An unmanned aerial vehicle cluster cooperative inspection autonomous task allocation scheduling method

Abstract

The application discloses a kind of unmanned aerial vehicle cluster cooperation inspection autonomous task allocation scheduling method, comprising the following steps: collecting inspection task data, unmanned aerial vehicle state data, environment data and communication data, construct unified scheduling model;Based on the scheduling model, initial task ownership result and task sequence are generated using distributed bidding method, and task conflict identification and resolution are realized through information interaction, conflict-free task allocation result and stable task sequence are formed;Current execution scheduling scheme is generated using rolling horizon optimization method, and threshold determination is carried out on the state change in the execution process, to identify the affected tasks;Based on the affected tasks, a local re-planning task set is formed, and under the condition of keeping the unaffected task section unchanged, the distributed calculation and rolling optimization are repeatedly executed to generate the updated task allocation result and execution scheduling scheme.The application realizes the efficient allocation and cooperative scheduling of unmanned aerial vehicle cluster tasks in complex dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of UAV swarm collaborative control and task scheduling technology, and in particular to a method for autonomous task allocation and scheduling of UAV swarm collaborative inspection. Background Technology

[0002] With the development of drone technology, drone swarms are widely used in scenarios such as power line inspection, engineering structure inspection, and complex environment monitoring. In existing technologies, the allocation and scheduling of multi-drone inspection tasks typically adopt centralized scheduling methods or rule-based distributed allocation methods. These methods allocate tasks by pre-setting task priorities or based on single factors such as distance and time, and combine them with fixed-cycle path planning and scheduling strategies to complete the inspection tasks.

[0003] In complex and dynamic environments, inspection tasks are characterized by uncertainty and dynamic changes, such as the real-time insertion of new tasks, changes in the remaining battery power of drones, fluctuations in communication links, and environmental disturbances. Existing technologies struggle to achieve efficient and stable task allocation and scheduling. On the one hand, centralized scheduling methods rely on global information, resulting in high computational complexity and stringent communication requirements, which can easily lead to computational latency and communication bottlenecks in large-scale drone swarms. On the other hand, traditional distributed methods typically employ full replanning or fixed-period update strategies when tasks change, leading to high computational overhead, slow response times, and potential disruption of the stability of already executed tasks.

[0004] Existing rolling optimization or path planning methods typically perform unified optimization on all tasks, lacking a protection mechanism for confirmed tasks. This can easily lead to frequent changes in task execution order, resulting in discontinuous paths or reduced resource utilization efficiency. Furthermore, for changes in local tasks, there is a lack of effective impact range identification mechanisms, often requiring the entire task sequence to be recalculated, making it difficult to balance real-time performance with global coordination.

[0005] Therefore, how to provide a method for autonomous task allocation and scheduling of collaborative inspection by UAV swarms is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an autonomous task allocation and scheduling method for collaborative inspection of UAV swarms. Based on a distributed bidding mechanism and a rolling optimization strategy, this invention dynamically allocates and schedules inspection tasks, and performs local replanning in conjunction with state changes during execution, thereby realizing the collaborative execution and real-time adjustment of multiple UAV tasks. It has the advantages of fast response speed, low computation and communication overhead, high task execution stability, and high resource utilization efficiency.

[0007] According to an embodiment of the present invention, a method for autonomous task allocation and scheduling of collaborative inspection by unmanned aerial vehicle (UAV) clusters includes the following steps: collecting inspection task data, UAV status data, environmental data and communication data, and constructing a unified scheduling model that includes task time window, task priority, current UAV position, resource constraints and link status parameters.

[0008] Based on a unified scheduling model, a consensus bundling algorithm is used for distributed bidding to generate candidate task bundles for each UAV, and to form initial task assignment results and corresponding initial task sequences.

[0009] Based on the initial task assignment results and initial task sequence, consistency propagation and conflict resolution are performed. Candidate task bundles with conflicts are released by suffix backtracking, and tasks are re-inserted and released starting from the current position of the UAV, generating conflict-free task assignment results and stable task sequences.

[0010] Based on the conflict-free task allocation results and stable task sequences, a rolling time-domain optimization algorithm is used to iteratively solve the task execution order, path cost, and time constraints to generate the current execution scheduling scheme and determine the prefix task segments that have been confirmed for execution.

[0011] During the execution of the current execution scheduling scheme, threshold judgment is performed on changes in state data. When the change exceeds the preset threshold, the affected suffix task segments in the unlocked tasks are identified, and a local replanning task set and corresponding computing node set are formed.

[0012] Based on the local replanning task set and the corresponding set of computing nodes, while keeping the constraints of the unaffected task segments unchanged, the distributed computing and rolling optimization are repeatedly executed, and the updated task allocation results and execution scheduling scheme are output.

[0013] Optionally, the construction of the unified scheduling model includes: collecting inspection task data, which includes task location coordinates, task priority level, task time window, and task service duration.

[0014] Collect drone status data, which includes the drone's current location coordinates, remaining battery power, maximum range, flight speed, and payload capacity.

[0015] Collect environmental data, including the distribution of obstacles, passable areas, and weather conditions within the inspection area.

[0016] The system collects communication data, including link quality, communication coverage, and communication stability between UAVs and between UAVs and ground nodes.

[0017] Based on inspection task data, UAV status data, environmental data, and communication data, a unified scheduling model is constructed. The task time window is mapped to a time constraint interval, the task priority is mapped to a scheduling weight, the current position of the UAV is mapped to the starting point of the initial path, the remaining battery power, range, and payload capacity are mapped to resource constraints, and the link quality is mapped to communication constraint parameters.

[0018] Optionally, the process of forming the initial task assignment result and the corresponding initial task sequence includes: inputting the inspection task data, UAV status data, task time window, communication data, task priority, UAV current position and resource constraints from the unified scheduling model into the consensus binding algorithm, and initializing the local task bundle, task bidding information and task winning record for each UAV.

[0019] Based on the current location of the UAV, the mission window, and resource constraints, candidate missions that meet the resource constraints are selected from the inspection mission data. The candidate missions are then inserted one by one into different positions in the current mission bundle of the UAV to determine the optimal insertion position and the corresponding bidding mission.

[0020] Based on the bidding task and its corresponding optimal insertion position, the bidding task is added to the local task bundle. After being added, the task execution order and resource occupancy status of the task bundle are updated until there are no candidate tasks that meet the task time window and resource constraints, thus forming the candidate task bundle and task execution order for each UAV.

[0021] The task bidding information is exchanged based on communication data, and the task winning record is updated consistently according to the received task bidding information to form a consistent task winning record for the same task. The local candidate task bundle is then corrected based on the consistent task winning record.

[0022] Once the mission winning records are consistent, the updated mission winning records are output as the initial mission assignment results, and the corrected mission ranking results in each UAV candidate mission bundle are output as the initial mission sequence corresponding to the initial mission assignment results.

[0023] Optionally, generating conflict-free task allocation results and stable task sequences includes: based on the initial task assignment results and the initial task sequence corresponding to the initial task assignment results, propagating task winning records and corresponding task sequence information among the UAVs based on communication data, and based on the received task winning records, identifying cases where the same task corresponds to multiple UAV identifiers to form a conflict task set.

[0024] Based on the set of conflict tasks, the winning records, bidding information and positions of the tasks in the initial task sequence are compared to determine the retained ownership records of each conflict task, and the UAV task sequences corresponding to the unretained ownership records are determined as conflict candidate task bundles.

[0025] Based on the conflict candidate task bundle, the position of the conflicting task is located in the corresponding initial task sequence, and the position of the conflicting task and the tasks after it are determined as suffix task segments. Backtracking and release are performed on the suffix task segments, the preceding task segments before the conflicting task are retained, and the task execution order and resource occupation status corresponding to the suffix task segments are deleted.

[0026] Based on the preceding task segments retained after the execution of the backtracking release, starting from the current position of the corresponding UAV, and combined with the end execution status of the preceding task segments, the release tasks in the suffix task segments are searched for and re-inserted one by one. After each re-insertion, the task execution order and resource occupation status are updated to form an updated task sequence.

[0027] Based on the updated task sequence, the process of propagating the task winning record, identifying conflicting tasks, backtracking and releasing suffix task segments, and re-inserting released tasks is repeated until there are no conflicting tasks in the task winning record. The task assignment result corresponding to the retained assignment record is output as the conflict-free task assignment result, and the corresponding updated task sequence is output as the stable task sequence.

[0028] Optionally, generating the current execution scheduling scheme includes: based on the conflict-free task allocation result, stable task sequence, UAV status data, environmental data, and communication data, detecting new task insertion, UAV status changes, communication status changes, and environmental disturbances, and extracting tasks directly related to the status changes from the stable task sequence to form the initial affected tasks.

[0029] Based on the initial affected tasks, the impact domain task set is expanded according to the temporal dependencies, task spatial adjacency relationships, and UAV resource coupling relationships in the stable task sequence. Then, the set of affected UAVs is determined based on the task attribution results corresponding to the impact domain task set.

[0030] A solution window is constructed using the set of tasks in the influence domain and the set of affected UAVs as the solution objects for rolling time-domain optimization. A protection window is constructed using task segments in the stable task sequence that do not belong to the set of tasks in the influence domain and have been confirmed to be executed, so that the tasks within the protection window remain locked during the rolling time-domain optimization process.

[0031] Extract the prefix task segment based on the protected window, and lock the task ownership, execution order, task start and end time and resource usage status corresponding to the prefix task segment.

[0032] Based on the current position of the corresponding UAV and the end execution state of the prefix task segment, a feasible connection domain for the suffix task segment within the solution window is constructed. The task search space of the suffix task segment is then limited to the feasible connection domain. With the task execution order, path cost, and time constraints as optimization objectives, the task execution order, task start and end times, and flight path of the suffix task segment are iteratively solved to generate candidate scheduling solutions.

[0033] Based on the candidate scheduling solution, the state continuity, task window constraints, resource constraints, and communication constraints between the prefix task segment and the suffix task segment are jointly verified.

[0034] When a candidate scheduling solution does not satisfy the joint verification, the feasible connection domain is reconstructed, and iterative solution is performed again based on the reconstructed feasible connection domain until an optimized scheduling solution that satisfies the joint verification is generated.

[0035] Based on the optimized scheduling solution that satisfies joint verification, the task execution order, flight path and task start and end times corresponding to the current execution step are extracted to generate the current execution scheduling scheme, and rolling time-domain optimization is achieved by advancing forward through a rolling window and repeatedly solving the problem.

[0036] Optionally, forming a local replanning task set and corresponding computing node set includes: during the execution of the current execution scheduling scheme, based on the real-time updates of UAV status data, communication data and environmental data, obtaining the task execution status, UAV location, remaining power, link quality and environmental parameters corresponding to the current moment, and comparing the task execution status at the current moment with the baseline status data used when executing the current execution scheduling scheme to obtain the change in each status data.

[0037] Based on the changes in each state data, threshold judgments are made for new task insertion, UAV state changes, communication state changes, and environmental disturbances. When the change in any state data exceeds the corresponding preset threshold, the corresponding state change is identified as a trigger event, and the task identifier, UAV identifier, and occurrence time of the trigger event are recorded.

[0038] Based on the triggering event, in the task sequence corresponding to the current execution scheduling scheme, the current execution step corresponding to the occurrence time is used as the boundary to determine the tasks that have entered the execution confirmation interval as locked task segments, and the tasks that have not entered the execution confirmation interval as unlocked task segments.

[0039] Based on the unlocked task segment, starting from the position of the task identifier corresponding to the triggering event in the task sequence, the task directly associated with the triggering event is identified as the initial affected task. Based on the sequential relationship of the initial affected task in the unlocked task segment, the tasks affected by task window constraints, resource constraints and path connection constraints are further identified to form a set of affected tasks.

[0040] Based on the distribution of the affected task set in the unlocked task segment, the continuous task segments following the initial affected tasks are extracted as the affected suffix task segments, and the tasks in the affected suffix task segments are identified as the local replanning task set.

[0041] Based on the task attribution results of each task in the local replanning task set, the set of UAVs participating in the local replanning task set processing is determined, forming a corresponding set of computing nodes.

[0042] Optionally, generating the updated task allocation result and execution scheduling scheme includes: based on the local replanning task set and the corresponding computing node set, extracting task segments that do not belong to the local replanning task set from the task sequence corresponding to the current execution scheduling scheme as unaffected task segments, and keeping the task affiliation, execution order, task start and end time, resource occupancy status and path connection relationship of the unaffected task segments unchanged.

[0043] Based on the unaffected task segments, the end execution state of the unaffected task segments and the corresponding current position of the UAV are determined as the boundary conditions for recalculating the local replanning task set, and the local replanning task set, the corresponding set of computing nodes, and the boundary conditions are input into the distributed computing process.

[0044] Based on the local replanning task set, the corresponding set of computing nodes, and boundary conditions, candidate task screening, task bundle construction, task bidding information exchange, and task winning record update are re-executed within the corresponding set of computing nodes, forming the local update task assignment results and local task sequence corresponding to the local replanning task set.

[0045] Based on the local update task attribution results and local task sequences, and with boundary conditions as constraints, rolling time-domain optimization is re-executed within the corresponding set of computing nodes. The task execution order, task start and end times, and flight paths in the local task sequences are iteratively solved to generate local candidate execution schemes corresponding to the local update task attribution results.

[0046] Based on local candidate execution schemes, consistency checks are performed on the task window constraints, resource constraints, communication constraints, and path connection constraints between the locally replanned task set and the unaffected task segments.

[0047] When a local candidate execution scheme fails to meet the consistency check, the local task sequence is updated based on the checked constraint relationship, and distributed computing and rolling time-domain optimization are performed again based on the updated local task sequence.

[0048] After the local candidate execution schemes satisfy the consistency check, the task assignment results of the locally updated task assignments are replaced with the task assignment results of the locally replanned task sets in the current execution scheduling scheme. The local candidate execution schemes are then inserted between the unaffected task segments to form the updated task assignment results and execution scheduling scheme.

[0049] The beneficial effects of this invention are as follows: By combining a unified scheduling model, consensus binding algorithm, conflict resolution mechanism, suffix backtracking release mechanism, and rolling time-domain optimization mechanism, this invention realizes closed-loop processing of UAV swarm inspection tasks from initial allocation, conflict resolution, execution scheduling to local replanning. This enables the system to dynamically identify and locally reconstruct affected tasks under conditions of new task insertion, UAV state changes, communication state changes, and environmental disturbances, thereby improving the real-time performance, stability, and continuity of task allocation and scheduling in complex dynamic environments.

[0050] This invention achieves local resolution of conflicting tasks and stable reconstruction of task sequences by performing consistency propagation, conflict identification, and suffix backtracking release on the initial task assignment results and initial task sequences, and re-inserting and releasing tasks starting from the current position of the UAV. This avoids the problems of large fluctuations in task order, discontinuous path connection, and excessive computational overhead caused by full replanning in the prior art, and improves the feasibility of task execution and sequence stability.

[0051] This invention constructs an influence domain task set, a solution window, and a protection window to lock the prefix task segment. Within the feasible connection domain, it iteratively solves the task execution order, task start and end times, and flight path of the suffix task segment. This achieves local rolling optimization while keeping the constraints of the unaffected task segment unchanged. It not only ensures the stability of the confirmed execution tasks but also improves the targeting and solution efficiency of scheduling updates under local state changes.

[0052] This invention identifies affected suffix task segments in unlocked tasks by thresholding task execution status, UAV location, remaining battery power, link quality, and environmental parameters. Based on local replanning of the task set and corresponding computing node set, it repeatedly performs distributed computing and rolling optimization, achieving rapid response to local changes in complex inspection scenarios. This reduces communication load and the scope of repeated computing, and improves the task collaboration capability, resource utilization efficiency, and overall inspection task completion efficiency of UAV clusters. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an autonomous task allocation and scheduling method for collaborative inspection of unmanned aerial vehicle (UAV) clusters proposed in this invention.

[0054] Figure 2 This is a schematic diagram of the execution scheduling process based on rolling time-domain optimization in this invention.

[0055] Figure 3 This is a schematic diagram illustrating the generation of the local replanning task set and computation node set in this invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0057] refer to Figure 1 - Figure 3 A method for autonomous task allocation and scheduling of collaborative inspection by unmanned aerial vehicle (UAV) swarms includes the following steps: collecting inspection task data, UAV status data, environmental data and communication data, and constructing a unified scheduling model that includes task time window, task priority, current UAV position, resource constraints and link status parameters.

[0058] Based on a unified scheduling model, a consensus bundling algorithm is used for distributed bidding to generate candidate task bundles for each UAV, and to form initial task assignment results and corresponding initial task sequences.

[0059] Based on the initial task assignment results and initial task sequence, consistency propagation and conflict resolution are performed. Candidate task bundles with conflicts are released by suffix backtracking, and tasks are re-inserted and released starting from the current position of the UAV, generating conflict-free task assignment results and stable task sequences.

[0060] Based on the conflict-free task allocation results and stable task sequences, a rolling time-domain optimization algorithm is used to iteratively solve the task execution order, path cost, and time constraints to generate the current execution scheduling scheme and determine the prefix task segments that have been confirmed for execution.

[0061] During the execution of the current execution scheduling scheme, threshold judgment is performed on changes in state data. When the change exceeds the preset threshold, the affected suffix task segments in the unlocked tasks are identified, and a local replanning task set and corresponding computing node set are formed.

[0062] Based on the local replanning task set and the corresponding set of computing nodes, while keeping the constraints of the unaffected task segments unchanged, the distributed computing and rolling optimization are repeatedly executed, and the updated task allocation results and execution scheduling scheme are output.

[0063] In this embodiment, the construction of the unified scheduling model includes: collecting inspection task data, which includes task location coordinates, task priority level, task time window, and task service duration.

[0064] Collect drone status data, which includes the drone's current location coordinates, remaining battery power, maximum range, flight speed, and payload capacity.

[0065] Collect environmental data, including the distribution of obstacles, passable areas, and weather conditions within the inspection area.

[0066] The system collects communication data, including link quality, communication coverage, and communication stability between UAVs and between UAVs and ground nodes.

[0067] Based on inspection task data, UAV status data, environmental data, and communication data, a unified scheduling model is constructed. The task time window is mapped to a time constraint interval, the task priority is mapped to a scheduling weight, the current position of the UAV is mapped to the starting point of the initial path, the remaining battery power, range, and payload capacity are mapped to resource constraints, and the link quality is mapped to communication constraint parameters.

[0068] The task time window is transformed into a time interval constraint that allows the task to be executed, and is used to limit the start and completion time range of the task. The task priority is transformed into a scheduling weight parameter, and is used to characterize the priority of the task in the allocation and sorting process. The current position of the UAV is used as the initial coordinates for path planning, and is used to calculate the path cost after the task is inserted. The remaining battery power, maximum range and payload capacity are transformed into resource constraints, and are used to limit the range of tasks that the UAV can execute. The link quality is transformed into a communication constraint parameter, and is used to constrain the reachability and stability of information interaction during the task allocation process.

[0069] In this embodiment, during the distributed bidding process using the consensus binding algorithm, each UAV calculates the task value based on inspection task data, UAV status data, task time window, task priority, UAV current position, resource constraints, and communication data from the unified scheduling model. For the first... The drone will be the first Each inspection task is inserted into the current task bundle of this machine. The In the case of a candidate location, first determine whether the candidate location meets the constraints of mission window, remaining battery power, range, payload capacity, and communication link; if the above constraints are not met, the candidate location will not participate in the bidding calculation; if the above constraints are met, the mission bid value corresponding to the candidate location will be calculated.

[0070] The task output value is determined based on task priority, path distance increment, time cost, energy consumption increment, and communication link quality, and is expressed as: ;in, Indicates the first The drone will be the first The inspection task was inserted into the first... The task yields value when there are multiple candidate positions; Indicates the first The task priority weight of each inspection task; Indicates the first The inspection task was inserted into the first... The path distance increased relative to the initial position after each candidate position; This indicates the time cost incurred after inserting the inspection task, which is determined by the task waiting time, the task time window deviation, or the task completion time increment. This indicates the energy consumption increment after inserting the inspection task; Indicates the first The drone carried out the first The communication link quality corresponding to each inspection task; , , , , These represent the weighting coefficients corresponding to task priority, path distance increment, time cost, energy consumption increment, and communication link quality, respectively.

[0071] For the same inspection task, the first The UAV traverses all candidate insertion positions in its current task bundle, calculates the task output value corresponding to each candidate insertion position, and selects the candidate insertion position with the highest task output value as the optimal insertion position for the inspection task in the UAV's task bundle. The final task output value of the UAV for this inspection task is expressed as: ;in, Indicates the first The drone to the first The ultimate goal of each inspection task is to generate value. Indicates the first The first inspection task was in the The set of candidate insertion positions in the current mission bundle of the UAV, wherein each candidate position in the set satisfies the constraints of mission window, remaining battery power, range, payload capacity, and communication link. When the set of candidate insertion positions is empty, the first... The drone is not the first Bidding is done for each inspection task.

[0072] During the task bidding information exchange process, each UAV sends its task identifier, UAV identifier, final task bid value, and corresponding insertion position. After receiving task bidding information from other UAVs, the final task bid values ​​corresponding to the same inspection task are compared, and the UAV identifier with the highest final task bid value is written into the task winning record. When multiple UAVs have the same final task bid value for the same inspection task, the task winning record is determined in the order of smaller path distance increment, smaller energy consumption increment, and higher communication link quality. Through the above method, the basis for calculating task bid value, the method for determining the optimal insertion position, and the update rules for task winning records are formed for the consensus binding algorithm.

[0073] In this embodiment, the formation of the initial task assignment result and the corresponding initial task sequence includes: inputting the inspection task data, UAV status data, task time window, communication data, task priority, UAV current position and resource constraints from the unified scheduling model into the consensus binding algorithm, and initializing the local task bundle, task bidding information and task winning record for each UAV.

[0074] Based on the current location of the UAV, the mission window, and resource constraints, candidate missions that meet the resource constraints are selected from the inspection mission data. The candidate missions are then inserted one by one into different positions in the current mission bundle of the UAV to determine the optimal insertion position and the corresponding bidding mission.

[0075] Based on the bidding task and its corresponding optimal insertion position, the bidding task is added to the local task bundle. After being added, the task execution order and resource occupancy status of the task bundle are updated until there are no candidate tasks that meet the task time window and resource constraints, thus forming the candidate task bundle and task execution order for each UAV.

[0076] After adding the bidding tasks to the task bundle, the task execution order is rearranged according to the task insertion position. The estimated start time and completion time of each task are calculated in turn according to the updated execution order. At the same time, the flight distance and energy consumption are accumulated. If subsequent candidate tasks do not meet the task time window constraints after being inserted at any position or cause the accumulated resource consumption to exceed the remaining power or range constraints of the UAV, it is determined that there are no candidate tasks that meet the constraints.

[0077] The task bidding information is exchanged based on communication data, and the task winning record is updated consistently according to the received task bidding information to form a consistent task winning record for the same task. The local candidate task bundle is then corrected based on the consistent task winning record.

[0078] The received task bid information is compared item by item with the local task bid set. For the same task, the drone with the better value is retained, and the local task winning record and corresponding candidate task bundle are updated accordingly.

[0079] Based on the consistent task winning record, delete the tasks that are no longer won by the local machine and their subsequent tasks from the local candidate task bundle, and re-execute the candidate task insertion and sorting update based on the remaining tasks.

[0080] Once the mission winning records are consistent, the updated mission winning records are output as the initial mission assignment results, and the corrected mission ranking results in each UAV candidate mission bundle are output as the initial mission sequence corresponding to the initial mission assignment results.

[0081] In this embodiment, generating conflict-free task allocation results and stable task sequences includes: based on the initial task assignment results and the initial task sequence corresponding to the initial task assignment results, propagating task winning records and corresponding task sequence information among the UAVs based on communication data, and based on the received task winning records, identifying cases where the same task corresponds to multiple UAV identifiers to form a conflict task set.

[0082] The received task winning records are compared item by item with the local task winning records. When the winning drone identifiers for the same task are inconsistent, the task is determined to be a conflicting task, and multiple corresponding drone identifiers are recorded.

[0083] Based on the set of conflict tasks, the winning records, bidding information and positions of the tasks in the initial task sequence are compared to determine the retained ownership records of each conflict task, and the UAV task sequences corresponding to the unretained ownership records are determined as conflict candidate task bundles.

[0084] The winning records of each drone in the same conflict task are summarized, and the task bidding information of each drone and the insertion position of the task in their respective initial task sequence are extracted. The drones are sorted according to their bid value. If the bid values ​​are the same, the resource consumption and time feasibility of the corresponding insertion positions are compared. The drone with the highest priority that meets the task time window and resource constraints is selected as the reserved ownership record of the conflict task.

[0085] Based on the conflict candidate task bundle, the position of the conflicting task is located in the corresponding initial task sequence, and the position of the conflicting task and the tasks after it are determined as suffix task segments. Backtracking and release are performed on the suffix task segments, the preceding task segments before the conflicting task are retained, and the task execution order and resource occupation status corresponding to the suffix task segments are deleted.

[0086] Locate the position of the conflicting task in the initial task sequence of the corresponding UAV, divide the position and all tasks thereafter into suffix task segments, remove the suffix task segments from the task sequence, retain only the preceding task segments before the conflicting task, and clear the execution order record and cumulative resource usage status of each task in the suffix task segments.

[0087] Based on the preceding task segments retained after the execution of the backtracking release, starting from the current position of the corresponding UAV, and combined with the end execution status of the preceding task segments, the release tasks in the suffix task segments are searched for and re-inserted one by one. After each re-insertion, the task execution order and resource occupation status are updated to form an updated task sequence.

[0088] Following the original order of each release task in the suffix task segment, each release task is sequentially inserted into the candidate positions between the reserved preceding task segment and the tasks that have been re-inserted. Then, it is determined whether the task time window and resource constraints are met after insertion. Once the insertion position that meets the constraints is determined, the re-insertion is completed.

[0089] Based on the updated task sequence, the process of propagating the task winning record, identifying conflicting tasks, backtracking and releasing suffix task segments, and re-inserting released tasks is repeated until there are no conflicting tasks in the task winning record. The task assignment result corresponding to the retained assignment record is output as the conflict-free task assignment result, and the corresponding updated task sequence is output as the stable task sequence.

[0090] In this embodiment, generating the current execution scheduling scheme includes: based on the conflict-free task allocation result, stable task sequence, UAV status data, environmental data, and communication data, detecting new task insertion, UAV status changes, communication status changes, and environmental disturbances, and extracting tasks directly related to the status changes from the stable task sequence to form the initial affected tasks.

[0091] Based on real-time updates of inspection task data, UAV status data, communication data, and environmental data, comparative analysis is performed on new task records, changes in UAV remaining battery power and location, changes in link quality, and changes in environmental parameters. When any data changes compared to the previous scheduling result, it is determined to be a status change.

[0092] Based on the initial affected tasks, the impact domain task set is expanded according to the temporal dependencies, task spatial adjacency relationships, and UAV resource coupling relationships in the stable task sequence. Then, the set of affected UAVs is determined based on the task attribution results corresponding to the impact domain task set.

[0093] Starting with the initially affected task, select the tasks adjacent to it in the stable task sequence, further select tasks whose spatial distance is less than a preset range, and tasks that share the same UAV remaining power, range or execution time constraints as them, as extended tasks and form an affected domain task set.

[0094] A solution window is constructed using the set of tasks in the influence domain and the set of affected UAVs as the solution objects for rolling time-domain optimization. A protection window is constructed using task segments in the stable task sequence that do not belong to the set of tasks in the influence domain and have been confirmed to be executed, so that the tasks within the protection window remain locked during the rolling time-domain optimization process.

[0095] After removing the tasks corresponding to the task set of the affected domain from the stable task sequence, select the continuous task segments that have entered the execution confirmation interval and whose execution order has been determined, use the continuous task segments as the protection window, and keep their task order and execution status unchanged.

[0096] Extract the prefix task segment based on the protected window, and lock the task ownership, execution order, task start and end time and resource usage status corresponding to the prefix task segment.

[0097] Based on the current position of the corresponding UAV and the end execution state of the prefix task segment, a feasible connection domain for the suffix task segment within the solution window is constructed. The task search space of the suffix task segment is then limited to the feasible connection domain. With the task execution order, path cost, and time constraints as optimization objectives, the task execution order, task start and end times, and flight path of the suffix task segment are iteratively solved to generate candidate scheduling solutions.

[0098] The task search space is a set of positions in the current task sequence that can be used to insert, adjust or rearrange tasks. The positions must simultaneously satisfy task time window constraints, resource constraints and path connection constraints, and are determined by the current position of the UAV and the end state of the preceding task.

[0099] Within the feasible connection domain, the insertion order of tasks in the suffix task segment is adjusted sequentially, and the start and end times of each task are recalculated according to the adjusted order. At the same time, the flight paths between adjacent tasks are updated. The order adjustment and timing and path updates are repeated until the task time window and resource constraints are met.

[0100] Based on the candidate scheduling solution, the state continuity, task window constraints, resource constraints, and communication constraints between the prefix task segment and the suffix task segment are jointly verified.

[0101] When a candidate scheduling solution does not satisfy the joint verification, the feasible connection domain is reconstructed, and iterative solution is performed again based on the reconstructed feasible connection domain until an optimized scheduling solution that satisfies the joint verification is generated.

[0102] Based on the optimized scheduling solution that satisfies joint verification, the task execution order, flight path and task start and end times corresponding to the current execution step are extracted to generate the current execution scheduling scheme, and rolling time-domain optimization is achieved by advancing forward through a rolling window and repeatedly solving the problem.

[0103] Select the task with the earliest time from the optimized scheduling solution that is located at the beginning of the current rolling window, obtain the execution order of the task in the task sequence, the corresponding flight path, and its start and end times, and use them as the scheduling result of the current execution step.

[0104] The current rolling step is executed based on the current execution scheduling scheme, and the unexecuted suffix task segments in the optimized scheduling solution are retained as the initial solution input for the next rolling time domain.

[0105] In this embodiment, the rolling time-domain optimization objective function and iterative solution method include: the rolling time-domain optimization takes the suffix task segment in the current rolling window as the solution object and the prefix task segment in the protected window as the locking object. Under the condition that the task ownership, execution order, task start and end time and resource occupation status of the prefix task segment remain unchanged, the task execution order, task start and end time and flight path of the suffix task segment are iteratively solved.

[0106] For the candidate scheduling solutions in the current rolling window Construct the rolling time-domain optimization objective function: ;in, Indicates candidate scheduling solutions The overall scheduling cost; This represents the path cost corresponding to the candidate scheduling solution; Indicates the time cost of the task; This indicates the energy consumption cost of drones; Indicates the cost of the communication link; This indicates the penalty for violating the constraint; , , , , These represent the weight coefficients of the corresponding cost terms.

[0107] Among them, path cost Determined based on the flight distance generated by the drone sequentially executing tasks within the current scrolling window; time cost. Determined based on task waiting time, task completion time, and task time window deviation; energy consumption cost. Determined based on the energy consumption generated by the UAV executing the current mission sequence; communication link cost. Determined based on the link quality between the UAV and adjacent computing or ground nodes during mission execution; constraints and penalties for violations. It is used to characterize the degree to which candidate scheduling solutions do not meet the task window constraint, remaining power constraint, range constraint, load constraint, communication constraint, or path connectivity constraint.

[0108] The constraints for rolling time-domain optimization include: ;in, This represents the feasible scheduling space determined by the mission window, the UAV's remaining battery power, range, payload capacity, communication link status, the execution status of the last segment of the prefix mission, and the UAV's current position. The optimization objective within the current rolling window is to find the candidate scheduling solution that minimizes the overall scheduling cost within the feasible scheduling space. ;in, This represents the optimized scheduling solution within the current scrolling window.

[0109] The specific algorithm for iterative solution includes: taking the suffix task segment in the stable task sequence that is located in the current rolling window as the initial task sequence, and taking the final execution state of the prefix task segment and the corresponding current position of the UAV as the initial boundary conditions to generate an initial candidate scheduling solution.

[0110] In the current candidate scheduling solution, a neighborhood update operation is performed on the suffix task segment. The neighborhood update operation includes at least one of task insertion, adjacent task swapping, and task retiring.

[0111] For each task sequence after neighborhood update, the corresponding task start time, task completion time, flight path between adjacent tasks, resource occupancy status, and communication link status are recalculated to form a new candidate scheduling solution.

[0112] Calculate the comprehensive scheduling cost corresponding to the new candidate scheduling solution. And compare it with the comprehensive scheduling cost of the current candidate scheduling solution.

[0113] When a new candidate scheduling solution satisfies the task window constraints, resource constraints, communication constraints, and path connectivity constraints, and its overall scheduling cost is less than that of the current candidate scheduling solution, the new candidate scheduling solution is updated to the current candidate scheduling solution.

[0114] If a new candidate scheduling solution does not meet the constraints or its overall scheduling cost is not reduced, discard the new candidate scheduling solution and continue to generate the next neighborhood candidate scheduling solution.

[0115] Repeat the process of neighborhood update, time sequence and path recalculation, constraint verification and comprehensive scheduling cost comparison until the termination condition is met. The termination condition includes that no candidate scheduling solution with a lower comprehensive scheduling cost is generated after a preset number of consecutive iterations, or that a preset maximum number of iterations is reached.

[0116] The current candidate scheduling solution retained at the time of termination is determined as the optimized scheduling solution in the current scrolling window, and the task execution order, flight path and task start and end time corresponding to the current execution step are extracted from the optimized scheduling solution to generate the current execution scheduling scheme.

[0117] In this embodiment, the formation of the local replanning task set and the corresponding computing node set includes: during the execution of the current execution scheduling scheme, based on the real-time updates of UAV status data, communication data and environmental data, obtaining the task execution status, UAV location, remaining power, link quality and environmental parameters corresponding to the current moment, and comparing the current task execution status with the baseline status data used when executing the current execution scheduling scheme to obtain the change in each status data.

[0118] The task execution status, drone location, remaining battery power, and link status collected at the current moment are matched one by one with the corresponding task start time, expected location, initial battery power, and link quality recorded when the current execution scheduling plan was generated. The difference of the status parameters under the same task and drone identifier is calculated to obtain the change in the corresponding status data.

[0119] Based on the changes in each state data, threshold judgments are made for new task insertion, UAV state changes, communication state changes, and environmental disturbances. When the change in any state data exceeds the corresponding preset threshold, the corresponding state change is identified as a trigger event, and the task identifier, UAV identifier, and occurrence time of the trigger event are recorded.

[0120] The system detects newly added task records to determine if a new task has been inserted; it compares changes in drone position deviation and remaining battery power; it compares changes in link quality; it compares changes in environmental parameters, and compares each change with a corresponding preset threshold. When any change exceeds its corresponding threshold, the task or drone corresponding to that change is identified as a trigger event.

[0121] The preset thresholds are the upper limits of various state changes that are pre-set when generating the current execution scheduling scheme, including the threshold for changes in the number of tasks, the threshold for location deviation, the threshold for changes in power consumption, the threshold for changes in link quality, and the threshold for changes in environmental parameters. These thresholds are used to determine whether state changes exceed the allowable range.

[0122] Based on the triggering event, in the task sequence corresponding to the current execution scheduling scheme, the current execution step corresponding to the occurrence time is used as the boundary to determine the tasks that have entered the execution confirmation interval as locked task segments, and the tasks that have not entered the execution confirmation interval as unlocked task segments.

[0123] Based on the unlocked task segment, starting from the position of the task identifier corresponding to the triggering event in the task sequence, the task directly associated with the triggering event is identified as the initial affected task. Based on the sequential relationship of the initial affected task in the unlocked task segment, the tasks affected by task window constraints, resource constraints and path connection constraints are further identified to form a set of affected tasks.

[0124] Starting from the position of the initially affected task in the unlocked task segment, the system traverses forward and backward according to the order of the task sequence, detecting tasks that have temporal dependencies on it. When a task's start time exceeds the task time window, its cumulative resource consumption exceeds the resource constraint, or its path connection is broken due to changes in the execution time of the preceding task, the task is identified as an affected task, and the system continues to expand along the task sequence direction to identify subsequent affected tasks.

[0125] Based on the distribution of the affected task set in the unlocked task segment, the continuous task segments following the initial affected tasks are extracted as the affected suffix task segments, and the tasks in the affected suffix task segments are identified as the local replanning task set.

[0126] Based on the task attribution results of each task in the local replanning task set, the set of UAVs participating in the local replanning task set processing is determined, forming a corresponding set of computing nodes.

[0127] In this embodiment, the threshold determination and local replanning triggering method includes: during the execution of the current execution scheduling scheme, the system collects UAV status data, communication data, and environmental data according to a preset sampling period, and compares the data collected at the current moment with the baseline status data used when generating the current execution scheduling scheme to obtain the change in status data. The change in status data includes UAV position deviation, change in remaining battery power, change in communication link quality, change in environmental disturbance, and change in newly added tasks.

[0128] For the A drone at the current moment The positional deviation is calculated as follows: ;in, Indicates the first A drone at the current moment Positional deviation, Indicates the first A drone at the current moment The actual location, Indicates the current execution scheduling scheme. A drone at the current moment The expected location. When Greater than the preset position deviation threshold When the change in the drone's position status meets the triggering condition.

[0129] For the A drone at the current moment The change in remaining battery power is calculated as follows: ;in, Indicates the first The change in the remaining battery power of the drone. Indicates the first The drone's current actual remaining battery power. Indicates the current execution scheduling scheme. A drone at the current moment The estimated remaining power. When Greater than the preset power change threshold ,or When the amount of power required to complete the unlocked task segment is less than the estimated power required, the change in the drone's resource status is determined to meet the trigger condition.

[0130] For changes in communication link quality, the current communication link quality is compared with the baseline communication link quality when the current execution scheduling plan was generated. When the current communication link quality between UAVs or between a UAV and a ground node falls below a preset communication quality threshold... Or the degradation of communication link quality exceeds the preset link degradation threshold. When the communication status change meets the triggering condition, it is determined that the change is true.

[0131] For changes in environmental disturbances, the current environmental parameters are compared with the baseline environmental parameters when the current execution scheduling scheme was generated. The environmental parameters include wind speed, rainfall intensity, changes in obstacle position, and changes in passable area. When the change in wind speed is greater than a preset wind speed change threshold, the change in obstacle position is greater than a preset obstacle change threshold, or the change in passable area is greater than a preset passable area change threshold, the environmental disturbance is determined to meet the triggering conditions.

[0132] For newly added task changes, the current inspection task data is compared with the inspection task data when the current execution scheduling scheme was generated; when a newly added inspection task not included in the current execution scheduling scheme is detected, and the task time window of the newly added inspection task intersects with the current remaining execution cycle, the newly added inspection task is identified as the task corresponding to the triggering event; when the task priority of the newly added inspection task is greater than the preset priority threshold, it is directly determined that the local replanning triggering condition is met.

[0133] The preset position deviation threshold Preset power change threshold Preset communication quality threshold Preset link descent threshold Both the preset thresholds corresponding to environmental disturbances and the preset thresholds are written into the unified scheduling model when generating the current execution scheduling scheme. In one implementation, the preset position deviation threshold... Set to 5% to 10% of the current adjacent task path length, preset power change threshold. Set the drone's battery level to 5% to 15% of its full charge, and preset the communication quality threshold. Set to 0.5 to 0.7 of the normalized communication link quality, with a preset link degradation threshold. Set to 20% to 40% of the baseline communication link quality. The preset threshold for environmental disturbances is set based on the wind speed limit of the inspection area, the obstacle detection accuracy, and the range of changes in the passable area.

[0134] When the change in any state data exceeds the corresponding preset threshold, the corresponding state change is identified as a trigger event, and the task identifier, drone identifier, and occurrence time of the trigger event are recorded. Based on the trigger event, in the task sequence corresponding to the current execution scheduling scheme, the current execution step corresponding to the occurrence time is used as the boundary, and tasks that have entered the execution confirmation interval are identified as locked task segments, while tasks that have not entered the execution confirmation interval are identified as unlocked task segments.

[0135] In the unlocked task segment, starting from the position of the task corresponding to the triggering event in the task sequence, the task directly associated with the triggering event is identified as the initial affected task, and the affected tasks continue to be identified along the task execution order of the unlocked task segment; when a subsequent task is caused by the triggering event to cause the task start time to exceed the task time window, the cumulative resource consumption to exceed the resource constraint, the communication link to fail to meet the communication constraint, or the path connection between adjacent tasks to fail to meet the passability constraint, the subsequent task is determined as the affected task; based on the distribution position of the affected tasks in the unlocked task segment, the continuous task segment after the initial affected task is extracted as the affected suffix task segment, and the tasks in the affected suffix task segment are determined as the local replanning task set.

[0136] Based on the original task attribution results of each task in the local replanning task set, the set of UAVs participating in the local replanning task set processing is determined, and a corresponding set of computing nodes is formed. For tasks that are not identified as affected suffix task segments, their task attribution, execution order, task start and end times, resource occupancy status, and path connection relationships remain unchanged.

[0137] In this embodiment, generating the updated task allocation result and execution scheduling scheme includes: based on the local replanning task set and the corresponding computing node set, extracting task segments that do not belong to the local replanning task set from the task sequence corresponding to the current execution scheduling scheme as unaffected task segments, and keeping the task affiliation, execution order, task start and end time, resource occupation status and path connection relationship of the unaffected task segments unchanged.

[0138] Based on the unaffected task segments, the end execution state of the unaffected task segments and the corresponding current position of the UAV are determined as the boundary conditions for recalculating the local replanning task set, and the local replanning task set, the corresponding set of computing nodes, and the boundary conditions are input into the distributed computing process.

[0139] Within the corresponding set of computing nodes, each UAV performs candidate task selection, task bundle construction, and task bid calculation based on the local replanning task set. They also exchange task bid information through communication data, update the received information in a consistent manner, and form a consistent task winning record and corresponding task attribution result.

[0140] Based on the local replanning task set, the corresponding set of computing nodes, and boundary conditions, candidate task screening, task bundle construction, task bidding information exchange, and task winning record update are re-executed within the corresponding set of computing nodes, forming the local update task assignment results and local task sequence corresponding to the local replanning task set.

[0141] Based on the local replanning task set and the corresponding computing node set, each UAV re-selects candidate tasks that meet the constraints according to the current UAV status and boundary conditions, inserts each candidate task into the construction task bundle and calculates the corresponding value, exchanges task bidding information through communication data, compares the bid values ​​of the same task, updates the task winning record, and corrects the local task bundle accordingly.

[0142] Based on the local update task attribution results and local task sequences, and with boundary conditions as constraints, rolling time-domain optimization is re-executed within the corresponding set of computing nodes. The task execution order, task start and end times, and flight paths in the local task sequences are iteratively solved to generate local candidate execution schemes corresponding to the local update task attribution results.

[0143] Based on the local task sequence, the position of each task in the sequence is adjusted sequentially, and the start time, end time and flight path between adjacent tasks are recalculated after each position adjustment. When the adjusted task execution order, task start and end times and flight path simultaneously satisfy the task time window constraint, resource constraint and path connection constraint, the adjustment result is retained; otherwise, the next sequence adjustment and recalculation are performed.

[0144] Based on local candidate execution schemes, consistency checks are performed on the task window constraints, resource constraints, communication constraints, and path connection constraints between the locally replanned task set and the unaffected task segments.

[0145] Connect the starting task of the local replanning task set in the local candidate execution scheme with the ending task of the unaffected task segment, compare whether the start time and end time meet the task time window constraint, calculate whether the connection path meets the resource constraint and path continuity, and determine whether the link in the connection process meets the communication constraint based on the communication data. When all the above constraints are met, it is determined to be consistent.

[0146] When a local candidate execution scheme fails to meet the consistency check, the local task sequence is updated based on the checked constraint relationship, and distributed computing and rolling time-domain optimization are performed again based on the updated local task sequence.

[0147] After the local candidate execution schemes satisfy the consistency check, the task assignment results of the locally updated task assignments are replaced with the task assignment results of the locally replanned task sets in the current execution scheduling scheme. The local candidate execution schemes are then inserted between the unaffected task segments to form the updated task assignment results and execution scheduling scheme.

[0148] To verify the feasibility of this invention in practice, it was applied to a large-scale infrastructure inspection scenario, employing multiple drones to conduct daily inspections of a wide-area structure. This inspection area includes complex spatial structures, various types of inspection tasks, and dynamically changing environmental factors, such as weather changes, communication quality fluctuations, and temporary adjustments to task priorities. In actual operation, the coordinated scheduling of dozens of inspection tasks and more than ten drones needs to be handled simultaneously. Traditional methods typically employ centralized scheduling or fixed-period update strategies, which can easily lead to significant computational latency, excessive communication load, and decreased task execution efficiency when the number of tasks is large or environmental changes are frequent.

[0149] In this scenario, inspection task data is collected, including the spatial location, priority, and time window of each task. Simultaneously, UAV status data is collected, including the UAV's current location, remaining battery power, flight speed, and payload capacity. A unified scheduling model is constructed by combining environmental and communication data. Based on this model, inspection tasks are initially allocated through a distributed bidding process. Each UAV generates a candidate task bundle based on its own resource constraints and current location, and a consistent task assignment result and initial task sequence are formed through task bidding information exchange.

[0150] In practical applications, after the initial system allocation, ten drones are assigned inspection tasks to different areas, totaling 48 tasks. Without the method of this invention, traditional scheduling methods require full path planning for all tasks, with each replanning taking an average of approximately 12.6 seconds and involving approximately 3.2MB of communication data exchange per round. During the inspection process, if a new task is inserted or the drone's status changes, the system typically needs to recalculate globally, causing task execution interruptions and delays for some tasks. The average task waiting time reaches 18.4 seconds, and the overall task completion time is 312 seconds.

[0151] After adopting the method of this invention, under the same task scale, the initial task allocation is completed through a distributed bidding mechanism, reducing the initial allocation time to 4.3 seconds and the communication data exchange volume to 1.1MB per round. During the execution scheduling phase, a rolling time-domain optimization method is used to optimize only the tasks within the current rolling window and lock the confirmed execution prefix task segments, ensuring the task execution order remains stable during local adjustments. During the inspection execution process, when new inspection tasks are detected, such as three new high-priority tasks, the system identifies the affected tasks through a state change detection mechanism and only replans the local task set consisting of the nine affected tasks, without affecting the remaining 39 tasks.

[0152] In this local replanning process, the computation node set only includes the four relevant UAVs, significantly reducing the communication load compared to traditional global scheduling which requires all ten UAVs to participate in the computation. The average computation time for local replanning is 2.1 seconds, and the communication data exchange volume is approximately 0.6MB, far lower than the overhead of global replanning. During task execution, a prefix locking and suffix release mechanism is adopted to ensure that executed or upcoming tasks are not interrupted, maintaining good path continuity. By comparing path smoothness metrics, the traditional method averages 2.7 path discontinuities per round, while the method of this invention reduces this to 0.4.

[0153] Regarding environmental disturbances, such as changes in wind speed leading to increased energy consumption for some drones, when the remaining battery power of a drone drops below a preset threshold, the system identifies the affected task segment and releases its subsequent tasks to a locally replanned task set. Through this reallocation, three of the five subsequent tasks originally assigned to that drone are reassigned to neighboring drones, while the remaining two tasks are still completed by the original drone through path adjustments. Throughout this process, the overall task completion rate remains at 100%, with no tasks lost.

[0154] Tests were conducted under conditions of multi-round dynamic task insertion and state changes, triggering 15 local replanning iterations during continuous operation. Statistical results show that, on average, each local replanning iteration involved 8.6 tasks, accounting for approximately 18% of the total number of tasks; and on average, 3.7 drones participated in the computation, accounting for approximately 37% of the total number of drones. In contrast, traditional methods require processing all tasks and all drones each time, resulting in a significantly larger computational scale.

[0155] In terms of overall performance, the method of this invention reduces the total task completion time from 312 seconds to 248 seconds, a reduction of approximately 20.5%; the average task waiting time is reduced from 18.4 seconds to 6.7 seconds; and the system response time for inserting new tasks is reduced from an average of 9.2 seconds to 2.4 seconds. Simultaneously, in terms of communication resource consumption, the total communication data volume is reduced from approximately 68MB to 24MB, a reduction of over 60%.

[0156] Regarding resource utilization, statistical analysis of UAV energy consumption efficiency reveals that traditional methods only achieve an energy utilization rate of 62% to 75% for some UAVs. However, the method of this invention, through dynamic task allocation and local replanning, increases the UAV energy utilization rate to 81% to 92%. Simultaneously, the task load of each UAV is more balanced, with the load standard deviation decreasing from 2.8 to 1.1, indicating a more rational task allocation.

[0157] Throughout the implementation process, the method of this invention, through a combination of distributed computing, local replanning, and rolling optimization, enables the system to quickly respond to task changes in complex and dynamic environments, avoiding the high computational overhead caused by global replanning, while maintaining the stability and path continuity of task execution. Through the above practical applications and data comparisons, it can be seen that this invention has significant effects in improving task allocation efficiency, reducing communication load, enhancing the system's real-time response capability, and improving overall resource utilization efficiency, effectively addressing the shortcomings of existing technologies in UAV swarm inspection and scheduling.

[0158] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for autonomous task allocation and scheduling of collaborative inspection by unmanned aerial vehicle (UAV) swarms, characterized in that, include: Collect inspection task data, UAV status data, environmental data and communication data, and construct a unified scheduling model that includes task time window, task priority, UAV current position, resource constraints and link status parameters; Based on a unified scheduling model, a consensus bundling algorithm is used for distributed bidding to generate candidate task bundles for each UAV, and to form initial task assignment results and corresponding initial task sequences. Based on the initial task assignment results and initial task sequence, consistency propagation and conflict resolution are performed. Candidate task bundles with conflicts are released by suffix backtracking, and tasks are re-inserted and released starting from the current position of the UAV, generating conflict-free task assignment results and stable task sequences. Based on the conflict-free task allocation results and stable task sequences, a rolling time-domain optimization algorithm is used to iteratively solve the task execution order, path cost and time constraints to generate the current execution scheduling scheme. During the execution of the current execution scheduling scheme, threshold judgment is performed on changes in state data. When the change exceeds the preset threshold, the affected suffix task segments in the unlocked tasks are identified, and a local replanning task set and corresponding computing node set are formed. Based on the local replanning task set and the corresponding set of computing nodes, while keeping the constraints of the unaffected task segments unchanged, the distributed computing and rolling optimization are repeatedly executed, and the updated task allocation results and execution scheduling scheme are output. The generation of conflict-free task allocation results and stable task sequences includes: Based on the initial task assignment results and the initial task sequence corresponding to the initial task assignment results, the task winning records and corresponding task sequence information are propagated among the UAVs based on communication data. Based on the received task winning records, the situation where the same task corresponds to multiple UAV identifiers is identified to form a set of conflicting tasks. Based on the set of conflict tasks, the winning records, bidding information and positions of the tasks in the initial task sequence are compared to determine the retained ownership records of each conflict task, and the UAV task sequences corresponding to the unretained ownership records are determined as conflict candidate task bundles. Based on the conflict candidate task bundle, locate the position of the conflicting task in the corresponding initial task sequence, and determine the position of the conflicting task and the tasks after it as the suffix task segment. Perform backtracking release on the suffix task segment, retain the preceding task segment before the conflicting task, and delete the task execution order and resource occupation status corresponding to the suffix task segment. Based on the preceding task segment retained after the execution of the backtrack release, starting from the current position of the corresponding UAV, and combined with the end execution state of the preceding task segment, the release tasks in the suffix task segment are searched for and re-inserted one by one. After each re-insertion, the task execution order and resource occupation status are updated to form an updated task sequence. Based on the updated task sequence, the process of propagating the task winning record, identifying conflicting tasks, backtracking and releasing the suffix task segment, and re-inserting the released task is repeated until there are no conflicting tasks in the task winning record. The task ownership result corresponding to the retained ownership record is output as the conflict-free task allocation result, and the corresponding updated task sequence is output as the stable task sequence. The process of generating the current execution scheduling scheme includes: Based on the conflict-free task allocation results, stable task sequences, UAV status data, environmental data, and communication data, new task insertion, UAV status changes, communication status changes, and environmental disturbances are detected. Tasks directly related to status changes are extracted from the stable task sequences to form the initial affected tasks. Based on the initial affected tasks, the impact domain task set is expanded according to the temporal dependency, task space adjacency and UAV resource coupling relationship in the stable task sequence, and the affected UAV set is determined based on the task attribution result corresponding to the impact domain task set. The solution window is constructed using the task set of the influence domain and the set of affected UAVs as the solution objects for rolling time-domain optimization, and the protection window is constructed using the task segments in the stable task sequence that do not belong to the task set of the influence domain and have been confirmed to be executed. Extract the prefix task segment based on the protected window, and lock the task ownership, execution order, task start and end time and resource usage status corresponding to the prefix task segment; Based on the current position of the corresponding UAV and the end execution state of the prefix task segment, a feasible connection domain for the suffix task segment is constructed within the solution window. The task search space of the suffix task segment is then limited to the feasible connection domain. With the task execution order, path cost, and time constraints as optimization objectives, the task execution order, task start and end times, and flight path of the suffix task segment are iteratively solved to generate candidate scheduling solutions. Based on the candidate scheduling solution, the state continuity, task window constraints, resource constraints and communication constraints between the prefix task segment and the suffix task segment are jointly verified. When a candidate scheduling solution does not satisfy the joint verification, the feasible connection domain is reconstructed, and iterative solution is performed again based on the reconstructed feasible connection domain until an optimized scheduling solution that satisfies the joint verification is generated. Based on the optimized scheduling solution that satisfies joint verification, the task execution order, flight path, and task start and end times corresponding to the current execution step are extracted to generate the current execution scheduling scheme.

2. The method for autonomous task allocation and scheduling of UAV swarm collaborative inspection according to claim 1, characterized in that, The construction of the unified scheduling model includes: Collect inspection task data, which includes task location coordinates, task priority level, task time window, and task service duration. Collect UAV status data, which includes the UAV's current location coordinates, remaining battery power, maximum range, flight speed, and payload capacity; Collect environmental data, including the distribution of obstacles, passable areas, and weather conditions within the inspection area; Collect communication data, including link quality, communication coverage, and communication stability between UAVs and between UAVs and ground nodes; Based on inspection task data, UAV status data, environmental data, and communication data, a unified scheduling model is constructed. The task time window is mapped to a time constraint interval, the task priority is mapped to a scheduling weight, the current position of the UAV is mapped to the starting point of the initial path, the remaining battery power, range, and payload capacity are mapped to resource constraints, and the link quality is mapped to communication constraint parameters.

3. The method for autonomous task allocation and scheduling of collaborative inspection by unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The formation of the initial task attribution result and the corresponding initial task sequence includes: The unified scheduling model inputs inspection task data, UAV status data, task time window, communication data, task priority, UAV current location and resource constraints into the consensus binding algorithm, and initializes the local task bundle, task bidding information and task winning record for each UAV. Based on the current location of the UAV, the mission window, and resource constraints, candidate missions that meet the resource constraints are selected from the inspection mission data, and the candidate missions are inserted one by one into different positions of the current mission bundle of the UAV to determine the corresponding optimal insertion position and the corresponding bidding mission. Based on the bidding task and its corresponding optimal insertion position, the bidding task is added to the local task bundle, and the task execution order and resource occupation status corresponding to the task bundle are updated after the addition, until there are no candidate tasks that meet the task time window and resource constraints, thus forming the candidate task bundle and task execution order for each UAV. Task bidding information is exchanged based on communication data, and the task winning record is updated consistently according to the received task bidding information to form a consistent task winning record for the same task. The local candidate task bundle is then corrected based on the consistent task winning record. Once the mission winning records are consistent, the updated mission winning records are output as the initial mission assignment results, and the corrected mission ranking results in each UAV candidate mission bundle are output as the initial mission sequence corresponding to the initial mission assignment results.

4. The method for autonomous task allocation and scheduling of collaborative inspection by unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The process of forming the local replanning task set and the corresponding set of computing nodes includes: During the execution of the current execution scheduling scheme, based on the real-time updates of UAV status data, communication data and environmental data, the task execution status, UAV location, remaining power, link quality and environmental parameters corresponding to the current moment are obtained, and the task execution status at the current moment is compared with the baseline status data used in the current execution scheduling scheme to obtain the change in each status data. Based on the amount of change in each state data, threshold judgments are made for new task insertion, UAV state change, communication state change and environmental disturbance. When the amount of change in any state data exceeds the corresponding preset threshold, the corresponding state change is determined as a trigger event, and the task identifier, UAV identifier and occurrence time of the trigger event are recorded. Based on the triggering event, in the task sequence corresponding to the current execution scheduling scheme, the current execution step corresponding to the time of occurrence is used as the boundary to determine the tasks that have entered the execution confirmation interval as locked task segments, and the tasks that have not entered the execution confirmation interval as unlocked task segments. Based on the unlocked task segment, starting from the position of the task identifier corresponding to the triggering event in the task sequence, the task directly associated with the triggering event is identified as the initial affected task. Based on the sequential relationship of the initial affected task in the unlocked task segment, the tasks affected by task window constraints, resource constraints and path connection constraints are further identified to form a set of affected tasks. Based on the distribution of the affected task set in the unlocked task segment, the continuous task segments following the initial affected tasks are extracted as the affected suffix task segments, and the tasks in the affected suffix task segments are identified as the local replanning task set. Based on the task attribution results of each task in the local replanning task set, the set of UAVs participating in the local replanning task set processing is determined, forming a corresponding set of computing nodes.

5. The method for autonomous task allocation and scheduling of UAV swarm collaborative inspection according to claim 1, characterized in that, The process of generating the updated task allocation results and execution scheduling scheme includes: Based on the local replanning task set and the corresponding computing node set, the task segments that do not belong to the local replanning task set are extracted from the task sequence corresponding to the current execution scheduling scheme as unaffected task segments, and the task ownership, execution order, task start and end time, resource occupation status and path connection relationship of the unaffected task segments remain unchanged. Based on the unaffected task segments, the end execution state of the unaffected task segments and the corresponding current position of the UAV are determined as the boundary conditions for the recalculation of the local replanning task set, and the local replanning task set, the corresponding set of computing nodes and the boundary conditions are input into the distributed computing process. Based on the local replanning task set, the corresponding set of computing nodes and boundary conditions, candidate task screening, task bundle construction, task bidding information interaction and task winning record update are re-executed within the corresponding set of computing nodes to form the local update task assignment result and local task sequence corresponding to the local replanning task set. Based on the local update task assignment results and local task sequences, and with boundary conditions as constraints, the rolling time-domain optimization is re-executed within the corresponding set of computing nodes. The task execution order, task start and end times and flight paths in the local task sequences are iteratively solved to generate local candidate execution schemes corresponding to the local update task assignment results. Based on local candidate execution schemes, consistency checks are performed on the task window constraints, resource constraints, communication constraints, and path connection constraints between the local replanning task set and the unaffected task segments. When a local candidate execution scheme fails to meet the consistency check, the local task sequence is updated based on the checked constraint relationship, and distributed computing and rolling time-domain optimization are performed again based on the updated local task sequence. After the local candidate execution schemes satisfy the consistency check, the task assignment results of the locally updated task assignments are replaced with the task assignment results of the locally replanned task sets in the current execution scheduling scheme. The local candidate execution schemes are then inserted between the unaffected task segments to form the updated task assignment results and execution scheduling scheme.

Images