Multi-unmanned aerial vehicle multi-vehicle collaborative inspection scheduling optimization and path planning method and system

Abstract

The invention discloses a multi-unmanned aerial vehicle and multi-vehicle collaborative inspection scheduling optimization and path planning method and system, and relates to the field of collaborative scheduling and optimization of multiple intelligent agents, and the method comprises the steps: firstly constructing a double-layer road network model comprising a ground layer and an air layer; secondly, by taking minimization of the maximum time for all vehicles to complete a task and return to an end point as an optimization target, establishing a comprehensive multi-constraint collaborative scheduling optimization model, then designing and applying an improved adaptive large neighborhood search algorithm to solve a vehicle path and an unmanned aerial vehicle path, and generating a collaborative inspection scheduling optimization scheme. According to the invention, through integrated collaborative optimization of the vehicle path, the unmanned aerial vehicle task allocation and path and the launching / recovery plan, space-time conflicts of ground nodes can be effectively avoided, the completion time of the whole inspection task is significantly shortened, the utilization efficiency of the vehicle and the unmanned aerial vehicle is improved, and the system is suitable for popularization and application. And a set of efficient and reliable scheduling optimization solution is provided for large-scale multi-agent collaborative inspection.

Description

Technical Field

[0001] The present invention relates to the field of multi-vehicle-multi-UAV collaborative inspection, and in particular to a multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method and system. Background Art

[0002] In the inspection tasks of complex environmental facilities such as power lines, oil and gas pipelines, urban transportation, and environmental monitoring, traditional manual inspection methods are time-consuming, labor-intensive, inefficient, and pose significant safety risks in high-risk environments. They are no longer able to meet the efficiency and safety requirements of modern inspection needs. With the rapid development of drone technology and intelligent vehicle technology, multi-drone and multi-vehicle collaborative operation systems have gradually become the mainstream model to replace traditional manual inspections. By combining the aerial flexibility of drones with the ground support capabilities of vehicles, this system can significantly improve inspection efficiency, reduce labor costs, and enhance operational safety. However, existing collaborative scheduling methods still face several key technical challenges in practical applications, especially in dealing with issues such as the collaboration of heterogeneous devices in three-dimensional space, dynamic task insertion, and resource conflict resolution. There are significant limitations, which restrict their application potential in complex environments.

[0003] First, existing collaborative scheduling methods are inefficient when handling the coordination of heterogeneous devices in three-dimensional space. As heterogeneous devices, drones and vehicles have fundamental differences in physical properties, motion patterns, and resource requirements. For example, drones operating in three-dimensional space are limited by battery life, flight altitude, and external environmental factors (such as wind speed); vehicles, on the other hand, are constrained by the topology, traffic conditions, and carrying capacity of the ground road network. Existing methods often employ a hierarchical optimization architecture that fails to fully account for these differences, resulting in inefficient collaboration between drones and vehicles. In large-scale missions, there is a lack of scientific and rational decision-making to plan vehicle routes, allocate drone tasks, and decide on vehicle recovery and launch. For example, drones may interrupt their inspection missions due to a lack of timely ground vehicle support (such as mobile charging stations), or vehicle scheduling may not effectively match the drone's inspection rhythm, resulting in wasted resources and time delays.

[0004] Secondly, existing technologies are slow to respond to dynamic task insertion. Temporary additions to inspection tasks or handling unexpected equipment failures are common during inspections of complex facilities. However, pre-defined inspection plans are difficult to plan in real time, cannot quickly respond to environmental changes, and lack efficient task allocation algorithms. This results in the system being unable to adjust work plans in response to dynamic demands, impacting the timeliness and effectiveness of inspections.

[0005] Furthermore, resource conflicts are a major bottleneck in existing technologies. When multiple drones and vehicles collaborate, resource conflicts are inevitable. For example, after completing their missions, multiple drones may compete for the same vehicle, resulting in insufficient vehicle capacity to recover all drones. Alternatively, multiple vehicles may encounter path conflicts in complex terrain (such as one-way lanes, winding terrain, and dense intersections), leading to a deadlock and preventing drones from being recovered in a timely manner, resulting in wasted resources. Existing scheduling algorithms often lack effective mechanisms for resolving these conflicts, resulting in hindered mission execution or reduced resource utilization.

[0006] Therefore, there is an urgent need for a multi-UAV and multi-vehicle collaborative inspection scheduling optimization and path planning method and system that can solve the above problems. Summary of the Invention

[0007] The purpose of the present invention is to provide a multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method and system. Aiming at the problems of node conflict constraints, task scheduling coupling and path planning in multi-vehicle and multi-UAV collaborative inspection tasks in complex environments, a multi-UAV and multi-vehicle collaborative inspection path planning model under a double-layer road network with the goal of minimizing inspection time is proposed, and an improved adaptive large neighborhood search algorithm (IALNS) is designed for solution. Then, a multi-vehicle and multi-UAV collaborative task allocation and scheduling scheme is planned in a complex environment to achieve collaborative and efficient operation.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] The present invention provides a multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method, comprising the following steps:

[0010] S1. Construct a two-layer road network model, including a ground layer and an aerial layer; the ground layer contains ground nodes where vehicles can drive and park and ground arcs connecting nodes, and the aerial layer contains aerial nodes that define drone inspection tasks and aerial arcs connecting nodes;

[0011] S2. Based on the two-layer road network model, plan the driving paths of multiple vehicles, the driving paths including starting from the starting ground node, visiting the ground node to perform the drone launch or recovery operation, and returning to the ending ground node;

[0012] S3. Planning inspection paths for multiple drones, wherein the drones are launched from a vehicle at a selected ground node, and are recovered by the vehicle at a selected ground node after performing an inspection mission along an air arc;

[0013] S4. Coordinate the scheduling of vehicle and UAV mission sequences to ensure that pre-set constraints are met, including mission coverage constraints, vehicle path constraints, UAV endurance constraints, vehicle capacity constraints, launch and recovery time constraints, time synchronization constraints, and space-time occupancy conflicts of ground nodes.

[0014] S5. Taking minimizing the maximum time required for all vehicles to complete their tasks as the optimization goal, the vehicle paths and UAV paths are solved through the improved adaptive large neighborhood search algorithm to generate a collaborative inspection scheduling optimization solution.

[0015] Preferably, the spatiotemporal occupancy conflict constraint of the ground node is determined by introducing a binary variable to determine the order in which two vehicles arrive at the same node, thereby ensuring that the departure time of the later arriving vehicle is later than the departure time of the earlier arriving vehicle.

[0016] Preferably, the improved adaptive large neighborhood search algorithm includes:

[0017] S51. Initialization step: using a clustering algorithm to assign aerial inspection tasks to different vehicles, and generating an initial path and UAV launch / recovery plan for each vehicle based on the clustering results;

[0018] S52. Destruction step: selecting random destruction, similar node destruction based on high-frequency conflict, or worst node destruction strategy to remove some nodes or tasks in the current solution;

[0019] S53. Repair step, using greedy strategy, regret value strategy or random noise strategy to reinsert the removed tasks and generate new feasible solutions.

[0020] Preferably, in the destruction step, when the ground node is removed, the associated drone task chain is deleted simultaneously; when the aerial arc task is removed, the drone path is adjusted according to the task position or the entire task is deleted.

[0021] Preferably, in the repairing step, when inserting a task, the optimal insertion scheme is selected based on the cost increment or regret value of the feasible insertion position, or a probabilistic selection mechanism is adopted after correcting the cost increment by adding random noise.

[0022] Preferably, the UAV continuously performs multiple aerial arc inspection missions in one launch mission, and the same UAV can be recovered by different vehicles at different ground nodes.

[0023] Preferably, the preset constraints also include: a dynamic balance constraint on the number of drone launches and recoveries, a vehicle path sub-ring elimination constraint, and a drone mission integrity constraint.

[0024] The present invention also provides a multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning system, comprising:

[0025] A model building module, used for building the double-layer road network model;

[0026] Path planning module, used to plan the vehicle's driving path and the drone's inspection path;

[0027] A collaborative scheduling module that determines the mission sequence of vehicles and drones and ensures that pre-set constraints are met;

[0028] The optimization solution module uses the IALNS algorithm to solve the optimization solution.

[0029] Preferably, the collaborative scheduling module enforces the spatiotemporal occupancy conflict constraints of ground nodes, ensuring that the same node is occupied by at most one vehicle in any time period.

[0030] Preferably, the IALNS algorithm of the optimization solution module includes an initialization submodule, a destruction submodule and a repair submodule, wherein the destruction submodule executes a destruction strategy and applies adjustment rules, and the repair submodule executes a repair strategy and applies insertion rules.

[0031] Compared with the prior art, the present invention has achieved the following beneficial technical effects:

[0032] The present invention provides a multi-UAV and multi-vehicle collaborative inspection scheduling optimization and path planning method and system. The method first constructs a two-layer road network model including a ground layer (for vehicle driving and parking) and an aerial layer (for UAV inspection operations), which accurately describes the operating environment and interactive relationship between vehicles and UAVs. Secondly, with the optimization goal of minimizing the maximum time for all vehicles to complete the task and return to the destination, a collaborative scheduling optimization model is established that comprehensively considers multiple constraints such as task coverage, vehicle path constraints, UAV endurance limitations, the number of vehicle carriers, launch and recovery time, vehicle and UAV recovery timing synchronization, and key ground node space-time occupancy conflicts. The model supports the execution of multiple inspection tasks by the UAV at one time after launch, and allows a flexible operation mode in which the UAV is launched by one vehicle and then recovered by the same launching vehicle or another different vehicle. Then, in order to efficiently solve this complex optimization problem, the present invention designs and applies an improved adaptive large neighborhood search (IALNS) algorithm, which includes a specific initialization strategy, a variety of destruction operators (such as those based on conflict similarity, removal of the worst node, etc.) and repair operators (such as greedy insertion, regret value insertion, random noise insertion, etc.) that are combined with the characteristics of the problem, and defines special adjustment and insertion rules to maintain the feasibility of the solution. The present invention can effectively avoid spatiotemporal conflicts of ground nodes by integrating and collaboratively optimizing vehicle paths, UAV task allocation and paths, and launch / recovery plans, significantly shortening the completion time of the overall inspection task, and improving the utilization efficiency of vehicles and UAVs, providing a set of efficient and reliable scheduling optimization solutions for large-scale multi-agent collaborative inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 A multi-UAV-multi-vehicle collaborative inspection path planning model in a double-layer road network;

[0035] Figure 2 It is a characteristic case of vehicle-UAV collaborative operation;

[0036] Figure 3 This is a schematic diagram of the improved adaptive large domain search algorithm framework;

[0037] Figure 4 Generate pseudocode for algorithm population initialization;

[0038] Figure 5 Pseudocode for vehicle recovery drone;

[0039] Figure 6 Pseudocode for launching drones for vehicles;

[0040] Figure 7 Schematic diagram of the destruction strategy;

[0041] Figure 8 This is a distribution map of inspection environment tasks. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] The purpose of the present invention is to provide a multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method and system to solve the problems existing in the prior art.

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Example 1:

[0046] This embodiment provides a multi-UAV and multi-vehicle collaborative inspection scheduling optimization and path planning method. First, a mathematical model system is constructed to describe the multi-UAV and multi-vehicle collaborative inspection path planning model under a double-layer road network, and an improved adaptive large neighborhood search algorithm (IALNS) is designed based on the model characteristics and its multi-dimensional constraints to solve the problem.

[0047] In order to simulate the road network conditions in a complex environment, the multi-UAV-multi-vehicle collaborative inspection path planning model under a double-layer road network is defined in a double-layer graph G = (N, A, M) consisting of nodes (N), arcs (A) and layers (M), as shown in Figure 1 As shown. It includes the ground layer (vehicle operating environment) and the air layer (drone operating environment), which are defined by T (ground layer) and F (air layer) in the set respectively. There is M = {T, F}. The set of air layer nodes can be expressed as There are a total of |N air network nodes F |. The set of ground layer nodes can be expressed as Total | N T +2|. The set of parking points that vehicles can go to in the ground layer nodes is There are a total of |N parking spots T | one, let Denote all nodes that the vehicle may leave, let Represents the combination of all nodes that a vehicle may reach. Set A represents the arcs between all nodes in the network, consisting of ground network arcs A T and aerial network arc A F In these two subsets, all arcs connect a pair of different nodes. It is worth noting that for A F and A T Any arc (a i ,a j ) and (g i ,g j ) all satisfy a i ≠a j , g i ≠g j ,The above statement ensures that each edge in the graph uniquely represents a path between two independent nodes.

[0048] During the mission, first, |V| trucks depart from warehouse g0, and each vehicle carries at most |D k | drones perform inspection tasks, there are |X| drones in total, each drone is represented by code x. For any truck v k All have v k ∈V,V={v1,v2,…,vk ,…,|V|}. Truck v k The drone with code name x can be represented as Among them D k represents the set of drones carried by the k-th vehicle, and the set of all drones is During the operation, the vehicle k Only the node set N at the ground level can be T In the process of moving, launching / recovering drone x, the launch time tl and the recovery time tr are generated. Similarly, after being launched, drone x can independently move in the air layer node set N F Perform inspection tasks. It is worth noting that Figure 1 In (a), the vehicle needs to meet strict ground constraints during the execution of the task. Any node on the ground layer can only be occupied by one vehicle at a time to perform the task, such as Figure 1 As shown in (a), vehicle v3 needs to perform the task from g1 to g5 in the time period t1→t5. Vehicle v3 can choose three routes (shown by the blue line) to go to g5. From the time occupation table in the figure, it can be seen that it is not feasible for vehicle v3 to go from g1→g3 or from g1→g2→g4. This is mainly because vehicles v1 and v2 performed the launch / recovery tasks in the corresponding time period, resulting in nodes g3 and g4 being occupied in a certain time period. In the current time period, vehicle v3 can only choose g1→g2→g5. The characteristics of drones in inspection operations are mainly reflected in two task execution modes ( Figure 1 (c) When a drone performs an inspection task in an aerial network, it: 1. It performs the inspection task in the aerial arc segment at a predefined inspection speed. 2. It quickly reaches the task node at the actual flight speed of the drone.

[0049] Figure 2 Example ① shows the vehicle-drone collaborative operation, where the drone The drone is carried by vehicle v1 and launched at g1 and retrieved at g2. When the drone passes through the arc segment formed by two arbitrary consecutive nodes a1→a2, the corresponding inspection task is considered completed. The vehicle can launch multiple drones at a node at one time to perform inspection tasks, such as Figure 2 The multi-drone and multi-vehicle coordinated inspection in Example ② is shown. The flexible operation of drones is introduced in Example ③ in the multiple drone operations and flexible landing. After completing the aerial node inspection task a3→a4, the a5→a6 task can be continued, reflecting the feature that the drone can perform multiple operations in one launch mission. After the vehicle v2 completes its mission at node g8, it returns to vehicle v1, which is parked at node g7, reflecting the free fall feature of the vehicle. Finally, when all the arcs in the upper network nodes are inspected by the drone, all vehicles return to the depot after retrieving the drone. The inspection task is considered completed

[0050] Based on the above description and definition, the established multi-UAV-multi-vehicle collaborative inspection path planning model under a two-layer road network is shown below.

[0051]

[0052] The optimization goal of CMDMVRP-TLRN is to minimize the maximum time it takes for a vehicle to complete a task. For vehicle v k The arrival time of returning to the warehouse after completing the task.

[0053]

[0054] Formula (2) ensures the coverage inspection constraint of each power grid line, while limiting each power grid line to be inspected by a drone at most once. Represents vehicle v k Whether the xth UAV executes the air network arc (a i →a j ) inspection tasks.

[0055] Vehicle path constraints:

[0056]

[0057] Formula (3) defines the starting path constraint of the vehicle scheduling model, and formula (4) defines the ending path constraint of the vehicle scheduling model. These two constraints together constitute the starting and ending boundary constraints of the vehicle path. Represents vehicle v k Whether it passes through the ground layer arc segment (a i →a j ).

[0058]

[0059] Formula (5) represents the access restriction of vehicles to each node.

[0060]

[0061] Formula (6) represents the balance constraint of vehicles flowing into and out of each node.

[0062]

[0063] Formula (7-8) represents the order constraint and the elimination constraint using MTZ elimination sub-ring, where pos is the order in which vehicles visit nodes. M represents an infinite number.

[0064] Vehicle time constraints:

[0065]

[0066] Formula (9) sets the vehicle v k At the time of the initial node, formula (10-11) represents the time constraint for vehicle scheduling at the ground network node. Indicates vehicle v k Leave node g i time. Represents vehicle v k Through arc segment g i →g j time. Indicates the average speed of the vehicle.

[0067]

[0068] Formula (12) expresses the vehicle v k Departure time constraint.

[0069]

[0070] Formula (13) requires that the vehicle must arrive at the node in advance to wait for and recover the drone. Indicates vehicle v k Is it at node g? i Launch the xth drone and at node g j Recycling and vehicle v m The xth drone launched.

[0071]

[0072] Formula (14) expresses the parking time constraint after the vehicle chooses to park in place; Represents vehicle v k At node g i The docking time is tl, which is the launch time of the UAV, and tr, which is the recovery time of the UAV.

[0073]

[0074] Constraint (15) ensures that the number of drones launched by any ground node must be equal to the number of drones recovered at the node, ensuring the dynamic balance between drone launch and recovery. Constraint (16) means that only when the vehicle v k When the node launches the drone x, it is allowed to be controlled by other vehicles v. mRecover at other nodes to avoid invalid cross-vehicle recovery associations. Constraint (17) states that only when a vehicle recovers a drone at a node can the drone be launched from other nodes. Constraint (18) restricts that if a drone x is launched by a vehicle v k Launched and captured by the vehicle m Recycling must meet and Represents vehicle v k Whether the x-th drone is carried from the ground network node g i Transmitted to airborne network node a i . Represents vehicle v k Is the x-th drone carried from the air network node a? j Return to ground network node g j ε represents a sufficiently small positive number.

[0075] Drone ownership:

[0076]

[0077] Formula (19) indicates that the total number of drones launched by the vehicle must not exceed the number of drones it actually carries at the node. Formula (20) indicates that the total number of drones recovered by the vehicle must not be less than the number of drones it should recover at the node. This ensures the ownership relationship between drones and the vehicles they carry at each node. It indicates whether drone x is at node g i By vehicle v k carry.

[0078] Drones and quantity restrictions:

[0079]

[0080]

[0081] Formulas (21-24) represent the constraints on the number of drones carried by the vehicle when it departs and the constraints on the number of drones carried by the vehicle after it performs the launch / recovery mission at the node; Represents vehicle v k Arriving at ground network node g i The number of drones carried at the time. It means that the vehicle leaves the ground network node g j The number of drones carried at the time.

[0082] Drone path constraints:

[0083]

[0084] Formula (25) ensures that the number of times a drone enters a node must be equal to the number of times it leaves the node. Constraint (26) links the drone's aerial movement to the ground launch operation, ensuring the rationality of the aerial path. Formula (27) ensures that the recovery operation is only for drones that actually arrive at the aerial node. Formula (28) enforces the integrity of the drone's mission through linear inequalities.

[0085]

[0086] Formula (29) expresses the endurance constraint of the UAV operation process.

[0087] Drone time calculation constraints:

[0088]

[0089] Constraint (30) ensures that the arrival time of UAV aδ is no earlier than that of vehicle v k Arrival at ground node g i The time is related to the total recovery time across vehicles. Constraint (32) ensures that the departure time of aδ follows the departure time of the launching vehicle v m The departure time, flight time and recovery time requirements. Constraint (33) requires the UAV to reach the ground node g i The time of must be synchronized with the global threshold aδ. Constraint (34) stipulates that the departure time of the UAV must be later than its arrival time.

[0090] Constraints (35-77) specify the time constraints for the UAV’s mission execution, which must take into account its possible recovery time, launch time calculation, and the restrictions on the UAV’s mission execution time after the vehicle performs multiple tasks at the same node. Constraint (38) ensures that the vehicle v k From the ground node g i The total mission time of the launched UAV x at least includes the sum of its launch, inspection and recovery flight time. Auxiliary variables aδ and dδ are used to measure the arrival / departure time of each UAV. Indicates the drone's average inspection speed.

[0091]

[0092] Formula (39-40) indicates that when two vehicles arrive at the same node, they must meet the time sequence. i The time coordination and conflict avoidance constraints of the multi-vehicle task are realized. Formula (41) represents the number of vehicles arriving at the same node at the same time. Represents vehicle v k Is the vehicle v m Leave node g before arrivingi .

[0093] The present invention designs an improved adaptive large domain search algorithm for solving related models. The designed algorithm consists of three parts: generation layer, initialization layer and optimization layer. Figure 3 As shown. In the generation layer, the node distribution of the vehicle-UAV mission, the state parameters of the vehicle and UAV, and the map scale are input to build a two-layer road network map and determine the connection relationship and distance between each node. In the initialization layer, by establishing the mapping relationship between the aerial nodes and the ground nodes, the initial path of each vehicle is generated according to the nearest neighbor algorithm, and the initial path of each vehicle is generated by the nearest neighbor algorithm. Figure 4-6 The pseudocode shown generates an initial vehicle-drone collaborative inspection task allocation scheme. Finally, at the optimization layer, this initial vehicle-drone collaborative inspection task allocation scheme is iteratively optimized by designing three targeted destruction strategies and three targeted repair strategies. The following is a detailed description of these tasks.

[0094] First, establish the cluster mapping relationship between air and ground nodes ( Figure 3 The number of clusters is set to the number of vehicles performing the task, and the vehicle inspection paths are planned based on the clustering results. This strategy avoids time conflicts caused by vehicles selecting the same ground node when generating the initial connection, reduces the difficulty of the drone's search in the aerial network, and thus accelerates the algorithm's convergence. Figure 3 The specific implementation process of pseudocode design is as follows:

[0095] First, based on the aerial network arc set A and the ground network arc set G, combined with the number of vehicles |V|, clusters are created to generate the task area cluster corresponding to each vehicle. Subsequently, a dedicated driving route (v.route) is planned for each vehicle v∈V. During the task execution phase, the system continuously monitors the status of the uninspected aerial arc set E. When there are pending tasks, the scheduling process for each vehicle is processed sequentially: for each node n in the vehicle route, it first determines whether the node is occupied at the current time vt. If so, the waiting time is updated; otherwise, the departure time is recorded and the node occupancy status is updated. The travel time (travel_time) from the current location (v.curr) to the target node n is then calculated, and the vehicle time and location status are updated simultaneously. In terms of carrying capacity control, when the vehicle's current capacity (v.current_capacity) is less than its maximum capacity (v.capacity), the drone recovery decision (DecideRecover) is executed, adjusting the system time based on the recovery flag (recover_flag). When the capacity is greater than zero, the drone launch decision (DecideLaunch) is executed, updating the time parameters based on the launch flag (launch_flag). This scheduling method achieves optimal spatiotemporal resource allocation in a multi-vehicle collaborative operation environment by dynamically coordinating key links such as vehicle path planning, node occupancy management, and drone deployment and retraction decisions.

[0096] Furthermore, before entering a node, the vehicle first calculates the expected arrival time and determines whether the target node is occupied during the current time period. If it is occupied, the vehicle will wait at the current node and update the node occupancy time; otherwise, it will go to the next node, record the departure time, and update the node occupancy status. After arriving at the target node, if the vehicle has the ability to recover the drone, it will press Figure 4 The pseudo code performs the recycling task, Figure 4 The specific implementation process of pseudocode design is as follows:

[0097] First, the recovery success flag (recover_success) is initialized to False. Then, each drone u in the set U of drones carried by vehicle v is traversed. If the drone is detected to be unlaunched or the vehicle's available capacity has reached its upper limit, the current drone is skipped. Otherwise, the vehicle's current position (v.curr) is further determined to determine whether it belongs to the drone's set of returnable nodes (u.return_nodes). If not, the next drone is detected. If so, the recovery operation is performed: the time (return_time) required for the drone to return from its current position (un) to the vehicle (v.curr) is calculated using the (calculate_time) function. The recovery success flag is set to True, the drone's current position is updated to the vehicle's node, the drone's state time parameter (ut) and launch state (u.launched) are simultaneously corrected, the vehicle's available capacity (v.available_capacity) is incremented, and the recovery node information is added to the vehicle's task list (v.task). This recovery decision mechanism ensures the feasibility of the recovery operation through dual conditional judgment and adopts a state synchronization update strategy to maintain system consistency. This achieves intelligent recovery timing determination and dynamic resource allocation in a multi-UAV collaborative operation environment.

[0098] After the recovery is completed, the vehicle will assess whether it has the conditions to launch the UAV mission and Figure 5 The pseudo code performs the launch task, and the detailed process is as follows:

[0099] The system first initializes the task success flag (launch_success) to False and clears the UAV task list (u.task). It then iterates over each UAV u in the set D of UAVs carried by vehicle v, resetting its launch status (u.launched) and restoring its remaining operating time (ue) to the rated value (ur). If there is an uninspected aerial arc set E, the system sets UAV d's current node to the vehicle's position v.curr, calculates the launch time d.launch_time, and updates the UAV's time parameter dt. Subject to the UAV's operating time capacity, the (FindNearestEdge) function dynamically retrieves the nearest edge to be inspected e from the current node, its inspection time δ, and the target node n. The remaining operating time ur is calculated in real time, and (GetReturnNodes) is called to retrieve the current set of returnable ground nodes R. If the remaining operating time is exhausted or no return node is reachable, the current task is terminated and the status flag is returned. Otherwise, the task success flag is updated, the UAV's position (u.node) and time (ut) are recorded, the inspection task δ is added to u.task, and the edge is removed from the uninspected set E. This method optimizes inspection routes through dynamic nearest neighbor search and combines a dual safety verification mechanism to ensure the drone's recyclability. This allows for autonomous mission planning and real-time status updates within a limited operating time constraint. Upon mission completion, the vehicle moves to the next node and repeats the above process until all aerial routes have been inspected.

[0100] Considering the strong dependencies between vehicle-UAV subpaths during the algorithm optimization process, and to ensure that the damaged and repaired paths meet the various performance constraints of the model, the following definitions are made to adjust for infeasible solutions in the new solution generated by the algorithm update:

[0101] Definition 1: In the destruction of vehicle paths, if the path task of the kth vehicle A node in If it is destroyed, the node is first judged based on the adjacency matrix Previous node With the next node Are they adjacent? If so, connect the node segments directly. If the nodes are not adjacent, the damaged ground path is repaired based on the nearest neighbor algorithm.

[0102] Definition 2: If there is any damaging node g n ∈N T So that vehicle k cannot be at node g n Execute the launch / recovery drone mission. If the drone is launched and recovered by the same vehicle at this node, only the aerial inspection mission and destruction node corresponding to the drone need to be deleted, such as Figure 6(a). If the UAV is not received by the original launching vehicle, then all subsequent potential tasks will be deleted at the same time as the node is deleted, including all subsequent potential tasks generated by the UAV after it is received by other vehicles. Figure 6 (b) shown.

[0103] Definition 3: If there is any damaging node g n ∈N T So that vehicle k cannot be at node g n Execute emission and node g j Perform the drone recovery mission, i.e. If the launch / recovery mission is performed by the same vehicle, that is, v k =v m , then directly delete the aerial inspection task and destruction node corresponding to the drone. The deleted task chain can be expressed as like Figure 7 (a). If the UAV x launched by node g is controlled by other vehicles v m Recycling, that is, v k ≠v m Then the vehicle v will be deleted m All potential task chains generated by launching UAV x in the subsequent process can be expressed as like Figure 7 (b) shown.

[0104] Definition 4: If the drone's aerial inspection mission is destroyed Any inspection task a i →a j Destroyed, if the destroyed inspection task is the start / end task of the aerial inspection task of drone x Modify the order of drone inspection tasks like Figure 7 Otherwise, the drone’s aerial inspection task is repaired according to the nearest neighbor method, as shown in (c). Figure 7 (d) If the number of drone aerial inspection tasks , directly delete the launch and recovery missions of this UAV, and at the same time delete the potential mission relay chain tasks according to Definition 3 to ensure that the constraints are met.

[0105] In the design of the destruction strategy, the present invention specifically designs three destruction strategies, namely, the random destruction strategy of air-ground nodes, the destruction strategy of similar nodes based on high-frequency conflicts, and the worst air-ground node destruction strategy. The following is a description of the destruction strategies:

[0106] (1) Random destruction strategy of air-ground nodes

[0107] In this destruction strategy, if the ground nodes are randomly selected Destroy, and according to Definition 2 and Definition 3, destroy the associated aerial inspection tasks and ground nodes simultaneously. i →a j If the solution is destroyed, the corresponding aerial inspection task and its potential task relay chain are deleted according to Definitions 3 and 4. This strategy effectively expands the search space of the algorithm by significantly destroying the current solution structure.

[0108] (2) Similarity destruction strategy based on high-frequency conflicts

[0109] The concentrated distribution of drone deployment tasks on adjacent nodes often leads to ground vehicle congestion and queuing in vehicle-drone collaborative scheduling tasks. This patent proposes a node optimization strategy that integrates task frequency and geographic spatial distribution characteristics, and determines the core high-frequency conflict nodes through formulas (41-42).

[0110]

[0111] Where: C(i) represents the centroid score of node i, F(i) represents the number of tasks for node i, D(i,j) represents the distance from node i to j, and W(i) represents the weighted sum of the task frequencies of adjacent nodes. γ and δ are weight coefficients used to balance the influence of task frequency and geographic location. ∈ is a small constant to prevent division by zero. Determine the centroid node with the highest score: c = argmax i After (C(i)), the similarity scores of other nodes and the center of gravity node are evaluated by formula (44). Finally, the nodes are sorted based on the node similarity scores, and task nodes are destroyed in order of priority, thereby optimizing the key areas with both high-frequency tasks and concentrated geographical locations, and improving the efficiency of overall collaborative scheduling.

[0112] In the repair strategy, in order to efficiently insert the damaged aerial inspection mission into the existing drone inspection mission or ground node (generating a new launch / recovery drone mission), the following definitions are made:

[0113] Definition 5: Launch / recovery mission for an existing drone Insert the inspection task after the damage (a i →a j ), which has three insertion methods, which can be inserted after the vehicle launch mission Before the recovery mission and drone inspection routes

[0114] Definition 6. To construct a new launch / recovery drone mission, for any vehicle v k , traverse its task path g1→g L, for the drone x it carries, obtain the ground node set where the air arc can be inserted, that is, the ground node segment where the vehicle carries the drone in a continuous path

[0115] First get the drone x in the vehicle v k All nodes that have been carried in the entire path (not necessarily continuous) To describe the UAV x in the vehicle path v k There are several index sets that carry consecutive segments. For each Define segments Where 1≤s r ≤e r ≤L is the start and end node index of the segment, which satisfies the following conditions:

[0116] 1. Carry throughout the section: From s r Starting from the node e, r During this journey, the drone x always accompanies the vehicle.

[0117] 2. No transmission task for adjacent nodes: s within the segment r ≤i≤e r There is no triggering of the drone launch mission on any arc segment.

[0118] 3. Boundary conditions (region start): When any of the following conditions are met, it is considered a new continuous carrying segment from node s r Officially starts: 1. The first node of the vehicle path starts to carry s r = 1. 2. Before this node, the drone was not attached to the vehicle. 3. In the arc segment Triggering the drone launch / recovery mission

[0119] 4. Boundary conditions (region end): When any of the following conditions is met, it is considered that this segment is "continuously carried" at node e r End. 1.e r = L means that this segment continues until the last node of the vehicle path. 2. The next node is not accompanied by the drone, i.e. 3. In the arc segment There is a drone launch mission, i.e.

[0120] The above constraints ensure that The continuous carrying section, in which each node is in the state of the drone still accompanying the vehicle. By combining such sections, we can get the complete time sequence division of the drone x "appearing with the vehicle" on the entire mission path. Where ∪ represents disjointness, meaning that the segments do not overlap in terms of node indices. Each node (if ∪ = 1) will belong to exactly one segment. If the node does not carry drone x, it does not belong to any of the above segments.

[0121] (1) Node insertion strategy based on greedy strategy

[0122] During the repair process, all removed air inspection arcs need to be reinserted into the solution space to restore the task integrity. First, based on the current vehicle-UAV collaborative path, all feasible insertion position sets S = (g n ,v k ,x,i) or S'=(g n ,v k ,x), S represents the insertion of vehicle v k At node g n The i-th mission index of launching drone x. S' represents vehicle v k At node g n Construct a new inspection task. Calculate the additional time increment ΔT generated in each feasible solution space and select the solution that minimizes the increment from all possible insertion positions. Expressed as or Repeat this operation until all damaged air inspection arcs are inserted back.

[0123] (2) Node insertion strategy based on regret value

[0124] In this repair strategy, based on the feasible insertion positions S and S' defined above, not only the direct time increment minimization is considered, but also a regret measurement mechanism is introduced to improve the robustness of the decision. Specifically, for each arc segment to be inserted, the additional time increment ΔT at each feasible position is first calculated and sorted in ascending order. Then, the time increment difference between the most feasible insertion position of the second segment and the most feasible insertion position is calculated to obtain the regret value R = ΔT second -ΔT best Finally, the insertion scheme with the largest regret value is selected, which is expressed as or

[0125] (3) Noise-based random node insertion strategy

[0126] Although the greedy insertion strategy can obtain local or potential optimal solutions to a certain extent, it is easy to fall into local optimality. By introducing random noise η, while retaining the preference for high-quality solutions, it gives inferior solutions a certain chance to be selected, thereby improving the overall solution quality. Specifically, for each air inspection arc to be inserted, first calculate its insertion position S = (g n ,v k,x,i) or S'=(g n ,v k ,x), and then the time increment is corrected by introducing random noise η through equations (44-45):

[0127]

[0128] Where: η represents the random noise assigned to the inserted task x, ΔT η This strategy uses a roulette wheel mechanism to determine the optimal insertion position based on the adjusted time cost. The probability p of each position is calculated using formula (46-47):

[0129]

[0130] In this embodiment, the Figure 8 The inspection environment task distribution diagram shown in the figure is used to verify the effectiveness of the multi-vehicle-multi-UAV collaborative inspection model designed by the present invention and the superiority of the improved adaptive large-area search algorithm. Figure 8 (ac) in the figure respectively show small-scale, medium-scale and large-scale aerial arc inspection tasks. Taking into account that the typical tower spacing is 0.3km-0.5km, 50 arc segments (covering a 30km*30km area) are set to simulate small-scale and medium-scale tasks, 100 arc segments (covering a 50km*50km area) simulate medium-scale tasks, and 200 arc segments (covering a 100km*100km area) simulate large-scale tasks. In the settings of the selected drone and vehicle parameters, the flight speed of the drone in non-inspection tasks (transferring power grid line nodes / going to task nodes / returning to vehicles) is set to 32.4km / h, and the inspection operation capacity during one mission is e=46min. In the inspection speed of the drone, the inspection operation specification standard for multi-rotor drones of overhead distribution lines (T / AOPA0053-2023) is referred to, and the inspection speed of the drone is set. In the pair Figure 8 In the test experiment, different numbers of vehicles and different vehicle-UAV ratios were selected for testing, and the optimal objective function value, the worst objective function value, the average objective function value Average Improvement Rate (AIR%) And stability indicators: To test and verify the superiority of the designed improved adaptive large neighborhood search algorithm. The test experiment table is shown in Table 1, and the experimental results are shown in Table 2.

[0131] Table 1: Examples of benchmark experiment groups

[0132]

[0133]

[0134] Table 2 Results of ALNS, VNS, and IALNS on small, medium, and large instances

[0135]

[0136]

[0137]

[0138]

[0139] As shown in Table 2, the average lower limit, average upper limit, comprehensive average value and stability value of the IALNS algorithm surpass those of mainstream algorithms such as ALNS and VNS in both small-scale and large-scale instance tasks. This shows that IALNS has advantages over similar benchmark algorithms in terms of solution quality stability and robustness. In small-scale instance scenarios, IALNS shows continuous optimization capabilities in 1-4 drone configuration schemes. Compared with other comparison algorithms (ALNS, VNS), the objective function value is reduced by an average of 4.9% and 4.78%. It is particularly noteworthy that as the number of drones increases (1-4), the algorithm objective function value decreases from 7.92 to 2.93, and the inspection efficiency is improved by 63%. This shows that in small-scale inspection tasks, (1) by adding multiple drones for collaborative inspection, the inspection time can be significantly shortened and the objective function value can be reduced. (2) The initial solution generation mechanism based on clustering-heuristic effectively guarantees the uniformity of the solution space; the inspection tasks can be evenly distributed to each drone, and a better allocation scheme is obtained in general. 3) The designed destruction / repair operator of the IALNS algorithm enhances the algorithm's local search capability within the solution space. This makes the algorithm more robust in small-scale exploration tasks and less likely to fall into local optimal solutions.

[0140] The IALNS algorithm also demonstrates significant performance advantages in medium- and large-scale inspection mission scenarios (100-200). In a medium-scale test case, compared to similar algorithms, its objective function value decreased by 13.63% and 13.31% for different vehicle-UAV configurations, respectively. When the mission scale was expanded to 200, the reductions further increased to 16.34% and 18.39%, respectively, representing an improvement of approximately three times compared to the small- and medium-scale examples. Furthermore, with the increase in inspection scale, number of vehicles, and number of UAV configurations, the algorithm's search space expands exponentially. This makes the compared algorithms prone to falling into local optima and struggles to find high-quality solutions that satisfy the constraints during the optimization process. However, thanks to the task chain destruction-repair mechanism proposed in the IALNS algorithm, the algorithm is able to efficiently search within the solution space that satisfies the constraints during the iterative optimization process of destruction and repair. Furthermore, the destroy-repair strategy based on air-ground collaboration not only ensures the efficient exploration capability of the IALNS algorithm, but also guarantees the quality of its local exploration, thereby significantly improving the algorithm's robustness. This is also the key to the IALNS algorithm's ability to consistently maintain efficient search and excellent performance in large-scale benchmark experimental tests.

[0141] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A multi-UAV and multi-vehicle collaborative inspection scheduling optimization and path planning method, characterized by: The following steps are involved: S1. Construct a two-layer road network model, including a ground layer and an aerial layer; the ground layer contains ground nodes where vehicles can drive and park and ground arcs connecting nodes, and the aerial layer contains aerial nodes that define drone inspection tasks and aerial arcs connecting nodes; S2. Based on the two-layer road network model, plan the driving paths of multiple vehicles, the driving paths including starting from the starting ground node, visiting the ground node to perform the drone launch or recovery operation, and returning to the ending ground node; S3. Planning inspection paths for multiple drones, wherein the drones are launched from a vehicle at a selected ground node, and are recovered by the vehicle at a selected ground node after performing an inspection mission along an air arc; S4. Coordinate the scheduling of vehicle and UAV mission sequences to ensure that pre-set constraints are met, including mission coverage constraints, vehicle path constraints, UAV endurance constraints, vehicle capacity constraints, launch and recovery time constraints, time synchronization constraints, and space-time occupancy conflicts of ground nodes. S5. Taking minimizing the maximum time required for all vehicles to complete their tasks as the optimization goal, the vehicle paths and UAV paths are solved through the improved adaptive large neighborhood search algorithm to generate a collaborative inspection scheduling optimization solution.

2. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to claim 1 is characterized in that: The space-time occupancy conflict constraint of the ground node introduces a binary variable to judge the order in which two vehicles arrive at the same node, ensuring that the departure time of the later arriving vehicle is later than the departure time of the earlier arriving vehicle.

3. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to claim 1 is characterized in that: The improved adaptive large neighborhood search algorithm includes: S51. Initialization step: using a clustering algorithm to assign aerial inspection tasks to different vehicles, and generating an initial path and UAV launch / recovery plan for each vehicle based on the clustering results; S52. Destruction step: selecting random destruction, similar node destruction based on high-frequency conflict, or worst node destruction strategy to remove some nodes or tasks in the current solution; S53. Repair step, using greedy strategy, regret value strategy or random noise strategy to reinsert the removed tasks and generate new feasible solutions.

4. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to claim 3 is characterized in that: In the destruction step, when the ground node is removed, the associated drone task chain is deleted simultaneously; when the aerial arc task is removed, the drone path is adjusted according to the task position or the entire task is deleted.

5. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to claim 3 is characterized in that: In the repair step, when inserting a task, the optimal insertion scheme is selected based on the cost increment or regret value of the feasible insertion position, or a probabilistic selection mechanism is adopted after correcting the cost increment by adding random noise.

6. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to claim 1 is characterized in that: The drone can continuously perform multiple aerial arc inspection missions in one launch mission, and the same drone can be recovered by different vehicles at different ground nodes.

7. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning method according to any one of claims 1-6 is characterized in that: The preset constraints also include: dynamic balance constraints on the number of drone launches and recoveries, vehicle path sub-ring elimination constraints, and drone mission integrity constraints. 8.Multi-UAV and multi-vehicle collaborative inspection scheduling optimization and path planning system, characterized by: include: A model building module, used for building the double-layer road network model; Path planning module, used to plan the vehicle's driving path and the drone's inspection path; A collaborative scheduling module that determines the mission sequence of vehicles and drones and ensures that pre-set constraints are met; The optimization solution module uses the IALNS algorithm to solve the optimization solution.

9. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning system according to claim 8 is characterized in that: The collaborative scheduling module enforces the spatiotemporal occupancy conflict constraints of ground nodes, ensuring that the same node is occupied by at most one vehicle in any time period.

10. The multi-UAV multi-vehicle collaborative inspection scheduling optimization and path planning system according to claim 8 is characterized in that: The IALNS algorithm of the optimization solution module includes an initialization submodule, a destruction submodule and a repair submodule, wherein the destruction submodule executes a destruction strategy and applies an adjustment rule, and the repair submodule executes a repair strategy and applies an insertion rule.

Images