A method and system for taking over a patrol task of a UAV cluster and a medium

Abstract

The application discloses a kind of unmanned aerial vehicle cluster's inspection task takeover method, system and medium, belong to unmanned aerial vehicle inspection technical field, method includes: obtaining each unmanned aerial vehicle battery remaining power and current flight power consumption, calculate remaining endurance time and determine low power unmanned aerial vehicle;Determine candidate unmanned aerial vehicle;According to shared airspace map, calculate the reachable path and estimated range cost of candidate unmanned aerial vehicle;Calculate the spatiotemporal similarity of low power unmanned aerial vehicle inspection task sequence and each candidate unmanned aerial vehicle current task sequence;According to the estimated range cost and spatiotemporal similarity of each candidate unmanned aerial vehicle, determine the task correlation degree and screen takeover unmanned aerial vehicle;According to the real-time constraint information of inspection task sequence and takeover unmanned aerial vehicle, construct multi-objective optimization model, and solve to obtain the cooperative flight path of takeover unmanned aerial vehicle, and control takeover unmanned aerial vehicle to execute the inspection task according to cooperative flight path, by implementing the application, the reliability of unmanned aerial vehicle task takeover continuation can be improved.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) inspection technology, and in particular to a method, system and medium for taking over inspection tasks of a UAV swarm. Background Technology

[0002] Transmission line inspection involves periodic, fault-related, and special checks and inspections of overhead transmission lines, towers, insulators, hardware, conductors, ground wires, and the environment along the line corridor. This is a core maintenance task to understand the equipment's operating status, promptly identify defects and potential external damage risks, and provide a basis for line maintenance and safe, stable operation. Unmanned aerial vehicle (UAV) swarms, with their multi-UAV distributed collaborative operation capabilities, have become an effective technical mode for performing large-scale transmission line corridor and tower inspection tasks. In long-distance, continuous inspection operations, the limited battery capacity of individual UAVs is a practical constraint affecting the overall stability of the swarm's task execution. Due to the long inspection cycle, once a single UAV's energy is depleted, the swarm system needs a collaborative takeover function to ensure that unfinished inspection tasks can be executed promptly and continuously.

[0003] Existing methods for taking over inspection tasks typically assign unfinished tasks from a single UAV to another UAV that is currently idle or closest to the task area. This method faces inherent difficulties in practical applications: relying solely on a single spatial or idle metric for takeover decisions fails to comprehensively consider multi-dimensional constraints such as energy reserves, spatiotemporal matching of task sequences, and cluster load balancing. This leads to interruptions in the takeover process when the selected UAV is tasked with new tasks due to insufficient energy or spatiotemporal conflicts in the task sequence, making it difficult to guarantee the continuity and overall efficiency of cluster inspection operations. Therefore, how to solve the problem of task interruption caused by the limited endurance of a single UAV during power line inspection by UAV clusters, thereby improving the reliability of UAV task takeover and continuation, is a pressing technical problem that needs to be addressed. Summary of the Invention

[0004] This application provides a method, system, and medium for taking over inspection tasks in a drone swarm, thereby improving the reliability of drone task takeover and continued inspection.

[0005] This application provides a method for taking over inspection tasks in a drone swarm, including: Obtain the remaining battery power and current flight power consumption of each drone in the drone cluster; calculate the remaining flight time of each drone based on the remaining battery power and the current flight power consumption, and identify the low-battery drone from the drone cluster based on the remaining flight time; Identify several candidate drones; Obtain a shared airspace map of the drone cluster; based on the shared airspace map, calculate the feasible path for each candidate drone to fly from its current location to the mission area of ​​the low-power drone, and obtain the estimated flight cost corresponding to each feasible path; Obtain the first inspection task sequence of the low-power UAV and the second inspection task sequence of each candidate UAV, and calculate the spatiotemporal similarity between each second inspection task sequence and the first inspection task sequence; Based on the estimated range cost and spatiotemporal similarity of each candidate drone, the task correlation between each candidate drone and the low-power drone is determined, and the takeover drone is selected from the candidate drones based on the task correlation. The system acquires real-time constraint information of the takeover drone, constructs a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information, solves the multi-objective optimization model to obtain the cooperative flight path of the takeover drone, and controls the takeover drone to perform the follow-up inspection task according to the cooperative flight path.

[0006] This application determines the trigger for task takeover by obtaining the remaining flight time of each UAV and identifying low-battery UAVs; it provides an initial set of candidates for subsequent screening by marking the remaining UAVs in the cluster as candidates; it quantifies the resource consumption of each candidate in performing the takeover task in space by calculating the feasible path and estimated range cost of each candidate UAV to the low-battery UAV's task area; it quantifies the matching degree of each candidate in the takeover task in the time and space dimensions by calculating the spatiotemporal similarity between the current task sequence of each candidate UAV and the unfinished task sequence of the low-battery UAV; it selects the takeover UAV from the candidate UAVs by combining the estimated range cost and spatiotemporal similarity of each candidate, so that the selected takeover UAV has a relative advantage in both spatial accessibility and task matching degree; it constructs and solves a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information of the takeover UAV, and optimizes the overall task completion degree and task interruption time under the constraints of flight time and load to obtain an executable cooperative flight path, and controls the takeover UAV to continue inspection according to the path, so that the cluster can still complete the original inspection task after the power of a single UAV is exhausted, reducing the task interruption and redundant energy consumption caused by the exit of a single UAV. Compared with existing technologies, this application generates task correlation degree through parallel analysis of spatial reachability and spatiotemporal similarity of task sequences when no offline preset scheme is available, so that the takeover selection has a clear quantitative basis. Furthermore, by constructing a multi-objective optimization model, it simultaneously optimizes task completion and interruption time under the dual constraints of endurance and load, enabling temporary takeover to have a globally quantifiable decision-making capability.

[0007] Further, the step of calculating feasible paths for each candidate UAV to fly from its current location to the low-power UAV's mission area based on the shared airspace map, and obtaining the estimated range cost corresponding to each feasible path, includes: Based on the obstacle distribution density in the shared airspace map, the shared airspace map is spatially divided to generate a spatial unit map containing several spatial units; Based on the current position of each candidate drone, a path search is performed in the spatial unit map, starting from the current position and ending at the mission area of ​​the low-power drone, until a preset convergence condition is met, resulting in several candidate feasible paths. The path search includes: constructing a search tree with the current position as the root node; during each expansion, selecting an expansion node to be expanded from the existing nodes in the search tree, expanding the expansion node to adjacent spatial units to obtain an expanded flight segment, and obtaining the path cost corresponding to the expanded flight segment; when the search tree expands to the low-power UAV mission area, recording the current search path as the candidate feasible path, and continuing to expand the search until the preset convergence condition is met; Based on the path cost, the estimated travel cost of each candidate feasible path is determined; the candidate feasible path with the smallest estimated travel cost is selected and determined as the feasible path.

[0008] By non-uniformly dividing the space according to the distribution density of obstacles in the shared airspace map, fine-grained cells are retained in densely obstacle areas to characterize the boundaries of feasible passages, while large cells are used to cover free areas to reduce computational load, thereby reducing the scale of spatial search while ensuring the accuracy of environmental characterization. By performing tree-like path search between the current position of each candidate UAV and the target area, the discovery process of feasible passages has a backtrackable exploration structure. By advancing only to adjacent spatial cells in each expansion and recording the path cost corresponding to the current expansion segment, the problem of accumulating the total cost of the entire path is decomposed into a sequential acquisition problem of single-step costs. By recording a path when the search tree first expands to the target area and continuing to expand until convergence, the problem of stopping when a feasible passage is first found and losing better paths is avoided, so that the final set of candidate paths covers multiple feasible solutions with different costs. By selecting the path with the minimum estimated range cost from multiple candidate paths as the feasible path, the cost of this path comes from the horizontal comparison of multiple candidate paths under the same search framework, rather than the isolated estimation of a single path, providing a comparable range consumption basis for subsequent UAV selection.

[0009] Further, calculating the spatiotemporal similarity between each of the second inspection task sequences and the first inspection task sequence includes: Extract the subsequences within the overlapping interval of the second inspection task sequence and the first inspection task sequence on the time axis to form a time-aligned double sequence pair; Using the smallest time window as the unit, slide the scan from the starting position of the double sequence pair, and calculate the spatial deviation metric of the double sequence pair within the current window in turn; The spatial deviation metric is compared with a preset threshold, and the double sequence pair is determined to be a spatiotemporally overlapping subsequence or a spatiotemporally non-overlapping subsequence in the minimum time window based on the comparison result. The spatiotemporal similarity is determined based on the first total duration of the spatiotemporally overlapping subsequences and the second total duration of the spatiotemporally non-overlapping subsequences.

[0010] By extracting the overlapping intervals of two sequences on the time axis, the comparison range is limited to the intervals where temporal overlap is possible, avoiding invalid calculations on non-overlapping parts. By sliding the scan from the starting position in units of the smallest time window, the continuous inspection task sequence is discretized into basic units that can be compared segment by segment, enabling the spatial matching relationship between sequences to be compared segment by segment at the same temporal granularity. By calculating the spatial deviation metric of the two sub-sequences within the current window and comparing it with a preset threshold, the continuous deviation degree of the two sequences in spatiotemporal position is transformed into a discrete overlap or non-overlap determination, making the qualitative division have a quantifiable basis for judgment. By determining the similarity based on the total duration of the spatiotemporally overlapping sub-sequences and the non-overlapping sub-sequences, the binary judgment results of each window are summarized into a global numerical similarity index with time weights. This index can be directly used for subsequent horizontal comparison and ranking between candidate UAVs.

[0011] Furthermore, solving the multi-objective optimization model to obtain the cooperative flight path of the takeover UAV includes: The first task to be taken over in the first inspection task sequence, as well as the current position, remaining power, and task capacity limit of each takeover drone, are input into the multi-objective optimization model. Solve the multi-objective optimization model to output a set of schemes including multiple candidate task allocation schemes; wherein, each candidate task allocation scheme includes the allocation relationship between the second task to be taken over and each of the takeover drones, and the task order in which each of the takeover drones executes the second task to be taken over; The optimal candidate task allocation scheme is selected from the set of schemes, and the cooperative flight path corresponding to each of the takeover UAVs is determined according to the optimal candidate task allocation scheme.

[0012] By inputting the spatial location and temporal attributes of the tasks to be taken over, the current location of each takeover drone, its remaining battery power, and the upper limit of the task capacity into a multi-objective optimization model, the allocation and sequencing of tasks are correlated with the real-time endurance and task capacity of each drone before the solution is started, providing calculable boundary conditions for subsequent scheme search. By solving the model and outputting a scheme set containing multiple candidate task allocation schemes, each scheme carries a complete allocation relationship and execution order, the model can generate multiple feasible allocation structures that meet the hard conditions under the dual constraints of endurance and load, rather than outputting only a single solution, providing a comparison object for subsequent selection under multi-objective trade-offs. By selecting the optimal scheme from the scheme set and determining the cooperative flight path accordingly, the task sequences of each takeover drone have a comparable basis for evaluation under the premise of meeting endurance and load constraints, and the abstract allocation scheme is transformed into a flight trajectory that each drone can actually execute.

[0013] Further, the step of selecting the optimal candidate task allocation scheme from the scheme set includes: Calculate the individual utility value of each takeover drone under each of the candidate task allocation schemes; Based on the individual utility value of each of the aforementioned takeover drones, a comprehensive evaluation index for each of the aforementioned candidate task allocation schemes is determined; The candidate task allocation scheme with the largest value of the comprehensive evaluation index is selected as the optimal candidate task allocation scheme.

[0014] By calculating the individual utility value of each takeover drone under each candidate scheme, the merits of each candidate task allocation scheme are quantified at the level of each individual drone, providing a comparable data basis for comparing the schemes on the dimension of "whether the benefits of each participant are balanced". By determining the comprehensive evaluation index of each candidate scheme based on the individual utility value of each drone, the multi-dimensional individual utility is aggregated into a single value at the scheme level, enabling the candidate schemes to be ranked horizontally on the same scale. By selecting the candidate scheme with the largest comprehensive evaluation index as the optimal scheme, the final selected scheme is superior to the unselected schemes in both overall task completion and the balance of benefits among drones, rather than sacrificing the task execution efficiency of some drones for the sole preference of overall task completion.

[0015] Further, the step of constructing a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information includes: The allocation relationship and task order are used as decision variables in the multi-objective optimization model. Based on the total number or total value of the third task to be taken over that has been successfully taken over corresponding to the decision variables, the first optimization objective is to maximize the total number or the total value. Based on the total delay of the third task to be taken over corresponding to the decision variable, the second optimization objective is to minimize the total delay. The constraints of the multi-objective optimization model are determined based on the remaining battery power of each of the takeover drones and the upper limit of the mission capacity. The multi-objective optimization model is constructed based on the decision variables, the first optimization objective, the second optimization objective, and the constraints.

[0016] By setting the allocation relationship and task order as decision variables, the model is given an adjustable degree of freedom during the solution process. By maximizing the total number or value of successfully taken-over third tasks as the primary optimization objective, the solver prioritizes accepting more tasks when searching for feasible allocation schemes, rather than actively discarding low-priority tasks during iterations to pursue other metrics. By minimizing the sum of the delays between the actual start time and the planned start time of each third task as the secondary optimization objective, the solver simultaneously evaluates the cumulative time offset of each task when selecting an allocation scheme, avoiding the pursuit of higher performance metrics at the expense of higher performance. Distributing a large number of tasks to the same drone due to the large number of tasks undertaken causes the overall sequence to shift backward. By determining constraints based on the remaining battery power and task capacity limit of each drone taking over, all candidate solutions are within the physically feasible boundary before execution, and no solution is discarded due to insufficient battery power or overload after the solution is solved. By combining decision variables, two optimization objectives and constraints into a unified optimization model, the solver can simultaneously handle the requirements of maintaining the number of tasks, maintaining time continuity and satisfying hard constraints in one optimization process, rather than making decisions in stages and item by item, which would lead to local optimal solutions damaging overall feasibility.

[0017] Further, the step of determining a matching score between each candidate drone and the low-power drone based on the estimated range cost and spatiotemporal similarity of each candidate drone, and selecting a takeover drone from the candidate drones based on the matching score, includes: Based on the estimated range cost of each candidate UAV and the corresponding spatiotemporal similarity, the task correlation degree corresponding to each candidate UAV is determined. Based on the task relevance, a matching score is determined for each of the candidate drones; The matching scores of each candidate drone are sorted, and the candidate drone with the highest matching score is selected as the takeover drone.

[0018] By integrating the estimated range cost with spatiotemporal similarity into a mission relevance score, the two heterogeneous dimensions of "spatial arrival cost" and "mission sequence matching degree" are normalized and compared within the same framework, providing a quantifiable initial basis for subsequent screening. By determining the matching degree score based on mission relevance, the constraints of additional range cost and remaining endurance safety margin are further introduced on the basis of mission relevance, so that the matching degree score covers both takeover suitability and physical feasibility, avoiding the selection of candidates that are spatially reachable but have insufficient power to support mission execution or have excessively high cost of route adjustment. By ranking the matching degree scores and selecting the highest one, the comparison of the merits of multiple candidate UAVs is completed on a unified scale. The ranking process itself is the execution path for screening decisions, without the need to introduce additional experience thresholds or manual intervention.

[0019] Further, the step of calculating the remaining flight time of each drone based on the remaining battery power and the current flight power consumption, and determining the low-battery drone based on the remaining flight time, includes: Obtain the ambient temperature and real-time wind speed of each of the aforementioned drones; The remaining battery power is adjusted based on the ambient temperature to obtain the adjusted usable capacity of each drone. The current flight power consumption is corrected based on the real-time wind speed to obtain the corrected actual power consumption of each UAV. The remaining flight time of each of the drones is calculated based on the ratio of the corrected available capacity to the corrected actual power consumption. Drones with remaining battery life below a preset threshold are identified as low-battery drones.

[0020] By acquiring the ambient temperature and real-time wind speed of each drone, external environmental data is provided for correcting the remaining battery power and current flight power consumption, ensuring that subsequent range calculations reflect real-world operating conditions. Correcting the remaining battery power based on ambient temperature yields a corrected usable capacity, incorporating the temperature-dependent characteristics of battery capacity into the calculation and avoiding the overestimation of range in low-temperature scenarios using nominal capacity at room temperature. Correcting the current flight power consumption based on real-time wind speed yields a corrected actual power consumption, incorporating the impact of wind resistance on flight energy consumption and avoiding the underestimation of energy consumption in headwind scenarios using power consumption values ​​under windless conditions. The remaining flight time for each drone is calculated based on the ratio of corrected usable capacity to corrected actual power consumption, ensuring that the calculation considers both the effects of temperature on usable capacity and wind speed on flight power consumption. Drones with remaining flight time below a preset threshold are identified as low-battery drones, ensuring that the low-battery determination is driven by the remaining flight time corrected for ambient temperature and real-time wind speed, rather than by a single battery percentage.

[0021] Another embodiment of this application provides a drone swarm inspection task takeover system, which is used to execute the drone swarm inspection task takeover method as described in this application.

[0022] Another embodiment of this application provides a computer-readable storage medium, including: a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the drone swarm inspection task takeover method as described in this application. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating one embodiment of the drone swarm inspection task takeover method provided in this application; Figure 2 This is a schematic diagram of the cooperative flight path solving process according to one embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0029] In long-endurance inspection operations of power lines, oil and gas pipelines, and bridges, drone swarms collaboratively execute task sequences covering multiple geographically dispersed locations in a distributed manner. Due to the limited battery capacity of individual drones, fluctuations in wind resistance, temperature, and load during flight make it difficult to accurately predict remaining flight time, frequently resulting in drones running out of power mid-mission to complete the remaining tasks. To address this, this application proposes a method for taking over inspection tasks from a drone swarm, aiming to solve the problem of task continuity failure caused by individual drones running out of power.

[0030] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the drone swarm inspection task takeover method provided in this application. To address the problem of task continuity failure due to individual drone battery depletion in a drone swarm and to improve the reliability of swarm task takeover, this application provides an embodiment of a drone swarm inspection task takeover method, which includes steps S1 to S6, each step as follows: Step S1: Obtain the remaining battery power and current flight power consumption of each drone in the drone cluster; calculate the remaining flight time of each drone based on the remaining battery power and current flight power consumption, and identify the drone with low battery power from the drone cluster based on the remaining flight time. The remaining battery capacity refers to the current available capacity or state of charge (SOC) read from the drone's Battery Management System (BMS). Current flight power consumption refers to the total power consumed by the drone in its current flight state, including basic power consumption to maintain flight, power consumption of onboard sensor payloads, and power consumption of communication and computing units. Remaining flight time is obtained by dividing the current available battery capacity by the current actual power consumption. A low-battery drone is one whose remaining flight time is below a preset threshold. For example, the preset threshold could be less than 10 minutes of remaining flight time or an SOC below 10%.

[0031] Step S2: Identify several candidate drones; Candidate drones are those in the drone swarm excluding drones with low battery power. The selection of candidate drones can be done by marking all other drones within or outside the swarm as candidates, or by further filtering based on geographical location or mission status to reduce subsequent computational overhead.

[0032] Step S3: Obtain the shared airspace map of the drone cluster; based on the shared airspace map, calculate the feasible path for each candidate drone to fly from its current location to the low-power drone mission area, and obtain the estimated flight cost corresponding to each feasible path. The shared airspace map includes information on the distribution of obstacles within the inspection area, no-fly zone boundaries, real-time locations of other drones, and markers for inspected areas. Feasible paths are flight routes that allow candidate drones to safely reach the low-battery drone's mission area from their current location, while avoiding obstacles and no-fly zones. Estimated range cost refers to the energy consumed or distance traveled by the candidate drone along this path, used to measure the resource consumption of the candidate drone in performing takeover operations in space.

[0033] Step S4: Obtain the first inspection task sequence of the low-power drone and the second inspection task sequence of each candidate drone, and calculate the spatiotemporal similarity between each second inspection task sequence and the first inspection task sequence. The first inspection task sequence consists of a set of inspection tasks that the low-power drone has not yet completed, including the spatial location of the task (the tower number to be inspected, the coordinates of the inspection point), the planned execution time, and the task attributes (such as tower inspection, insulator inspection, hardware inspection, etc.). The second inspection task sequence consists of a set of tasks that each candidate drone is currently executing or has not yet completed, containing the same spatiotemporal and attribute information. Spatiotemporal similarity is a numerical index that quantifies the degree of overlap between the two task sequences on the time axis and in spatial location. The higher the degree of overlap, the less the candidate drone needs to adjust its original route when taking over the task from the low-power drone, and the smoother the task transition.

[0034] Step S5: Based on the estimated range cost and spatiotemporal similarity of each candidate drone, determine the mission correlation between each candidate drone and the low-power drone, and select the takeover drone from the candidate drones based on the mission correlation. Among them, task relevance is a quantitative indicator that comprehensively measures whether a candidate UAV is suitable to take over the mission of a low-battery UAV. This indicator is obtained by weighted fusion of two factors: estimated range cost and spatiotemporal similarity. The higher the task relevance, the more likely the candidate UAV can reach the mission area of ​​the low-battery UAV with lower range consumption and take over the remaining mission with less adjustment of its own mission. The takeover UAV is the candidate UAV with the best task relevance, which takes over the remaining inspection mission of the low-battery UAV.

[0035] Step S6: Obtain the real-time constraint information of the takeover drone, construct a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information, solve the multi-objective optimization model to obtain the cooperative flight path of the takeover drone, and control the takeover drone to perform the follow-up inspection task according to the cooperative flight path.

[0036] The real-time constraint information includes the current location, remaining battery power, and task capacity limit of the takeover drone. The task capacity limit refers to the maximum number of inspection tasks or task points a single drone can execute in a single sortie, given its endurance and payload. The multi-objective optimization model is a mathematical model that simultaneously considers multiple optimization objectives. Its objectives include maximizing overall task completion (i.e., successfully taking over as many tasks from the original low-battery drones as possible) and minimizing task interruption time (i.e., the sum of delays between the actual start time and the originally planned time for each task). The model's constraints include the endurance of each takeover drone (the estimated total energy consumption for executing the assigned tasks does not exceed its current remaining battery power) and task load (the number of assigned tasks does not exceed its task capacity limit). Solving this model yields the optimal task allocation scheme for each takeover drone under the constraints. The cooperative flight path is a continuous flight trajectory planned for each takeover drone based on this optimal allocation scheme, integrating its original task area and the newly taken-over task area. The takeover drone executes the follow-up inspection tasks according to this trajectory.

[0037] In some embodiments, obtaining the remaining battery power and current flight power consumption of each drone in the drone cluster in step S1 is specifically achieved in the following way: when the drone cluster is performing an inspection task, it collects a multimodal dataset containing image data, sensor data and battery status information in real time. The battery parameters of each drone are parsed and extracted from the multimodal dataset. The battery parameters include the remaining battery percentage, voltage and current. At the same time, the total power consumption data of each drone in the current flight state is obtained from the multimodal dataset. The total power consumption data includes the basic consumption for maintaining flight, payload operation and communication.

[0038] In some embodiments, step S1 calculates the remaining flight time of each drone based on the remaining battery power and the current flight power consumption, and determines the low-battery drone from the drone cluster based on the remaining flight time, including steps S101 to S105, each step as follows: Step S101: Obtain the ambient temperature and real-time wind speed of each drone; Specifically, each drone in the drone swarm collects real-time ambient temperature and wind speed data using onboard sensors. Ambient temperature is obtained through onboard temperature sensors, and real-time wind speed is measured through onboard anemometers or obtained from meteorological data sources via the swarm communication link.

[0039] Step S102: Correct the remaining battery power based on the ambient temperature to obtain the corrected usable capacity of each drone; Specifically, the controller acquires the remaining battery power and corresponding ambient temperature for each drone. Based on the ambient temperature, it determines a temperature decay coefficient, which is positively correlated with ambient temperature; that is, the lower the ambient temperature, the smaller the temperature decay coefficient. Multiplying the remaining battery power by this temperature decay coefficient yields the corrected usable capacity. By correcting the remaining battery power based on ambient temperature, the impact of low temperatures on reduced battery activity and decreased usable capacity on the remaining range estimation is eliminated.

[0040] Step S103: Correct the current flight power consumption based on the real-time wind speed to obtain the corrected actual power consumption of each UAV. Specifically, the controller acquires the current flight power consumption and corresponding real-time wind speed of each drone, as well as the drone's current heading information. The controller calculates additional wind resistance power consumption based on the relationship between real-time wind speed, wind direction, and the drone's current heading: positive additional wind resistance power consumption occurs when the real-time wind direction is opposite to the drone's heading (headwind), with higher headwind speeds resulting in higher additional wind resistance power consumption; negative or zero additional wind resistance power consumption occurs when the real-time wind direction is the same as the drone's heading (tailwind). The current flight power consumption is added to the additional wind resistance power consumption to obtain the corrected actual power consumption.

[0041] Step S104: Calculate the remaining flight time of each drone based on the ratio of the corrected available capacity to the corrected actual power consumption. Specifically, the controller divides the corrected available capacity obtained in step S102 by the corrected actual power consumption obtained in step S103 to obtain an estimate of the remaining flight time of each UAV.

[0042] Step S105: Determine drones with remaining battery life below a preset threshold as low-battery drones.

[0043] Specifically, the controller obtains a preset low battery threshold, compares the remaining flight time of each drone with the preset threshold one by one, and marks drones with remaining flight time lower than the preset threshold as low battery drones. The preset threshold is set in advance according to actual operational needs, such as remaining battery level below 10% or flight time less than 10 minutes.

[0044] In some implementations, step S2 marks the remaining drones in the drone swarm as candidate drones and sends a pre-takeover command. Upon receiving the pre-takeover command, the candidate drones send a data request via the swarm communication network to the swarm data center or the current task leader, including their own ID and the low-power drone's ID. The swarm communication network employs a hybrid structure based on a self-organizing wireless mesh network, enabling point-to-point direct communication between drones. It also uses dynamically designated cluster head nodes or fixed ground control stations as logical centers to coordinate and synchronize critical global information. After the data center or leader verifies the request, it packages the shared airspace map synchronized with the low-power drone and the incomplete inspection task sequence into a data packet and sends it to each candidate drone via the swarm communication network.

[0045] In some embodiments, step S3 calculates feasible paths for each candidate UAV to fly from its current location to the low-power UAV mission area based on the shared airspace map, and obtains the estimated range cost corresponding to each feasible path, including steps S301 to S304, each step as follows: Step S301: Based on the obstacle distribution density in the shared airspace map, the shared airspace map is spatially divided to generate a spatial unit map containing several spatial units; Specifically, after receiving the pre-takeover command, the candidate drone obtains a shared airspace map synchronized with the low-power drone through the cluster communication network. The shared airspace map includes real-time updated no-fly zones, obstacle locations, other drone locations, and information on inspected flight segments.

[0046] Based on the acquired shared airspace map, the 3D space is divided into a hierarchical structure composed of cubic units of different granularities. The granularity of the division is dynamically determined by the obstacle distribution density: fine-grained division is performed in areas with dense obstacles to accurately characterize the boundaries of narrow passages; coarse-grained division is performed in free space areas to reduce computational complexity. Each cubic unit is treated as a spatial unit and connected to all its spatially adjacent units (including face-adjacent, edge-adjacent, and point-adjacent units), forming a multi-resolution spatial unit map composed of units of different granularities. An independent Gaussian process model is associated with the connecting edge between every two adjacent spatial units. This model is used to characterize the uncertainty of the travel cost when traveling along the edge. Each Gaussian process is defined by a mean function and a covariance function. The mean function is initialized with a heuristic cost based on Euclidean distance, and the covariance function uses a squared exponential kernel. The expression for the squared exponential kernel function is: , in, K Represents the covariance function. x a , x bThe input feature vector specifically represents the features of the edge, such as the three-dimensional coordinates of the edge's start and end points, the average obstacle density of the spatial cell containing the edge, the cell granularity level, etc. For signal variance, L This is the length scaling matrix, which determines the smoothness of the function's variation in the input space or the rate of correlation decay. If... L A diagonal matrix, then l d Indicates in d Length scale on each feature dimension l A larger value indicates a smoother change in that dimension, maintaining a high correlation even when the two points are far apart. In multi-resolution spatial cell maps, different length scales can be set for different granularity regions or different directions, such as the vertical or horizontal direction. It is the squared Mahalanobis distance after length scaling. This indicates that the length scale matrix is ​​inverted and then squared. This term measures the effective distance between two points in the feature space after anisotropic scaling. To observe the noise variance, Let Kronecker function be used when i = j (That is, when calculating the covariance of the same point itself), =1, otherwise 0.

[0047] Step S302: Based on the current position of each candidate UAV, perform path search in the spatial unit map with the current position as the starting point and the low-battery UAV mission area as the ending point until the preset convergence condition is met, and obtain several candidate feasible paths.

[0048] Specifically, each candidate UAV, starting from its current position and ending at the spatial cell corresponding to the low-power UAV's mission area, performs a path search in the spatial cell map generated in step S301. The goal of the search is to find one or more paths composed of connected spatial cells from the starting point to the ending point. Each path corresponds to a feasible flight path from the candidate UAV's current position to the low-power UAV's mission area. The path search continues until a preset convergence condition is met, resulting in several candidate feasible paths.

[0049] Step S303, Path Search includes: Constructing a search tree with the current position as the root node; Selecting an expansion node to be expanded from the existing nodes in the search tree during each expansion, expanding the expansion node to adjacent spatial units to obtain an expanded flight segment, and obtaining the path cost corresponding to the expanded flight segment; When the search tree expands to the low-power UAV mission area, the current search path is recorded as a candidate feasible path, and the search continues to expand until the preset convergence condition is met. Specifically, the path search employs an iterative search method based on a search tree. First, the current position of the candidate UAV is used as the root node, and a search tree is initialized on the multi-resolution spatial cell map. For each node in the tree, an access value estimate is maintained, which represents the expected value of the estimated remaining cost from the current node to the target area.

[0050] During each expansion, the node with the highest upper confidence interval for the access value estimate is selected from the existing nodes of the search tree as the node to be expanded. The calculation of the upper confidence interval uses a preset exploration coefficient to balance utilization and exploration. A larger coefficient value indicates a greater tendency to explore regions with high uncertainty, while a smaller value indicates a greater tendency to utilize known low-cost regions. The exploration coefficient can be adjusted within the range of [0.1, 2.0] according to the problem size and can be optimized by performing a grid search in a standard simulation environment, using the convergence speed and stability of the final path cost as indicators.

[0051] Starting from the expanded node, the optimal expanded edge is selected from all its connected edges that minimizes the uncertainty of the overall path cost from the root node to the target area. The expanded edge extends to adjacent spatial units, resulting in an expanded flight segment. Simultaneously, the Gaussian process model associated with this edge is invoked to evaluate the observed flight cost generated by this virtual expansion, and this observed value is used as the path cost corresponding to the expanded flight segment. Subsequently, the Gaussian process model parameters are updated using this path cost observation, and the updated cost information is backtracked along the search tree from the newly expanded node towards the root node, updating the probability distribution of the cumulative path cost from the root node to the new node, and simultaneously updating the access value estimates of all relevant nodes on the backtracked path.

[0052] The Gaussian process model is trained online and incrementally, with training data derived from virtual expansion observations during the algorithm's search process. Each time a virtual expansion is performed along an edge, a cost evaluation function is called to generate a range cost observation. This observation is then paired with the edge's feature vector and input into the Gaussian process model. The model's hyperparameters for the affected edge are then updated using either gradient descent or conjugate gradient methods.

[0053] When the search tree first expands to the spatial cell corresponding to the low-power UAV mission area, the entire search path from the root node to that cell is recorded as a candidate feasible path. The search does not stop at this point but continues iteratively until a preset convergence condition is met, resulting in several candidate feasible paths. The preset convergence condition is at least one of the following: the number of search tree iterations reaches a preset maximum value, the confidence interval width of the optimal path cost is less than a preset threshold, and the optimal path found in consecutive iterations has not changed.

[0054] Step S304: Determine the estimated travel cost of each candidate feasible path based on the path cost; select the candidate feasible path with the minimum estimated travel cost and determine it as the feasible path.

[0055] Specifically, for each candidate feasible path obtained in step S303, the path cost of each extended segment included in the path is obtained, and the path costs of each extended segment are summed to obtain the estimated range cost of the candidate feasible path. The estimated range costs of each candidate feasible path are compared, and the candidate feasible path with the smallest estimated range cost is selected as the final feasible path for the candidate UAV to fly from its current location to the low-power UAV mission area. The estimated range cost corresponding to this path is the estimated range cost of the candidate UAV.

[0056] In some embodiments, step S4 calculates the spatiotemporal similarity between each second inspection task sequence and the first inspection task sequence, including steps S401 to S404, each step being as follows: Step S401: Extract the subsequences within the overlapping interval of the second inspection task sequence and the first inspection task sequence on the time axis to form a time-aligned double sequence pair; Specifically, for each candidate UAV, the second inspection task sequence of that candidate UAV and the first inspection task sequence of the low-battery UAV are obtained. The execution times of the first and last positioning points in the two task sequences are read to determine the time intervals covered by each task sequence. The intersection of the two time intervals is taken, and subsequences falling into this intersection are extracted from the two task sequences to form a time-aligned double sequence pair. After time alignment, the two subsequences in the double sequence pair cover the same time interval, so that subsequent spatial deviation measurements can correspond one-to-one in time.

[0057] Step S402: Using the smallest time window as the unit, slide the scan from the starting position of the double sequence pair and calculate the spatial deviation measure of the double sequence pair within the current window in turn. Specifically, using a preset minimum time window as the basic scanning unit, the scan gradually slides along the time axis, starting from the initial position of the time-aligned double sequence pair. Within each current window, the two sub-task sequences falling into that window from the double sequence pair are extracted, and the spatial deviation metric between these two sub-task sequences is calculated. The spatial deviation metric is the maximum distance between corresponding task points in the two sub-sequences, i.e., the Hausdorff distance. Given sets A and B of sub-task sequences within two time windows, each task point contains spatial coordinates and a timestamp. The calculation of the Hausdorff distance H(A,B) involves two steps: for each task point in set A, calculate its distance to all task points in set B and take the minimum value; find the maximum value among all these minimum values, denoted as h(A,B). That is: in, This represents the spatial distance between task point a and task point b. For example, time can be used as a dimension and combined with the spatial distance in a weighted manner, or Euclidean distance in three-dimensional space can be used.

[0058] Similarly, calculate the maximum of the minimum distances from each task point in point set B to all task points in point set A, denoted as h(B,A): The final Hausdorff distance H(A,B) is the maximum of h(A,B) and h(B,A): Hausdorff distance measures the matching degree between two task sequences by finding the maximum deviation between two point sets. Only when the deviation of all corresponding points within the entire window is controlled within a threshold is the sequence considered overlapping. This distance value is the spatial deviation metric for the current window, used for subsequent comparisons with a preset threshold.

[0059] Step S403: The spatial deviation metric is compared with a preset threshold, and the comparison result determines whether the double sequence pair is a spatiotemporally overlapping subsequence or a spatiotemporally non-overlapping subsequence within the minimum time window. Specifically, the Hausdorff distance calculated in step S402 is compared with a preset spatial deviation threshold. If the Hausdorff distance is less than or equal to the preset threshold, it indicates that the spatial deviation between the two sub-task sequences within the current window is within an acceptable range, and the current window is determined to be a potentially overlapping window. At this point, the end boundary of the current window is gradually expanded, and the Hausdorff distance between the two sub-task sequences within the window is recalculated after each expansion until the Hausdorff distance exceeds the preset threshold. The longest consecutive sub-sequence that meets the condition before expansion is recorded as a spatiotemporally overlapping sub-task sequence. If the Hausdorff distance at the beginning of the current window is greater than the preset threshold, the two sub-sequences within the current window are determined to be spatiotemporally non-overlapping sub-sequences, and the current window is directly recorded as a spatiotemporally non-overlapping sub-task sequence. After completing one determination, the scanning window is moved to the next smallest time window after the end position of the current window, and the sliding scan continues until the entire time interval of the double sequence pair is scanned.

[0060] Step S404: Determine the spatiotemporal similarity by comparing the first total duration of the spatiotemporally overlapping subsequences with the second total duration of the spatiotemporally non-overlapping subsequences.

[0061] Specifically, the total time length of all spatiotemporally overlapping sub-task sequences identified in step S403 is taken as the first total time; the total time length of all spatiotemporally non-overlapping sub-task sequences is taken as the second total time. The ratio obtained by dividing the first total time by the sum of the first and second total times is the spatiotemporal similarity coefficient between the candidate UAV's second inspection task sequence and the low-power UAV's first inspection task sequence. This coefficient ranges from 0 to 1; a higher coefficient value indicates a greater degree of spatiotemporal overlap between the two task sequences, and a smaller adjustment range is required when the candidate UAV takes over the low-power UAV's task.

[0062] In some embodiments, step S5 determines the matching score between each candidate drone and the low-power drone based on the estimated range cost and spatiotemporal similarity of each candidate drone, and selects the takeover drone from the candidate drones based on the matching score, including steps S501 to S503, each step as follows: Step S501: Determine the task correlation degree of each candidate UAV based on the estimated range cost and corresponding spatiotemporal similarity of each candidate UAV. Specifically, upon receiving the pre-takeover command, each candidate UAV simultaneously initiates two independent analysis threads: one thread performs spatial reachability analysis, and the other performs task sequence spatiotemporal similarity analysis. After each thread completes its analysis, the estimated range cost and spatiotemporal similarity are quantified into dimensionless spatial reachability scores and task similarity scores, enabling the two heterogeneous indicators to be fused and compared on the same scale. The spatial reachability score is negatively correlated with the estimated range cost; the lower the estimated range cost, the higher the score. The task similarity score is positively correlated with spatiotemporal similarity; the higher the spatiotemporal similarity, the higher the score.

[0063] Preset weighting coefficients were assigned to both the spatial accessibility score and the task similarity score. The two scores were then weighted and summed to obtain the task correlation degree between each candidate UAV and the low-power UAV. This task correlation degree comprehensively reflects the feasibility of the candidate UAV reaching the low-power UAV's task area in space and the degree of matching in the task sequence for continuing the low-power UAV's inspection task. A higher task correlation degree indicates better overall performance of the candidate UAV in terms of both spatial accessibility and task matching.

[0064] Step S502: Determine the matching score of each candidate UAV based on the task relevance. Specifically, based on task relevance, the additional range cost required for each candidate drone to take over its mission and its remaining range safety margin are further obtained. The additional range cost refers to the additional flight distance or estimated energy consumption required for a candidate drone to fly from its current location to the low-battery drone's mission area and then back to its original inspection route. The remaining range safety margin refers to the remaining battery power of the candidate drone after deducting the power required to perform its original mission; it measures how many additional missions the candidate drone can undertake after completing its own mission. The task relevance, additional range cost, and remaining range safety margin of each candidate drone are input into a pre-built evaluation function for comprehensive calculation, resulting in a matching score for each candidate drone.

[0065] Step S503: Sort the matching scores of each candidate drone and select the candidate drone with the highest matching score as the takeover drone.

[0066] Specifically, each candidate drone calculates its own matching score locally and then reports it to the central node (ground station or cluster leader). The central node sorts all reported matching scores in descending order and selects the candidate drone with the highest matching score as the optimal takeover drone, forming a patrol task takeover pair between the low-battery drone and the optimal takeover drone, and broadcasts the matching result to the entire cluster.

[0067] In some embodiments, step S6 includes steps S61 to S62. Step S61, which constructs a multi-objective optimization model based on the first inspection task sequence and real-time constraint information, includes steps S601 to S605, each of which is detailed below: Step S601: Use the allocation relationship and task order as decision variables for the multi-objective optimization model; Specifically, the allocation relationship between the third task to be taken over and each takeover drone, as well as the execution order of the assigned tasks by each takeover drone, are determined as decision variables for the multi-objective optimization model. Decision variables include binary decision variables and continuous variables.

[0068] Binary decision variables Indicates the third task to be taken over Assign to takeover drone , Indicates the third task to be taken over Assigned to take over drone And execute at a certain sequential position. Indicates no assignment. Auxiliary binary decision variable. Indicates the order in which tasks are executed. Indicating the takeover of the drone In the task execution sequence, task Closely following the mission Then execute, Indicates that they are not adjacent. Continuous variables represent the actual start time of each third task to be taken over. .

[0069] Step S602: Based on the total number or total value of the third task to be taken over that has been successfully taken over corresponding to the decision variables, take maximizing the total number or total value as the first optimization objective. Specifically, the goal is to maximize the overall task completion rate, which is the total value or number of the third task to be taken over successfully. The first optimization objective function is expressed as: in, Indicates the third task to be taken over Priority or value weight, This indicates the task index (after traversing all third-party tasks awaiting takeover). This represents the drone index (traversing all takeover drones). The objective function is to maximize the number of high-value tasks that can be successfully taken over.

[0070] Step S603: Based on the total delay of the third task to be taken over corresponding to the decision variable, minimize the total delay as the second optimization objective; Specifically, the goal is to minimize the total task interruption time, which is the sum of the differences between the originally planned start time and the actual start time of all the third tasks to be taken over. The second optimization objective function is expressed as: in, Indicates the third task to be taken over Assigned to take over drone The actual start time after that. Indicates the third task to be taken over The objective function aims to minimize the actual execution time of each task, making it as close as possible to the original plan and reducing overall latency caused by task interruptions.

[0071] Step S604: Determine the constraints of the multi-objective optimization model based on the remaining battery power and mission capacity limit of each takeover drone; Specifically, the constraints include endurance constraints, task load constraints, task uniqueness and sequence constraints, and time window constraints.

[0072] Endurance constraints: For each takeover drone Its total energy consumption for performing all assigned tasks and flying between tasks must be less than or equal to its current remaining battery power, that is: in, To take over drones Execute the third takeover mission Energy consumption for task execution To take over drones In the third consecutive mission to be taken over and tasks Flight energy consumption between missions (if mission) and In drones If the execution sequence is not adjacent, then this term is 0. To assist binary decision variables Indicating the takeover of the drone In the task execution sequence, task Closely following the mission After that, otherwise 0), Indicates taking over the drone The current remaining battery power. This constraint ensures that the energy consumption of each takeover drone does not exceed its currently available battery power.

[0073] Task load constraint: The number of tasks assigned to each takeover drone shall not exceed its task capacity limit.

[0074] Task uniqueness and sequence constraints: Each third takeover task can be assigned to at most one takeover drone, and the task sequence on each takeover drone must satisfy the time sequence logic. The next task can only begin after the previous one is completed, and the flight time between tasks must be taken into account.

[0075] Time window constraint: The execution time of each third task to be taken over must be completed within its original planned time window.

[0076] Step S605: Construct a multi-objective optimization model based on the decision variables, the first optimization objective, the second optimization objective, and the constraints.

[0077] Specifically, the decision variables, first optimization objective, second optimization objective, and constraints determined in steps S601 to S604 are combined to construct a multi-objective optimization model. The multi-objective optimization model is a multi-objective mixed-integer linear programming model, containing a linear objective function, linear constraints, and integer decision variables. During the solution process, the weighted sum method or ε-constraint method can be used to transform it into a single-objective problem, or a multi-objective optimization solver can be directly used to find the Pareto front. All constraints are explicitly written in the form of linear equations or inequalities.

[0078] In some embodiments, please refer to Figure 2 , Figure 2This is a schematic diagram of the cooperative flight path solving process according to an embodiment of this application; step S62 solves the multi-objective optimization model to obtain the cooperative flight path of the takeover UAV, including steps S621 to S623, each step is as follows: Step S621: Input the first task to be taken over in the first inspection task sequence, as well as the current position, remaining power and task capacity limit of each takeover drone into the multi-objective optimization model; Specifically, the first tasks to be taken over in the first inspection task sequence (i.e., inspection tasks not completed by low-battery drones), as well as the current position, current remaining battery power, and task capacity limit of each takeover drone, are used as input parameters and input into the multi-objective optimization model constructed in step S61 as the basic data for solving the multi-objective optimization model. Among them, the task capacity limit is used to determine the load constraints of subsequent tasks.

[0079] Step S622: Solve the multi-objective optimization model and output a set of schemes including multiple candidate task allocation schemes; wherein each candidate task allocation scheme includes the allocation relationship between the second task to be taken over and each takeover drone, as well as the task order in which each takeover drone executes the second task to be taken over. Specifically, a multi-objective optimization solver (such as the NSGA-II variant) is used to solve the multi-objective optimization model constructed in step S61. During the solution process, new candidate task allocation schemes are generated iteratively, that is, the expected completion status of each task is calculated according to the current decision variable values ​​(allocation relationship and task order), forming a preliminary set of candidate task allocation schemes.

[0080] In each generation of the solution process, the fairness of each generated candidate task allocation scheme is verified and corrected through cooperative game theory. The specific verification and correction process is as follows: Each takeover drone is treated as a game participant, and an individual utility function is defined for each participant, which is the comprehensive benefit obtained by the drone in performing the assigned task under the current candidate scheme; at the same time, a coalition characteristic function is defined for the takeover drone coalition, which is the maximum total utility that the coalition can obtain by independently performing the task. The coalition characteristic function value serves as the benchmark value for fairness constraints. The total utility obtained by any takeover drone coalition under the candidate scheme must not be lower than the maximum total utility that the coalition can obtain when it leaves the larger group and performs the task independently.

[0081] A repair operator is introduced to verify whether any candidate task allocation scheme satisfies the fairness constraints of all alliances. If the current scheme satisfies the fairness constraints of all alliances, it is retained as a feasible solution; otherwise, a coordination mechanism is triggered to iteratively adjust the task allocation relationship, task execution order, or resource scheduling in the current scheme, gradually correcting the unfair parts until the scheme satisfies the fairness constraints of all alliances or approaches the constraint boundary.

[0082] After repair, candidate solutions that become feasible solutions are evaluated simultaneously using the first and second optimization objectives set in step S61 to obtain the target values ​​for each candidate solution on both objectives. Based on the target values ​​for task completion and downtime, and according to the dominance relationship and distribution density of the solutions, the candidate solution that is better in the multi-objective trade-off is selected to enter the next iteration. Specifically, for a candidate solution A and another candidate solution B, if A is not inferior to B in terms of task completion, not inferior to B in terms of downtime, and superior to B in at least one objective, then A is said to dominate B; during the iteration, only non-dominated solutions are retained and the population distribution density is kept uniform.

[0083] Repeat the above generation, verification and repair, evaluation and screening process until the maximum number of iterations is met. Output the Pareto solution set as a set of solutions. Each solution in the Pareto solution set corresponds to a candidate task allocation scheme that has a trade-off between task completion and interruption time and satisfies the fairness of cooperative game theory. Each candidate task allocation scheme includes the allocation relationship between the second task to be taken over and each takeover drone, as well as the task order in which each takeover drone executes the second task to be taken over.

[0084] Among them, the second task to be taken over is the task instance of the first task to be taken over after being assigned to each takeover drone, that is, the task of the first task to be taken over has been assigned to a specific takeover drone in the candidate scheme.

[0085] Step S623: Select the optimal candidate task allocation scheme from the scheme set, and determine the cooperative flight path corresponding to each takeover UAV based on the optimal candidate task allocation scheme, including: calculating the individual utility value of each takeover UAV under each candidate task allocation scheme; determining the comprehensive evaluation index of each candidate task allocation scheme based on the individual utility value of each takeover UAV; and selecting the candidate task allocation scheme with the largest comprehensive evaluation index value as the optimal candidate task allocation scheme.

[0086] Specifically, from the Pareto solution set output in step S622, i.e., the candidate solution set, the individual utility value of each takeover drone under each candidate task allocation scheme is calculated. The individual utility value represents the comprehensive benefit obtained by the drone in performing the assigned task under the current candidate scheme, which is determined by a combination of multiple dimensions such as the drone's contribution to the completion of the assigned task, flight energy efficiency, and task interruption delay.

[0087] Based on the individual utility values ​​of each takeover drone, a comprehensive evaluation index is calculated for each candidate task allocation scheme. The comprehensive evaluation index uses the Nash welfare function, which is the product of the individual utility values ​​of all takeover drones in each candidate scheme. The candidate task allocation scheme with the highest comprehensive evaluation index value is selected as the optimal candidate task allocation scheme. The product property of the Nash welfare function is extremely sensitive to any decrease in the utility of any participant, automatically tending to avoid exploited participants with extremely low utility, thus leading to an equilibrium scheme that ensures both good overall task completion and relatively satisfactory benefits for each participating drone.

[0088] Based on the optimal candidate task allocation scheme, a collaborative flight path is planned for each takeover UAV. After determining the second takeover task undertaken by each takeover UAV and its execution order, the original task waypoints of each takeover UAV and the waypoints of the newly takenover second takeover task are used as path nodes for unified replanning, generating a continuous flight trajectory starting from the current position and passing through all assigned task points (including original tasks and newly takenover tasks). The planning of the collaborative flight path inherits the optimization objectives and constraints of the multi-objective optimization model, including maximizing the overall task completion rate, minimizing the total task downtime, endurance constraints, task load constraints, task execution order logic constraints, and task time window constraints. Each takeover UAV performs the follow-up inspection task according to the planned collaborative flight path.

[0089] In some embodiments, this application provides a drone swarm inspection task takeover system, which is used to execute the drone swarm inspection task takeover method as described in this application.

[0090] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium, including: a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the drone swarm inspection task takeover method as described in any of the above-described method embodiments of this application.

[0091] The modules / units integrated into the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0092] The above are preferred embodiments of this application. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A method for taking over inspection tasks in a drone swarm, characterized in that, include: Obtain the remaining battery power and current flight power consumption of each drone in the drone cluster; calculate the remaining flight time of each drone based on the remaining battery power and the current flight power consumption, and identify the low-battery drone from the drone cluster based on the remaining flight time; Identify several candidate drones; Obtain a shared airspace map of the drone cluster; based on the shared airspace map, calculate the feasible path for each candidate drone to fly from its current location to the mission area of ​​the low-power drone, and obtain the estimated flight cost corresponding to each feasible path; Obtain the first inspection task sequence of the low-power UAV and the second inspection task sequence of each candidate UAV, and calculate the spatiotemporal similarity between each second inspection task sequence and the first inspection task sequence; Based on the estimated range cost and spatiotemporal similarity of each candidate drone, a matching score is determined between each candidate drone and the low-power drone, and a takeover drone is selected from the candidate drones based on the matching score. The system acquires real-time constraint information of the takeover drone, constructs a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information, solves the multi-objective optimization model to obtain the cooperative flight path of the takeover drone, and controls the takeover drone to perform the follow-up inspection task according to the cooperative flight path.

2. The method for taking over inspection tasks of a drone swarm according to claim 1, characterized in that, The step of calculating feasible paths for each candidate UAV to fly from its current location to the mission area of ​​the low-power UAV based on the shared airspace map, and obtaining the estimated range cost corresponding to each feasible path, includes: Based on the obstacle distribution density in the shared airspace map, the shared airspace map is spatially divided to generate a spatial unit map containing several spatial units; Based on the current position of each candidate drone, a path search is performed in the spatial unit map, starting from the current position and ending at the mission area of ​​the low-power drone, until a preset convergence condition is met, resulting in several candidate feasible paths. The path search includes: constructing a search tree with the current position as the root node; during each expansion, selecting an expansion node to be expanded from the existing nodes in the search tree, expanding the expansion node to adjacent spatial units to obtain an expanded flight segment, and obtaining the path cost corresponding to the expanded flight segment; when the search tree expands to the low-power UAV mission area, recording the current search path as the candidate feasible path, and continuing to expand the search until the preset convergence condition is met; Based on the path cost, the estimated travel cost of each candidate feasible path is determined; the candidate feasible path with the smallest estimated travel cost is selected and determined as the feasible path.

3. The method for taking over inspection tasks of a drone swarm according to claim 1, characterized in that, The calculation of the spatiotemporal similarity between each of the second inspection task sequences and the first inspection task sequence includes: Extract the subsequences within the overlapping interval of the second inspection task sequence and the first inspection task sequence on the time axis to form a time-aligned double sequence pair; Using the smallest time window as the unit, slide the scan from the starting position of the double sequence pair, and calculate the spatial deviation metric of the double sequence pair within the current window in turn; The spatial deviation metric is compared with a preset threshold, and the double sequence pair is determined to be a spatiotemporally overlapping subsequence or a spatiotemporally non-overlapping subsequence in the minimum time window based on the comparison result. The spatiotemporal similarity is determined based on the first total duration of the spatiotemporally overlapping subsequences and the second total duration of the spatiotemporally non-overlapping subsequences.

4. The method for taking over inspection tasks of a drone swarm according to claim 1, characterized in that, The process of solving the multi-objective optimization model to obtain the cooperative flight path of the takeover UAV includes: The first task to be taken over in the first inspection task sequence, as well as the current position, remaining power, and task capacity limit of each takeover drone, are input into the multi-objective optimization model. Solve the multi-objective optimization model to output a set of schemes including multiple candidate task allocation schemes; wherein, each candidate task allocation scheme includes the allocation relationship between the second task to be taken over and each of the takeover drones, and the task order in which each of the takeover drones executes the second task to be taken over; The optimal candidate task allocation scheme is selected from the set of schemes, and the cooperative flight path corresponding to each of the takeover UAVs is determined according to the optimal candidate task allocation scheme.

5. The method for taking over inspection tasks of a drone swarm according to claim 4, characterized in that, The step of selecting the optimal candidate task allocation scheme from the scheme set includes: Calculate the individual utility value of each takeover drone under each of the candidate task allocation schemes; Based on the individual utility value of each of the aforementioned takeover drones, a comprehensive evaluation index for each of the aforementioned candidate task allocation schemes is determined; The candidate task allocation scheme with the largest value of the comprehensive evaluation index is selected as the optimal candidate task allocation scheme.

6. The method for taking over inspection tasks of a drone swarm according to claim 4, characterized in that, The step of constructing a multi-objective optimization model based on the first inspection task sequence and the real-time constraint information includes: The allocation relationship and task order are used as decision variables in the multi-objective optimization model. Based on the total number or total value of the third task to be taken over that has been successfully taken over corresponding to the decision variables, the first optimization objective is to maximize the total number or the total value. Based on the total delay of the third task to be taken over corresponding to the decision variable, the second optimization objective is to minimize the total delay. The constraints of the multi-objective optimization model are determined based on the remaining battery power of each of the takeover drones and the upper limit of the mission capacity. The multi-objective optimization model is constructed based on the decision variables, the first optimization objective, the second optimization objective, and the constraints.

7. The method for taking over inspection tasks of a drone swarm according to claim 1, characterized in that, The process of determining a matching score between each candidate drone and the low-power drone based on the estimated range cost and spatiotemporal similarity of each candidate drone, and selecting a takeover drone from the candidate drones based on the matching score, includes: Based on the estimated range cost of each candidate UAV and the corresponding spatiotemporal similarity, the task correlation degree corresponding to each candidate UAV is determined. Based on the task relevance, a matching score is determined for each of the candidate drones; The matching scores of each candidate drone are sorted, and the candidate drone with the highest matching score is selected as the takeover drone.

8. The method for taking over inspection tasks of a drone swarm according to claim 1, characterized in that, The step of calculating the remaining flight time of each drone based on the remaining battery power and the current flight power consumption, and identifying low-battery drones from the drone cluster based on the remaining flight time, includes: Obtain the ambient temperature and real-time wind speed of each of the aforementioned drones; The remaining battery power is adjusted based on the ambient temperature to obtain the adjusted usable capacity of each drone. The current flight power consumption is corrected based on the real-time wind speed to obtain the corrected actual power consumption of each UAV. The remaining flight time of each of the drones is calculated based on the ratio of the corrected available capacity to the corrected actual power consumption. Drones with remaining battery life below a preset threshold are identified as low-battery drones.

9. A system for taking over inspection tasks in a drone swarm, characterized in that, The inspection task takeover system of the drone swarm is used to execute the inspection task takeover method of the drone swarm as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, the device containing the computer-readable storage medium is controlled to perform the inspection task takeover method for the drone swarm as described in any one of claims 1-8.

Images