Therefore, when the drone's battery level falls below this threshold, the drone will trigger a charging event at the current inspection point, and the calculation will be performed. The energy replenished for the drone at the current inspection point, of which The receiving power of the drone already includes the transmitting power and energy conversion efficiency of the charging station. Charging time for the drone; Step 1-3-3) Calculation The energy consumption of the drone flying from the previous inspection point to the current inspection point is obtained, among which... It is a power function that gives the total power required to maintain flight at a speed of V. For the flight time of the drone; (Steps 1-3-4) Therefore, it is possible to calculate the total energy consumption of the drone in the current time interval. ,in Energy consumption for drone hovering. This represents the total transmission energy consumption of the drone; (Steps 1-3-5) Simultaneously, the remaining energy of the drone before it departs for the next inspection point can be calculated. ,in This is the maximum battery capacity of the drone. The remaining energy of the drone when it leaves the previous inspection point. This represents the total energy consumption of the drone during the current time interval. Steps 1-3-6) Calculation The total time the drone spends processing inspection tasks within that time slot is obtained, including the drone's flight time. The time required for drones to process tasks And drone charging time ,in , This refers to the total mission transmission time between the drone and the high-altitude platform.

[0029] In step S2, the objective function of the constructed joint optimization model is: In the formula, The set of computing resource decisions allocated to drones for handling tasks. For the inspection trajectory decision set of drones, The decision set of the HAP is used to offload part of the mission for the drone.

[0030] The constraints on the objective function in the constructed joint optimization model are as follows: 1) The processing time for each task must not exceed the maximum tolerable time inherent to the task itself; 2) The computing resources allocated to a drone for a mission cannot exceed the computing resources it possesses; 3) The remaining battery power of the drone must not exceed the maximum battery capacity, nor fall below the battery power threshold; 4) The inspection sequence must be followed; 5) The remaining battery power of the drone before it departs from any inspection point must be sufficient to support the energy consumption requirements of subsequent missions.

[0031] Step S3 includes: In step 3-1, the overall optimization objective is divided into two sub-problems: first, optimizing the inspection sequence based on global information; and second, after determining the inspection sequence, having the UAV inspect in order, and then allocating the unloading ratio and computational resources at each point.

[0032] In step 3-2, a task-aware genetic algorithm is used to solve the inspection path optimization sub-problem to generate a globally optimized UAV inspection sequence.

[0033] In step 3-3, after the inspection sequence is determined, a deep reinforcement learning algorithm based on TD3 is used to dynamically decide the task unloading ratio and the computing resource allocation strategy of the edge server at each inspection point for the task unloading and computing resource allocation sub-problem.

[0034] The specific implementation of the genetic algorithm and the TD3 algorithm includes the following steps: 3-4-1) First, initialize the power grid inspection area information, task point coordinates, and task urgency weights; during the genetic algorithm optimization process, construct a fitness function that includes flight distance cost and task urgency, and iteratively update the particle state through selection, crossover, and mutation operations to search for and obtain the optimal UAV inspection path sequence. This is used to determine the order in which the drone visits each mission point and its flight path.

[0035] 3-4-2) Determining the inspection route Then, the hovering process of the UAV at each task point is modeled as a Markov decision process; the UAV, as an intelligent agent, outputs the optimal action strategy based on the current system state using the TD3 algorithm; the action strategy specifically includes: task offloading ratio. and the drone's local computing frequency The TD3 algorithm evaluates action value through a dual-commenter network and a target policy smoothing mechanism. It is trained with the objective function of minimizing the total inspection completion time of the UAV. That is, under the premise of meeting the task deadline constraint, the algorithm minimizes the total time cost for the UAV to complete all inspection tasks by dynamically adjusting the unloading decision and computing resources.

[0036] Example: Figure 1 This is a schematic diagram of a high-efficiency unloading and trajectory planning method for power grid inspection drones provided in an embodiment of the present invention.

[0037] Figure 2 A system model diagram is presented, describing an edge-cloud collaborative system for smart grid inspection. In this system, drones obtain energy replenishment via ground charging stations and offload part of the inspection task to a high-altitude platform for computation via an aerial link. The model optimizes the drone access sequence, task offloading ratio, and resource allocation to balance computational load and energy consumption, ensuring efficient and stable continuous inspection operations within task deadlines and energy consumption constraints.

[0038] Figure 3 The convergence of different comparison algorithms is described. Although the cumulative reward obtained by the agent fluctuates in different training rounds, the final reward value remains stable in the positive range and continues to increase, which proves the feasibility and reward guarantee capability of the proposed TD3 algorithm in joint optimization of task unloading and trajectory. In addition, the reward convergence curve of TD3 tends to be closer to the theoretical optimal upper bound, indicating that the proposed collaborative optimization mechanism can significantly reduce the total inspection time of UAVs while ensuring the reliable completion of UAV inspection tasks.

[0039] Table 1. Total UAV inspection completion time for each algorithm scheme under different numbers of inspection points;

[0040] Figure 4Table 1 shows the comparison results of the proposed TGTD method with six baseline algorithms (GADD, GTD, FUCC, FHCC, DOGT, and RTGTD) in the total inspection completion time in the smart grid drone inspection scenario, as the number of inspection points increases from 8 to 12. As can be seen from the specific data in Table 1, the total time of all algorithms increases with the expansion of the task scale, but TGTD consistently achieves the lowest value at each test point. For example, it takes only 513.28 seconds with 10 inspection points, which is significantly better than the suboptimal solution GTD's 526.62 seconds and other methods. Figure 4 The trend curve further confirms this advantage, with a gentler growth slope, demonstrating stronger scalability. In contrast, FHCC, due to uploading all tasks to the high-altitude platform, experiences a surge in communication overhead with the data volume, resulting in the worst performance, reaching 804.9 seconds at 12 points. FUCC, limited by onboard computing power, suffers from significant processing latency. While other heuristic or partially optimized methods alleviate resource bottlenecks to some extent, their overall efficiency remains inferior to TGTD due to the lack of deep collaboration between trajectory planning and task offloading. In summary, Figure 4 Together with Table 1, this verifies the effectiveness and superiority of the proposed two-stage collaborative optimization framework in reducing the total system completion time and improving inspection efficiency.

[0041] Table 2. The impact of UAV cruise speed on the total inspection completion time of UAVs under various algorithms;

[0042] Figure 5 Table 2 shows the comparison of the TGTD method with baseline algorithms such as GADD, GTD, FUCC, FHCC, DOGT, and RTGTD in the total inspection completion time in the smart grid drone inspection scenario when the cruise speed increases from 10 m / s to 30 m / s. As the speed increases, the total time of all schemes decreases, but TGTD is always the best, with a time of 513.28 seconds at 20 m / s and decreasing to 473.28 seconds at 30 m / s. Figure 5 The proposed method exhibits a smooth and stable downward trend, demonstrating good adaptability. In contrast, FHCC suffers from the worst performance (855.34 seconds at 10 m / s) due to the high communication overhead of full data upload, while FUCC lags significantly behind due to limitations in onboard computing power. Although other methods show some improvement with increasing speed, their efficiency is still lower than TGTD due to the lack of coordinated optimization of trajectory and unloading. The results show that the proposed method can effectively utilize speed gain, balancing flight efficiency and mission processing, and has stronger robustness.

[0043] Table 3. The impact of the maximum battery capacity of the UAV on the total inspection completion time of the UAV under each algorithm;

[0044] Figure 6 Table 3 shows the performance comparison of the TGTD method with baseline algorithms such as GADD, GTD, FUCC, FHCC, DOGT, and RTGTD in the total inspection completion time in the smart grid drone inspection scenario when the maximum battery capacity of the drone increases from 10 units to 30 units. As shown in Table 3, the total task completion time of all schemes decreases significantly with the increase of battery capacity. This is mainly attributed to the fact that the drone can reduce the number of charging times or shorten the charging time. Among them, TGTD consistently performs best under all capacity settings, for example, 513.28 seconds at 20 units capacity and further reduced to 462.25 seconds at 30 units. In contrast, FHCC is more sensitive to energy due to high communication energy consumption caused by full battery unloading, reaching as high as 767.75 seconds at low capacity (10 units), which is the worst performance. Although FUCC has local computing capabilities, its efficiency improvement is limited by computing power. While other methods have been optimized with the improvement of energy supply, they are still inferior to TGTD overall because they do not consider the trajectory and unloading strategy in a coordinated manner. Figure 6 The trend indicates that the proposed method can utilize available energy resources more efficiently and exhibits stronger adaptability and robustness under different battery capacity constraints.

[0045] Table 4. Impact of inspection task size on total UAV inspection completion time under each algorithm;

[0046] Figure 7 Table 4 shows the performance comparison of the TGTD method with baseline algorithms such as GADD, GTD, FUCC, FHCC, DOGT, and RTGTD in the total inspection completion time in the smart grid UAV inspection scenario, as the task data volume of a single inspection point increases from 100 units to 250 units. As shown in Table 4, the total time of all schemes increases significantly with the increase of task size, but TGTD always performs best, with a time of 545.5 seconds when the task size is 200, which is significantly lower than the second-best scheme GTD's 567.36 seconds. In contrast, FHCC, due to its full offload strategy, experiences a surge in communication transmission overhead with the data volume, reaching as high as 845.1 seconds when the task volume is 250, resulting in the worst performance. Although FUCC processes tasks locally, its processing latency increases rapidly due to limitations in onboard computing power. Other methods have some optimization, but their efficiency improvement is limited due to the lack of dynamic adaptability to changes in task size. Figure 7 The trend indicates that TGTD can more effectively cope with different task loads by co-optimizing trajectories and dynamically adjusting the offloading ratio, and shows stronger scalability and robustness under various task scales.

Claims

1. A method for high-efficiency unloading and trajectory planning of a power grid inspection drone, characterized in that, Includes the following steps: Step S1: Establish a communication transmission model between the UAV and the high-altitude platform, a model of the UAV's energy consumption and replenishment, and a model of the UAV's mission processing time; Step S2: Define the optimization objective as minimizing the total inspection completion time. Then, determine the resource allocation strategy, inspection trajectory decision, and task unloading decision as variables to be optimized. Combine the model from Step S1 to construct a joint optimization model with the objective of minimizing the total inspection completion time, and impose constraints such as task latency, resource capacity, energy feasibility, and inspection order. Step S3: Decompose and solve the joint optimization problem. First, use a task-aware genetic algorithm to determine the optimal UAV inspection path sequence. Then, based on the path, dynamically decide the task unloading ratio and computing resource allocation strategy for each inspection point.

2. The method according to claim 1, characterized in that, Step S1 includes the following steps: Step 1-1: Describe the system architecture of edge-cloud collaborative drone inspection, and determine the set of inspection points and the set of charging stations and their spatial relationship; Steps 1-2: Establish a communication transmission model between the UAV and the high-altitude platform; Steps 1-3: Establish the UAV's own task processing and energy consumption and replenishment model, as well as the task processing time model.

3. The method according to claim 2, characterized in that, In step 1-1, the attributes of the drone, high-altitude platform, charging base station, and inspection point in the system are represented as follows: 1-1-1) Define the system as including a drone. A high-altitude platform containing a cloud monitoring center And a set of charging stations deployed around the inspection area to recharge the drones. Meanwhile, a set of inspection points is set up within the inspection area. The set of inspection points is represented as follows: ,in This represents the k-th inspection point, where k is the total number of inspection points; the set of charging stations is represented as... ,in Let m represent the m-th charging station, where m is the total number of charging stations; define the set of time slots for task execution as... ,in This refers to the nth discrete time slot, where N refers to the total number of time slots for task execution. Define inspection points The attribute is ,in Here are the coordinates of the inspection point. This represents the size of the task data generated by the drone at this inspection point in the nth time slot. Assign time constraints to the task; and define the inspection access sequence as follows: ,in Indicates the first k The indexes of the accessed inspection, and , representing the start and end points of the drone, respectively; Define drones The attribute is ,in For the drone's coordinates, For the drone's remaining energy, Provides onboard computing resources for the drone; the drone starts from a preset starting point within the mission cycle. Take off and ascend to the preset operating altitude Perform inspection tasks and return to the preset destination after completing all tasks. It also stipulates that all inspection tasks must be completed no later than [date missing]. Finish; 1-1-2) The channel gain between the UAV and the high-altitude platform is calculated; where This represents the probability of a line-of-sight link, used to characterize the probability of line-of-sight propagation. This refers to the additional attenuation coefficient of non-line-of-sight links; This refers to the reference channel gain at a distance of 1 meter; This refers to the Euclidean distance between the drone and the high-altitude platform; 1-1-3) indicates that the drone's charging decision is a binary variable. ,Right now Each drone will choose the nearest charging station to recharge its energy via wireless charging. Based on the attributes of the drone, high-altitude platform, charging base station, and inspection point in step 1-1, a drone inspection system process is established, in which the drone performs inspections in sequence; then, when stopping at each inspection point, tasks are assigned, and some tasks are offloaded to the high-altitude platform for calculation; finally, the drone will determine whether its own battery needs to be replenished. This framework effectively alleviates the computational burden during drone inspections.

4. The method according to claim 2, characterized in that, In steps 1-2, the communication transmission model between the UAV and the high-altitude platform is as follows: 1-2-1) Using the calculated The communication rate between the UAV and the high-altitude platform at inspection point k is obtained. ,in Channel bandwidth allocated to high-altitude platforms For the drone's transmission power, For noise power, For the channel gain between the UAV and the high-altitude platform; calculate The time required to transmit the drone mission to the high-altitude platform; This indicates the proportion of drone unloading tasks; 1-2-2) Define C as the CPU cycles required to process a unit of data size, and calculate... Obtain the number of CPU cycles required for the high-altitude platform to execute the task; calculate The computation time of the high-altitude platform is obtained, and Substitute Obtain the total mission transmission time between the drone and the high-altitude platform; in This refers to the computing resources of the high-altitude platform; 1-2-3) Calculation The total transmission energy consumption of the drone is obtained, where the definition is... This refers to the drone's launch power.

5. The method according to claim 2, characterized in that, In steps 1-3, the drone needs to handle its own tasks and replenish its energy. The drone's energy consumption, energy replenishment, and task processing time model are as follows: Step 1-3-1) The energy consumption of the UAV for processing tasks consists of hovering energy consumption and computing energy consumption. The hovering energy consumption of the drone was obtained, among which This refers to the hovering power of the drone. This refers to the hovering time of the drone; calculation The energy consumption of the UAV computing task is obtained, among which This refers to an effective capacitor switch. This refers to the computing resources allocated to the drone for this mission. This refers to the time consumed by the drone; Step 1-3-2) Define the charging threshold for the drone as follows: Therefore, when the drone's battery level falls below this threshold, the drone will trigger a charging event at the current inspection point, and the calculation will be performed. The energy replenished for the drone at the current inspection point, of which The receiving power of the drone already includes the transmitting power and energy conversion efficiency of the charging station. Charging time for the drone; Step 1-3-3) Calculation The energy consumption of the drone flying from the previous inspection point to the current inspection point is obtained, among which... It is a power function that gives the total power required to maintain flight at a speed of V. For the flight time of the drone; (Steps 1-3-4) Therefore, it is possible to calculate the total energy consumption of the drone in the current time interval. ,in Energy consumption for drone hovering. This represents the total transmission energy consumption of the drone; (Steps 1-3-5) Simultaneously, the remaining energy of the drone before it departs for the next inspection point can be calculated. ,in The maximum battery capacity of the drone, The remaining energy of the drone when it leaves the previous inspection point. This represents the total energy consumption of the drone during the current time interval. Steps 1-3-6) Calculation The total time the drone spends processing inspection tasks within that time slot is obtained, including the drone's flight time. The time required for drones to process tasks And drone charging time ,in , This refers to the total mission transmission time between the drone and the high-altitude platform.

6. The method according to claim 1, characterized in that, In step S2, the objective function of the constructed joint optimization model is: In the formula, The set of computing resource decisions allocated to drones for handling tasks. For the inspection trajectory decision set of drones, The decision set of the HAP is used to offload part of the mission for the drone.

7. The method according to claim 6, characterized in that, The constraints on the objective function in the constructed joint optimization model are as follows: 1) The processing time for each task must not exceed the maximum tolerable time inherent to the task itself; 2) The computing resources allocated to a drone for a mission cannot exceed the computing resources it possesses; 3) The remaining battery power of the drone must not exceed the maximum battery capacity, nor fall below the battery power threshold; 4) The inspection sequence must be followed; 5) The remaining battery power of the drone before it departs from any inspection point must be sufficient to support the energy consumption requirements of subsequent missions.

8. The method according to claim 7, characterized in that, Step S3 includes the following sub-steps: In step 3-1, the overall optimization objective is divided into two sub-problems: first, optimizing the inspection sequence based on global information; and second, determining the inspection sequence and then having the UAV inspect in order, followed by calculating the unloading ratio and resource allocation at each point. In step 3-2, a task-aware genetic algorithm is used to solve the sub-problem of inspection path optimization in order to generate a globally optimized UAV inspection sequence. In step 3-3, after the inspection sequence is determined, a deep reinforcement learning algorithm based on TD3 is used to dynamically decide the task unloading ratio and the computing resource allocation strategy of the edge server at each inspection point for the sub-problem of task unloading and computing resource allocation.

9. The method according to claim 8, characterized in that: The specific implementation of the genetic algorithm and the TD3 algorithm includes the following steps: 3-4-1) First, initialize the power grid inspection area information, task point coordinates, and task urgency weights; during the genetic algorithm optimization process, construct a fitness function that includes flight distance cost and task urgency, and iteratively update the particle state through selection, crossover, and mutation operations to search for and obtain the optimal UAV inspection path sequence. This is used to determine the order in which the drone visits each mission point and its flight path; 3-4-2) Determining the inspection route Then, the hovering process of the UAV at each task point is modeled as a Markov decision process; the UAV, as an intelligent agent, outputs the optimal action strategy based on the current system state using the TD3 algorithm.

10. The method according to claim 9, characterized in that: The action strategy specifically includes: task unloading ratio. and the drone's local computing frequency The TD3 algorithm evaluates action value through a dual-commenter network and a target policy smoothing mechanism. It is trained with the objective function of minimizing the total inspection completion time of the UAV. That is, under the premise of meeting the task deadline constraint, the algorithm minimizes the total time cost for the UAV to complete all inspection tasks by dynamically adjusting the unloading decision and computing resources.

Images