Route planning method and system for single-vehicle recycling multi-unmanned aerial vehicle under road network restriction

Abstract

The application discloses a route planning method and system for single-vehicle recycling of multiple unmanned aerial vehicles under road network restriction, and belongs to the field of cooperative task planning of heterogeneous multi-machine systems.The application adopts a planning solution reconstruction mode of "node removal+insertion" of ALNS, and innovatively designs customized heuristic removal and insertion operations according to the characteristics of the route planning problem for single-vehicle recycling of multiple unmanned aerial vehicles under road network restriction.These customized operations contain several innovative scoring methods for removal and insertion operation candidate schemes of unmanned aerial vehicles and recycling locations, and only a few times of analysis based on the recycling results of the planning solution are needed to optimally select and execute the removal and insertion operation schemes, so that the problem that the calculation time of the removal and insertion operations of ALNS increases exponentially with the increase of the problem scale is solved, and the solution speed of the route planning problem is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of collaborative task planning for heterogeneous multi-machine systems, and more specifically, relates to a route planning method and system for recovering multiple drones under road network constraints by a single vehicle. Background Technology

[0002] The Vehicle Carrying Multi-Drones System (VCmDS) is an ideal operational mode for enabling collaborative tasks involving multiple drones across large areas. VCmDS consists of multiple drones performing the mission, and a vehicle serving as the carrier and mobile resupply station for these drones. The VCmDS mission execution process is as follows: a vehicle carrying multiple drones departs from a base, moves along a road network within the scene, and visits multiple stop points; when the vehicle arrives at a stop point, it can launch one or more drones; these drones will then proceed to a nearby target location or area to perform operations; when these drones complete their assigned tasks or are about to run out of power, they will fly to a pre-arranged stop point to rendezvous with the vehicle, where they will be retrieved and recharged; after launching this batch of drones, the vehicle will proceed to the next stop point to launch another batch of drones or retrieve any drones still in flight. During this process, the vehicle can not only flexibly retrieve drones and provide timely energy replenishment, but also carry drones for long-distance movement within the scene, conserving drone energy. In this way, VCmDS can overcome the limitations of limited endurance energy in traditional multi-UAV operations on the size of the mission area and the number of targets, and can perform large-area, multi-node missions.

[0003] CN114924593A discloses a rapid planning method for rendezvous routes between vehicles and multiple UAVs, comprising a multi-UAV-vehicle rendezvous route planning problem model under multi-time-window constraints and a two-layer heuristic planning model. The two-layer heuristic planning model has the following characteristics: 1) It includes two iterations, where the first layer optimizes the combination of "rendezvous order - rendezvous position," and the second layer optimizes both the rendezvous order and rendezvous position; 2) It includes a heuristic strategy to escape local optima, which can attempt to escape local optima by setting the rendezvous order; 3) It includes a heuristic method to generate initial solutions for "rendezvous order - rendezvous position." Considering the real-time configuration of vehicles and UAVs (initial positions of vehicles and UAVs, remaining endurance of UAVs, and available time windows, etc.), the method of this invention can quickly obtain a set of rendezvous route solutions that allow UAVs to safely return to the vehicle.

[0004] However, due to the following unique characteristics of the route planning problem for vehicle-to-drone recovery under road network constraints, this method cannot be directly used to solve VCmDS recovery routes under road network constraints: This method is only applicable to scenarios with continuous regions, where vehicles can travel freely and meet up with drones at any point for recovery; however, under road network constraints, vehicles can only travel along the road network and can only meet up with drones at road network nodes for recovery. Furthermore, this method is computationally slow and cannot meet the practical requirements of route planning under road network constraints. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a route planning method and system for recovering multiple drones by a single vehicle under road network constraints, aiming to solve the problem that the existing methods cannot meet the computational speed requirements under road network constraints.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a route planning method for recovering multiple drones under road network constraints, applicable when the original route fails, including:

[0007] S1. Acquire road network data, the current location and speed of vehicles, the current location of each unrecovered drone, the recovery time window constraints and flight speed, and construct a route planning mathematical model by integrating all data;

[0008] S2. Based on the improved adaptive large-scale neighborhood search algorithm, solve the route planning mathematical model to obtain a set of recovery routes that enable all flying UAVs to safely return to the vehicle;

[0009] In each iteration, the improved adaptive large-scale neighborhood search algorithm obtains a new solution by first randomly selecting and performing a heuristic removal operation, and then randomly selecting and performing a heuristic insertion operation.

[0010] The heuristic removal operation is any of the following: removal of the worst-rated drone, removal of the worst-rated crew, removal of a single drone, or removal from the recycling site;

[0011] The heuristic insertion operation is any of the following: adjustable drone preferred insertion, recovery location insertion, single drone insertion, or adjustable drone random insertion;

[0012] The main steps of removing multiple operational drones by evaluating the worst-performing drone are as follows: First, analyze the current solution Slt and the recovery result RS; then, based on RS, use an evaluation method for candidate removal operation schemes to score all drones; randomly select... The high-scoring drones are used to form a set of drones to be inserted. The drone to be inserted The location of the destroyed solution Slt″ is removed from the current solution Slt. n is the number of drones;

[0013] The main steps of removing multiple operational drones by the worst-performing group are as follows: First, analyze the current recovery result RS of the solution Slt; then, based on RS, score all drones using an evaluation method for removal operation candidate schemes; finally, select the drone with the highest score. Add it to the collection of drones to be inserted. Remove it and its recovery location from the current solution Slt; calculate the remaining drones in Slt relative to... The similarity was then randomly selected. Add a highly similar drone to the list. And remove them and their respective recycling locations from Slt, resulting in the corrupted solution Slt″;

[0014] The single drone removal process can only operate one drone at a time. The main steps are as follows: First, analyze the current recovery result (RS) of the solution Slt. If no drone recovery timeout occurs, randomly select a drone with a short idle time. If a drone recovery timeout occurs, randomly select a drone with a long timeout. Next, add the selected drone to the set of drones to be inserted. Remove it and its recycling location from the current solution Slt to obtain the destroyed solution Slt″;

[0015] The recovery site removal operation can handle multiple drones, but can only operate on their recovery sites. Its main steps are: first, analyze the current recovery result RS of the solution Slt; then, based on RS, use an evaluation method for candidate removal operation schemes to score the recovery sites of all drones; randomly select... For drones with high scores, set the node number of the selected drone's recovery location to 0 in Slt to obtain the destroyed solution Slt″.

[0016] The adjustable UAV preferred insertion operation mainly involves the following steps: from The process involves iteratively selecting drones with small recovery time windows. Insert it into Slt″ until a complete new solution Slt′ is obtained; in each iteration, first analyze the recycling result RS″ corresponding to the current Slt″; then, based on RS″, use an evaluation method for candidate solutions of the insertion operation to evaluate... All candidate combinations of "recycle order insertion position - recycling location" are scored; then, a set of high-scoring combinations is randomly selected, and new elements are formed accordingly. i ,p i And insert it into Slt″; finally, as needed, adjust... The recycling sequence number and recycling location after inserting Slt″ have been optimized.

[0017] The main steps of the recovery location insertion operation are as follows: First, identify all drones with invalid recovery location numbers in Slt″, and select the road network node closest to their initial position as the new recovery location for each drone; for the above drones, iteratively adjust the recovery locations in descending order according to their recovery sequence number in Slt″; the specific recovery location adjustment method is as follows: for each drone, first analyze the recovery result RS″ of Slt″, and then use the evaluation method for candidate schemes of insertion operation to score all candidate recovery locations of the drone based on RS″; select the location with the best score as the new recovery location of the drone, and modify Slt″ accordingly;

[0018] The single-drone insertion operation processes only one drone, and its main steps are as follows: First, insert the drone that needs to be processed. Insert the corrupted solution Slt″ into the current solution Slt; then, analyze the recovery result RS of the current solution Slt; next, based on RS... In the event of a timeout during recycling, appropriate adjustment methods should be selected. Adjust the recovery sequence number in the current solution SLT, and simultaneously prioritize drones. The recycling location, thus forming a new solution Slt′;

[0019] The adjustable random insertion operation of the drone mainly involves the following steps: first, cyclically starting from... The drone with the narrowest time window is selected and randomly inserted into Slt″. Then, an evaluation method for the candidate solutions of the insertion operation is used to re-optimize the drone recovery location, forming a new solution Slt′. Finally, in the new solution Slt′, the "number - recovery location node number" element of all inserted drones is adjusted [q]. i ,p i ], [q i ,p i ] represents the number of the i-th drone that met up with the vehicle and was recovered, and the corresponding road network node number of the recovery location, respectively.

[0020] Preferably, the evaluation method for the removal operation candidate schemes is to score the UAVs in Slt based on the current recovery results of the solution Slt. The higher the score of the UAV, the more likely the removal operation candidate scheme for it is to improve the recovery results of the route planning solution.

[0021] The main evaluation indicators include: the drone's timeout duration, idle time, vehicle waiting time, distance between the initial position and the vehicle's initial position, the end time of the recovery time window, the distance between its recovery location and the recovery locations of adjacent drones in the recovery sequence, and the similarity with other drones that do not time out.

[0022] Specifically, drones with the following characteristics score highly: 1) long timeout duration or short idle duration; 2) long waiting time for vehicles; 3) far initial position from vehicle initial position; 4) short recovery time window deadline; 5) long distance from adjacent drones in the recovery sequence at the recovery location; 6) low similarity to other drones that do not time out.

[0023] The evaluation method for candidate insertion operation schemes is based on the complete recovery results of the destroyed solution Slt″ before insertion and the partial recovery results of the destroyed solution Slt″ after insertion. The candidate insertion position and candidate recovery location of the UAV to be inserted in the recovery sequence of Slt″ are scored. The higher the score, the less likely it is to affect the recovery results of the planned solution and the objective function value after insertion according to this scheme.

[0024] The main evaluation indicators include: the number of drones that require vehicles to wait and the number of drones that time out of recovery after the candidate insertion position in the recovery sequence before insertion; the recovery timeout duration and idle time of the inserted drone after insertion; and the delay in the rendezvous time of the next drone after the insertion position caused by insertion.

[0025] Specifically, candidate insertion positions and candidate recycling locations with the following characteristics score highly:

[0026] 1) In the corrupted solution Slt″ before insertion, there are a large number of drones that require vehicles to wait after the candidate insertion position in the recovery sequence;

[0027] 2) In the corrupted solution Slt″ before insertion, the number of drones that timed out after the candidate insertion position is small;

[0028] 3) After inserting the UAV into the destroyed solution Slt″ according to the candidate insertion position and candidate recovery location in the recovery order, the inserted UAV does not time out and has a large idle time;

[0029] 4) After inserting the UAV into the destroyed solution slt″ according to the candidate insertion position and candidate recovery location in the recovery sequence, the rendezvous time of the next UAV after the insertion position in the recovery sequence is delayed by a small amount.

[0030] Preferably, in each iteration of the improved adaptive large-scale neighborhood search algorithm, after obtaining a new solution, if the new solution meets the acceptance condition, the current solution or the optimal solution is updated, and the updated current solution is recorded.

[0031] The new solution satisfies the following acceptance conditions: the new solution Slt′ is not recorded, and the objective function value is better than the optimal solution Slt. bIf the objective function value of the new solution Slt is lower than that of the current solution Slt, the new solution will be accepted; however, if the objective function value of the new solution Slt′ is lower than that of the current solution Slt, the probability function of simulated annealing will be used. Let ξ be a random number, f() be the objective function, and T be an auxiliary variable. If the objective function value of the new solution Slt′ is greater than or equal to 1 and the objective function value of the current solution Slt is less than 1, then... If the objective function values ​​of both the new solution Slt′ and the current solution Slt are greater than or equal to 1, then If the objective function values ​​of both the new solution Slt′ and the current solution Slt are less than 1, then T = η is the coefficient controlling the attenuation, 0 < η < 1, α o The acceptable ratio of the number of drones that have exceeded the recovery timeout limit to the total number of drones, where n is the number of drones, and x4 to c6 are the weighted parameters for the status of drones that have exceeded the recovery timeout limit. For an acceptable average timeout duration, o max The maximum acceptable timeout duration, is a coefficient.

[0032] Preferably, for removal or insertion operation i, its weight w in the (j+1)th interval i(j+1) :

[0033]

[0034] Where, θ i and π i These represent the total number of removal or insertion operations i executed and the total score in the previous iteration interval, respectively. This is the adjustment factor for the weights.

[0035] Preferably, the route planning mathematical model uses the recovery sequence number and recovery location selection of each UAV as decision variables, and vehicle driving constraints, UAV flight constraints, and recovery time window constraints of each UAV as constraints. If all UAVs do not exceed the recovery time limit, the goal is to shorten the time required to complete the recovery of all UAVs and increase the idle time of each UAV. If there are UAVs that exceed the recovery time limit, the goal is to shorten the number of UAVs that exceed the time limit, the average time limit of UAVs, and the maximum time limit of UAVs.

[0036] Preferably, the objective function of the route planning mathematical model is as follows:

[0037]

[0038]

[0039]

[0040]

[0041] Where C1 and C2 are switches for switching the objective function calculation method for two scenarios: none of the drones were recovered within the timeout period, and some drones experienced a recovery timeout period. c1 to c6 are the weight parameters of each indicator in the objective function. c The moment when the vehicle completes the recovery of all drones, u i This is the cutoff time for the recovery window of drone number i, where i represents the drone number, U is the set of drone numbers, n is the number of drones, and r is the number of drones. k When the k-th drone rendezvous with and is recovered by the vehicle, the time when the drone completes its recovery is l. qk For the time window that opens for the recovery of the kth recovered drone, u qk This is the deadline for the recovery time window of the k-th recovered drone, where k is the recovery sequence number and N is the number of drones recovered. o To recover the number of drones that have exceeded their time limit, o k The duration of the recovery timeout for the kth drone when it is successfully recovered.

[0042] Preferably, the constraints of the route planning mathematical model include:

[0043]

[0044]

[0045]

[0046]

[0047] r0 = 0

[0048] t c =r n

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062] Where n is the number of drones, m is the number of road network nodes, i is the drone ID, j is the road network node ID, k is the recovery sequence number, and x is the drone ID. ijk Let q be the decision variable. A value of 1 indicates that the k-th drone recovered by the vehicle is numbered i, and the road network node corresponding to its recovery location is numbered j. U is the set of drone numbers, and q k Let p be the serial number of the kth drone that rendezvoused with the vehicle and was recovered. k Let r0 be the road network node number of the recovery location where the k-th drone rendezvous with the vehicle, and t be a parameter agreed upon to ensure the integrity of the mathematical model. c The moment when the vehicle completes the recovery of all drones, r n The moment when the nth drone rendezvous with and is recovered by the vehicle, and when the drone completes its recovery. When the k-th drone rendezvouses and is recovered by the vehicle, the time when the vehicle arrives at the recovery location. The time required for a vehicle to travel from the recovery location of the (k-1)th recovered drone to the recovery location of the kth recovered drone is given. When the k-th drone rendezvous with the vehicle and is recovered, the time when the drone arrives at the recovery location is... For the flight time required for the k-th recovered drone to travel from its current location to its recovery location, l qk r is the opening time of the recovery time window for the k-th recovered drone. k When the k-th drone rendezvous with and is recovered by the vehicle, the time o represents the moment when the drone completes its recovery. k u represents the timeout duration for the recovery of the k-th drone upon completion of its recovery. qk e represents the end time of the recovery window for the k-th recovered drone. k b represents the idle time remaining between the end of the time window for the k-th drone and the completion of its recovery. k N represents the time N is the duration N is spent by the vehicle waiting for the drone at the recovery location when the k-th drone rendezvous and is recovered. o To recover the number of drones that have exceeded their time limit, Q iP is the recovery sequence number of the drone with the serial number i. i Let R be the road network node number where the recovery location of drone number i is located. i O is the time when the vehicle completes the task of recovering the drone numbered i. i E represents the timeout duration for the recovery of drone number i. i B represents the idle time remaining until the end of the time window when drone number i completes its recovery. i The duration the vehicle waits at the recovery site when drone number i is successfully recovered.

[0063] To achieve the above objectives, in a second aspect, the present invention provides a route planning system for recovering multiple drones under road network constraints, comprising: a processor and a memory; the memory for storing computer execution instructions; and the processor for executing the computer execution instructions, thereby causing the method described in the first aspect to be executed.

[0064] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art:

[0065] This invention provides a route planning method and system for recovering multiple drones by a single vehicle under road network constraints. It adopts the "node removal + insertion" planning and reconstruction method of ALNS (Alternating Range and Retrieval System). Based on the characteristics of the route planning problem for recovering multiple drones by a vehicle under road network constraints, it innovatively designs customized heuristic removal and insertion operations. These customized operations include several innovative scoring methods for candidate removal and insertion operations for drones and recovery locations. Only a few analyses based on the recovery results of the planning solution are needed to optimize the removal and insertion operation scheme, thereby solving the problem that the computation time of removal and insertion operations in ALNS increases exponentially with the problem size, and greatly improving the solution speed of the route planning problem. Attached Figure Description

[0066] Figure 1 This is a flowchart of a route planning method for recovering multiple drones by ordering a vehicle with road network restrictions, provided by the present invention.

[0067] Figure 2 This invention provides an embodiment of how to insert a drone during the insertion operation. Based on the combination of "recycling order insertion position - recycling location" A schematic diagram inserted into the corrupted solution Slt″.

[0068] Figure 3 This invention provides an embodiment of a preferred insertion operation for an adjustable drone, which involves adjusting the drone to be inserted. A schematic diagram illustrating the detailed method for determining the sequence number in the recovery order of the corrupted solution Slt″.

[0069] Figure 4 This is a scenario example of a vehicle recovering multiple drones (15 drones) under certain road network restrictions provided in an embodiment of the present invention.

[0070] Figure 5 This is the recovery route planning result of a scenario example of vehicle recovery of multiple drones (15 drones) under certain road network constraints provided by an embodiment of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0072] like Figure 1 As shown, this invention provides a route planning method for recovering multiple drones under road network constraints using a single vehicle. This method is applied when the original route fails, including:

[0073] S1. Acquire road network data, the current location and speed of vehicles, the current location of each unrecovered drone, the recovery time window constraints and flight speed, and construct a route planning mathematical model by integrating all the data.

[0074] The problem scenarios to which the method of this invention is applicable have the following characteristics:

[0075] Regarding environmental characteristics: there are multiple roads within the scene area; these roads are all two-way traffic lanes, forming the scene's road network; vehicles can only travel along the road network; there are multiple nodes on the road network, which are parking points for vehicles; vehicles can only stop at road network nodes, and vehicles can only meet and retrieve drones at road network nodes (i.e., the drone's retrieval location can only be a road network node); the movement of drones is not restricted by the scene environment.

[0076] Regarding initial conditions (i.e., the scene conditions at the initial moment): the vehicle is located at a road network node; the drones are scattered throughout the scene area; the initial positions of the drones and vehicles are known; the remaining battery life of each drone can be different, i.e., the end time of the recovery time window is different; there can be drones that are performing tasks, and these drones can wait to finish their current tasks before setting off to fly to the rendezvous point, i.e., the opening time of the recovery time window is different.

[0077] In terms of execution: The vehicle departs from its initial position and travels along the road network to the nodes designated as drone recovery points; each drone also leaves its initial position and flies to its designated recovery point after its recovery time window opens. Both the vehicle and the drone can remain at the recovery point to await the other's arrival. Once both have arrived at the recovery point, the drone lands on the vehicle to complete the recovery. After successfully recovering a drone, the vehicle moves to the next drone's recovery point according to the recovery order. Each drone must be recovered by the vehicle before the end of its recovery time window, i.e., before its remaining battery power is exhausted.

[0078] When using the method of the present invention, it is necessary to construct a corresponding mathematical model based on the above-mentioned characteristics of the specific problem scenario, which will serve as the input model for the method of the present invention.

[0079] The simplified assumptions are as follows:

[0080] 1) The initial positions of the drone and vehicle are known.

[0081] 2) The vehicle travels along the shortest path between the two drone recovery locations.

[0082] 3) The drone flies from its initial position to the recovery location in a straight line.

[0083] 4) The drone can only land and be recovered by the vehicle when it is located at the same road network node (recovery location) as the vehicle within its recovery time window.

[0084] 5) The landing time of the drone when it is recovered by the vehicle needs to be considered. That is, after the drone and the vehicle are at the same recovery point, a period of time is still needed for the drone to land on the vehicle before the recovery can be completed.

[0085] 6) Both vehicles and drones move at a constant speed, and all drones move at the same speed.

[0086] 7) The energy consumption of a drone's endurance is only related to flight time, and the energy consumption rate per unit time of all drones is consistent and constant.

[0087] The recovery route planning solution includes: 1) the flight path of each UAV from its initial position to its recovery location, and 2) the driving path of the vehicle to visit all UAV recovery locations. The feasible solution requires that all UAVs meet the vehicle, land, and complete the recovery within their respective recovery time windows.

[0088] The recovery route solution of the method of this invention is expressed as a combination of "recovery order - road network node number where the recovery location is located", and is defined as the planning solution Slt, which is:

[0089] Slt=[q,p]∈ID n×2

[0090] q = [q1, q2, ..., q n ]′

[0091] p = [p1, p2, ..., p n ]′

[0092] The solution Slt is an n*2 ID matrix composed of two column vectors q and p, where q represents the drone recovery order and p represents the road network node number where each drone recovery location is located. The k-th row element of the solution Slt [q...] k ,p k ] represents the number of the k-th drone that met up with the vehicle and was recovered, and the corresponding road network node number of the recovery location, respectively.

[0093] The decision variables are the formal representations of the planning solutions defined in the problem definition. The relationship between them can be expressed by the following equation:

[0094]

[0095] x ijk It can simultaneously represent the recovery sequence number and recovery location selection of a specific drone. ijk It is a 0-1 variable. When x ijk When the value is 1, it means that the number of the kth drone recovered by the vehicle is i, and the number of the road network node corresponding to its recovery location is j.

[0096] The planning objective is to find a feasible solution for a set of recovery routes in the shortest possible computation time, and to comprehensively optimize 1) the total time to complete the recovery of all drones and 2) the remaining endurance of each drone when the recovery task is completed, while satisfying the recovery time window constraints of all drones.

[0097] The known conditions include: the composition of the road network, the number of drones, their initial positions, recovery time windows, flight speeds, landing times, and the initial positions and speeds of the vehicles.

[0098] The parameters of the mathematical model are shown in Table 1.

[0099] Table 1

[0100]

[0101] The auxiliary variables of the mathematical model are shown in Table 2. The auxiliary variables mainly consist of the vehicle-UAV rendezvous and recovery results corresponding to the planning solution, and are important indicators for evaluating the quality of the planning solution.

[0102] Table 2

[0103]

[0104]

[0105] Preferably, the route planning mathematical model uses the recovery sequence number and recovery location selection of each UAV as decision variables, and vehicle driving constraints, UAV flight constraints, and recovery time window constraints of each UAV as constraints. If all UAVs do not exceed the recovery time limit, the goal is to shorten the time required to complete the recovery of all UAVs and increase the idle time of each UAV. If there are UAVs that exceed the recovery time limit, the goal is to shorten the number of UAVs that exceed the time limit, the average time limit of UAVs, and the maximum time limit of UAVs.

[0106] Preferably, the objective function of the route planning mathematical model is as follows:

[0107]

[0108]

[0109]

[0110] Where C1 and C2 are switches for switching the objective function calculation method for two scenarios: none of the drones were recovered within the timeout period, and some drones experienced a recovery timeout period. c1 to c6 are the weight parameters of each indicator in the objective function. c The moment when the vehicle completes the recovery of all drones, u i This is the cutoff time for the recovery window of drone number i, where i represents the drone number, U is the set of drone numbers, n is the number of drones, and r is the number of drones. k When the k-th drone rendezvous with and is recovered by the vehicle, the time when the drone completes its recovery is l. qk For the time window that opens for the recovery of the kth recovered drone, u qk This is the deadline for the recovery time window of the k-th recovered drone, where k is the recovery sequence number and N is the number of drones recovered. o To recover the number of drones that have exceeded their time limit, o k The duration of the recovery timeout for the kth drone when it is successfully recovered.

[0111] The objective function is designed with two parts, corresponding to the scenarios where all drones recover without timeout and where drone recovery timeouts occur, switched by variables C1 and C2. The latter part of the objective function is a penalty function for the drone recovery time window constraint, used to accelerate the search for feasible solutions; the first part is an evaluation function for feasible solutions, used to minimize the time required to recover all drones and maximize the idle time of each drone, thereby making the feasible solution tend towards overall optimality.

[0112] Specifically, if the vehicle-drone rendezvous and recovery result corresponding to the planning solution is that none of the drones timed out, then the planning solution is feasible. In this case, the evaluation function in the first half of the objective function comes into play, and the evaluation indicators are the average of the completion time of all drone recovery tasks, the drone time window usage ratio (the ratio of the time elapsed from the drone's time window opening time to the drone's recovery time to the time window width when the drone is successfully recovered), and the minimum drone idle time (the difference between the recovery time and the time window closing time when the drone is successfully recovered). If there are drones that timed out in the rendezvous and recovery result of the planning solution, then the planning solution is infeasible. In this case, the penalty function in the second half of the objective function comes into play, considering only the number of drones that timed out, the average timeout duration of the drones, and the maximum drone timeout duration. The weights of each indicator in the objective function are determined by parameters c1 to c6.

[0113] Preferably, the constraints of the route planning mathematical model include:

[0114] Constraint 1

[0115] Constraint 2

[0116] Constraint 3

[0117] Constraint 4

[0118] Constraint 5r0 = 0

[0119] Constraint 6t c =r n

[0120] Constraint 7

[0121] Constraint 8

[0122] Constraint 9

[0123] Constraint 10

[0124] Constraint 11

[0125] Constraint 12

[0126] Constraint 13

[0127] Constraint 14

[0128] Constraint 15

[0129] Constraint 16

[0130] Constraint 17

[0131] Constraint 18

[0132] Constraint 19

[0133] Where n is the number of drones, m is the number of road network nodes, i is the drone ID, j is the road network node ID, k is the recovery sequence number, and x is the drone ID. ijk Let q be the decision variable. A value of 1 indicates that the k-th drone recovered by the vehicle is numbered i, and the road network node corresponding to its recovery location is numbered j. U is the set of drone numbers, and q k Let p be the serial number of the kth drone that rendezvoused with the vehicle and was recovered. k Let r0 be the road network node number of the recovery location where the k-th drone rendezvous with the vehicle, and t be a parameter agreed upon to ensure the integrity of the mathematical model. c The moment when the vehicle completes the recovery of all drones, r n The moment when the nth drone rendezvous with and is recovered by the vehicle, and when the drone completes its recovery. When the k-th drone rendezvouses and is recovered by the vehicle, the time when the vehicle arrives at the recovery location. The time required for a vehicle to travel from the recovery location of the (k-1)th recovered drone to the recovery location of the kth recovered drone is given. When the k-th drone rendezvous with the vehicle and is recovered, the time when the drone arrives at the recovery location is... For the flight time required for the k-th recovered drone to travel from its current location to its recovery location, l qk r is the opening time of the recovery time window for the k-th recovered drone. k When the k-th drone rendezvous with and is recovered by the vehicle, the time o represents the moment when the drone completes its recovery. k u represents the timeout duration for the recovery of the k-th drone upon completion of its recovery. qk e represents the end time of the recovery window for the k-th recovered drone. k b represents the idle time remaining between the end of the time window for the k-th drone and the completion of its recovery. k N represents the time N is the duration N is spent by the vehicle waiting for the drone at the recovery location when the k-th drone rendezvous and is recovered. o To recover the number of drones that have exceeded their time limit, Q i P is the recovery sequence number of the drone with the serial number i.i Let R be the road network node number where the recovery location of drone number i is located. i O is the time when the vehicle completes the task of recovering the drone numbered i. i E represents the timeout duration for the recovery of drone number i. i B represents the idle time remaining until the end of the time window when drone number i completes its recovery. i The duration the vehicle waits at the recovery site when drone number i is successfully recovered.

[0134] Constraints 1 and 2 ensure that the decision variable x ijk The logical correctness of the constraints is as follows: Constraint 1 states that for any given drone, it can only be recovered at one recovery location; moreover, the vehicle can only perform the recovery operation for one drone at a time, meaning that when a drone is landing for recovery, other drones at the same recovery location must hover and wait for it to complete its recovery before they can begin landing. Constraint 2 states that each drone can only be recovered by the vehicle once, and only at one recovery location. Constraints 3 and 4 provide the logic based on the decision variable x. ijk This describes a method for determining the ID of the k-th recovered drone and the road network node ID at the recovery location. Constraint 5 specifies r0 to ensure the completeness of the mathematical model. Constraint 6 defines the time when the vehicle completes the recovery of all drones, i.e., the time when the vehicle completes the recovery of the last drone. Constraints 7-9 specify the times when the vehicle arrives at the recovery location. The moment the drone arrives at the recovery site And the time when the drone and vehicle are fully recovered. k The calculation method is as follows. Because the drone takes time to land during vehicle recovery, r... k Must be greater than and Constraints 10-13 specify the calculation methods for UAV recovery timeout duration, UAV idle time, vehicle waiting time for UAV, and the number of UAVs with recovery timeouts. Constraints 14-19, for a specific UAV, provide calculation methods for its recovery sequence number, recovery location, recovery timeout duration, idle time, vehicle waiting time, and other combined recovery result information.

[0135] S2. Based on the improved adaptive large-scale neighborhood search algorithm, the route planning mathematical model is solved to obtain a set of recovery routes that enable all flying UAVs to safely return to the vehicle.

[0136] The Adaptive Large-Scale Neighborhood Search (ALNS) algorithm specifically includes the following steps:

[0137] (1) Generate an initial solution Slt0 using a heuristic or random method; and use it to compare the current solution Slt and the optimal solution Slt. b Perform initialization. Then initialize the weights, scores, and execution counts for each heuristic removal and insertion operation. Initialize the iteration count i = 1.

[0138] (2) Start the i-th iteration. If i is exactly the weight update period n of the heuristic removal and insertion operation. seg Integer multiples of (i.e., weight update period n after one round of operation) seg If the operation fails, the ALNS framework's adaptive mechanism for removal and insertion operations is used to reset the weights of each removal and insertion operation and restore the operation score and execution count.

[0139] (3) Based on the weight, randomly select a removal and insertion operation respectively (the operation with higher weight is more likely to be selected); execute the operation to modify the current solution Slt and generate a new solution Slt'.

[0140] (4) If the objective function value of the new solution Slt' is better than that of the optimal solution Slt b If the objective function value is obtained, then the optimal solution Slt is updated. b =Slt'; If the new solution Slt' satisfies the acceptance condition, then update the current solution Slt = Slt' and record the updated current solution Slt.

[0141] (5) Update the number of times and scores of the removal and insertion operations selected in this iteration.

[0142] (6) If the iteration exit condition is not met, update the iteration count i = i + 1, return to step (2), and start a new round of iteration; if the iteration exit condition is met, end the iteration and output the optimal solution Slt. b This serves as a planning solution to the route planning problem of vehicle recovery of multiple drones under road network constraints.

[0143] Based on the adaptive large-scale neighborhood search algorithm, this invention innovatively designs four customized heuristic removal operations and four heuristic insertion operations; and in the above customized operations, it innovatively designs a scoring method for candidate schemes of removal and insertion operations for drones and recycling sites to shorten the computation time of removal and insertion operations; and it also innovatively designs a planning solution acceptance criterion that considers the recycling results of the planning solution.

[0144] In each iteration, the improved adaptive large-scale neighborhood search algorithm obtains a new solution by first randomly selecting and performing a heuristic removal operation, and then randomly selecting and performing a heuristic insertion operation.

[0145] The heuristic removal operation is any of the following: removal of the worst-rated drone, removal of the worst-rated crew, removal of a single drone, or removal from the recycling site.

[0146] The heuristic insertion operation is any of the following: adjustable drone preferred insertion, recovery location insertion, single drone insertion, or adjustable drone random insertion.

[0147] The main steps of removing multiple operational drones by evaluating the worst-performing drone are as follows: First, analyze the current solution Slt and the recovery result RS; then, based on RS, use an evaluation method for candidate removal operation schemes to score all drones; randomly select... The high-scoring drones are used to form a set of drones to be inserted. The drone to be inserted The location of the destroyed solution Slt″ is removed from the current solution Slt. n represents the number of drones.

[0148] The main steps of removing multiple operational drones by the worst-performing group are as follows: First, analyze the current recovery result RS of the solution Slt; then, based on RS, score all drones using an evaluation method for removal operation candidate schemes; finally, select the drone with the highest score. Add it to the collection of drones to be inserted. Remove it and its recovery location from the current solution Slt; calculate the remaining drones in Slt relative to... The similarity was then randomly selected. Add a highly similar drone to the list. Remove them and their respective recycling locations from Slt to obtain the corrupted solution Slt″.

[0149] The single drone removal process can only operate one drone at a time. The main steps are as follows: First, analyze the current recovery result (RS) of the solution Slt. If no drone recovery timeout occurs, randomly select a drone with a short idle time. If a drone recovery timeout occurs, randomly select a drone with a long timeout. Next, add the selected drone to the set of drones to be inserted. Remove it and its recycling location from the current solution Slt to obtain the corrupted solution Slt″.

[0150] The recovery site removal operation can handle multiple drones, but can only operate on their recovery sites. Its main steps are: first, analyze the current recovery result RS of the solution Slt; then, based on RS, use an evaluation method for candidate removal operation schemes to score the recovery sites of all drones; randomly select... For drones with high scores, set the node number of the selected drone's recovery location to 0 in Slt to obtain the destroyed solution Slt″.

[0151] The adjustable UAV preferred insertion operation mainly involves the following steps: from The process involves iteratively selecting drones with small recovery time windows. Insert it into Slt″ until a complete new solution Slt′ is obtained; in each iteration, first analyze the recycling result RS″ corresponding to the current Skt″; then, based on RS″, use an evaluation method for candidate solutions of the insertion operation to evaluate... All candidate combinations of "recycle order insertion position - recycling location" are scored; then, a set of high-scoring combinations is randomly selected, and new elements are formed accordingly. i ,p i And insert it into Slt″; finally, as needed, you can... The recycling sequence number and recycling location after inserting Slt″ have been optimized.

[0152] The main steps of the recovery location insertion operation are as follows: First, identify all drones with invalid recovery location numbers in Slt″, and select the road network node closest to their initial position as the new recovery location for each drone. For the above drones, adjust the recovery locations iteratively in descending order according to their recovery sequence number in Slt″. The specific recovery location adjustment method is as follows: For each drone, first analyze the recovery result RS″ of Slt″, and then use the evaluation method for candidate schemes of insertion operation to score all candidate recovery locations of the drone based on RS″. Select the location with the best score as the new recovery location of the drone, and modify Slt″ accordingly.

[0153] The single-drone insertion operation processes only one drone, and its main steps are as follows: First, insert the drone that needs to be processed. Insert the corrupted solution Slt″ into the current solution Slt; then, analyze the recovery result RS of the current solution Slt; next, based on RS... In the event of a timeout during recycling, appropriate adjustment methods should be selected. Adjust the recovery sequence number in the current solution SLT, and simultaneously prioritize drones. The recycling location thus forms a new solution Slt′.

[0154] The adjustable random insertion operation of the drone mainly involves the following steps: first, cyclically starting from... The drone with the narrowest time window is selected and randomly inserted into Slt″. Then, an evaluation method for the candidate solutions of the insertion operation is used to re-optimize the drone recovery location, forming a new solution Slt′. Finally, in the new solution Slt′, the "number - recovery location node number" element of all inserted drones is adjusted [q]. i ,p i ], [q i ,pi ] represents the number of the i-th drone that met up with the vehicle and was recovered, and the corresponding road network node number of the recovery location, respectively.

[0155] Based on the combination of "recycling order insertion position - recycling location" Will Insert into the corrupted solution Slt″. For example... Figure 2 As shown, when the drone The recycling order insertion position is At that time, the adjustable drone preferred insertion operation will... and its recycling location number Inserted in Slt″ After that position. After inserting Slt″, from Removed from the middle. When If all UAVs are inserted into Slt″, a new solution Slt′ is formed.

[0156] In the adjustable drone preferred insertion operation In the adjustment method after inserting Slt″, adjust The detailed method for the sequence number in the Slt″ recycling order is as follows: Figure 3 As shown. In Figure 3 In the middle, if the adjustable position is i, it means It can be adjusted to be inserted before the drone with serial number i.

[0157] Preferably, the evaluation method for the removal operation candidate schemes is to score the UAVs in Slt based on the current recovery results of the solution Slt. The higher the score of the UAV, the more likely the removal operation candidate scheme for it is to improve the recovery results of the route planning solution.

[0158] The main evaluation indicators include: the drone's timeout duration, idle time, waiting time required by the vehicle, distance between the initial position and the vehicle's initial position, the end time of the recovery time window, the distance between its recovery location and the recovery locations of adjacent drones in the recovery sequence, and similarity with other drones that did not time out.

[0159] Specifically, drones with the following characteristics score highly: 1) long timeout or short idle time; 2) long waiting time for vehicles; 3) far initial position from vehicle initial position; 4) short recovery time window deadline; 5) long distance from adjacent drones in the recovery sequence at the recovery location; 6) low similarity to other drones that do not time out.

[0160] The evaluation method for candidate insertion operation schemes is based on the complete recovery results of the destroyed solution Slt″ before insertion and the partial recovery results of the destroyed solution Slt″ after insertion. The candidate insertion position and candidate recovery location of the UAV to be inserted in the recovery sequence of Slt″ are scored. The higher the score, the less likely it is to affect the recovery results of the planned solution and the objective function value after insertion according to this scheme.

[0161] The main evaluation indicators include: the number of drones that require vehicles to wait and the number of drones that time out of recovery after the candidate insertion position in the recovery sequence before insertion; the recovery timeout duration and idle time of the inserted drone after insertion; and the delay in the rendezvous time of the next drone after the insertion position caused by insertion.

[0162] Specifically, candidate insertion positions and candidate recycling locations with the following characteristics score highly:

[0163] 1) In the corrupted solution Slt″ before insertion, there are a large number of drones that require vehicles to wait after the candidate insertion position in the recovery sequence;

[0164] 2) In the corrupted solution Slt″ before insertion, the number of drones that timed out after the candidate insertion position is small;

[0165] 3) After inserting the UAV into the destroyed solution Slt″ according to the candidate insertion position and candidate recovery location in the recovery order, the inserted UAV does not time out and has a large idle time;

[0166] 4) After inserting the UAV into the destroyed solution Slt″ according to the candidate insertion position and candidate recovery location in the recovery sequence, the rendezvous time of the next UAV after the insertion position in the recovery sequence is delayed by a small amount.

[0167] Preferably, in each iteration, after obtaining a new solution, the improved adaptive large-scale neighborhood search algorithm updates the current solution or the optimal solution if the new solution meets the acceptance criteria, and records the updated current solution.

[0168] The new solution satisfies the following acceptance condition: if the objective function value of the new solution Slt′ is greater than or equal to 1 and the objective function value of the current solution Slt is less than 1, then If the objective function values ​​of both the new solution Slt′ and the current solution Slt are greater than or equal to 1, then If the objective function values ​​of both the new solution Slt′ and the current solution Slt are less than 1, then η is the coefficient controlling the attenuation, 0 < η < 1, α o This represents the acceptable ratio of the number of drones that have exceeded the recovery timeout limit to the total number of drones, where n is the number of drones, and c4 to c6 are the weighted parameters for the status of drones that have exceeded the recovery timeout limit. For an acceptable average timeout duration, o max The maximum acceptable timeout duration, is a coefficient.

[0169] Preferably, for removal or insertion operation i, its weight w in the (j+1)th interval i(j+1) :

[0170]

[0171] Where, θ i and π i These represent the total number of removal or insertion operations i executed and the total score in the previous iteration interval, respectively. This is the adjustment factor for the weights.

[0172] Evaluation method for candidate removal operations

[0173] The evaluation method for candidate removal operations is used to select the drone "number - recovery location node number" element to be removed from the current solution Slt [q]. i ,p i The evaluation method assigns a score to each drone in the current solution Slt based on its recovery results. A higher score indicates a greater likelihood that the proposed removal operation will improve the recovery results of the route planning solution. Key evaluation metrics include: drone timeout duration, idle time, vehicle waiting time, distance between initial position and vehicle initial position, recovery time window deadline, distance between the drone's recovery location and the recovery locations of adjacent drones in the recovery sequence, and similarity to other drones that have not timed out.

[0174] For heuristic removal operations, two types of evaluation methods were used: evaluation method 1 and its variant, evaluation method 1.1.

[0175] Evaluation Method 1

[0176] Evaluation Method 1 evaluates the UAV's recovery sequence number or the combination of "recovery sequence number - recovery location". Based on the recovery results from the planning solution, Evaluation Method 1 comprehensively evaluates the characteristics of the UAV's recovery outcome, thereby simplifying and indirectly assessing the potential benefits of changing the UAV's recovery sequence number. Evaluation Method 1 is applied to evaluate the worst-performing UAV removal operation and the worst-performing fleet removal operation.

[0177] Specifically, for drone i, evaluation method 1 uses the following formula to score it. Wherein, The overall score for drone i These are the scoring items for the recovery results of drone i. That is the corresponding coefficient.

[0178]

[0179] In the formula, The score representing the recovery timeout duration of drone i is calculated using the following formula. The longer the timeout duration, the higher the score for this item.

[0180]

[0181] The score representing the idle time of drone i is calculated using the following formula. The shorter the idle time, the higher the score for this item.

[0182]

[0183] The score representing the vehicle waiting time for drone i is calculated using the following formula. The longer the vehicle waiting time, the higher the score for this item.

[0184]

[0185] The score representing the distance between the initial position of UAV i and the initial position of the vehicle is calculated using the following formula. The greater the distance between the initial position and the initial position of the vehicle, the higher the score for this item.

[0186]

[0187] The score for the deadline of the recovery time window for drone i is calculated using the following formula. The smaller the deadline of the recovery time window, the higher the score for this item.

[0188]

[0189] The score representing the distance between drone i and its neighboring drones in the recovery sequence at the recovery location is calculated using the following formula. The greater the distance between neighboring drones in the recovery sequence at the recovery location, the higher this score.

[0190]

[0191] This represents the similarity score between drone i and other drones that have not exceeded their timeout period, based on their initial position and recovery time window. It is calculated using the following formula: The lower the similarity with other drones that have not exceeded their timeout period, the higher this score.

[0192]

[0193]

[0194] Furthermore, in the above formula, This represents the similarity between drone i and drone j. As shown in the following formula, Includes initial position similarity Correlation with time window Two parts.

[0195]

[0196] For drone i, its reachable road network node A i and their quantity It can be defined by the following formula. The common reachable road network node A of drones i and j. ij It is then defined by the following formula.

[0197]

[0198]

[0199]

[0200] As shown in the following formula, the initial position similarity between UAV i and UAV j is... Two main factors are considered: (1) the proportion of the common reachable nodes of UAV i and j in the reachable road network nodes of UAV j; (2) the ratio of the distances of the common reachable nodes of UAV i and j to the initial positions of UAV i and j. is a coefficient.

[0201]

[0202] As shown in the following formula, the time window correlation between drone i and drone j Two main factors are considered: (1) the degree of overlap between the time windows of UAV i and UAV j if they intersect; and (2) the degree of proximity between the time windows of UAV i and UAV j if they do not intersect. and is a coefficient.

[0203]

[0204] Evaluation Method 1.1

[0205] Evaluation Method 1.1 is derived from a simplified evaluation method 1. Its purpose is to simplify and indirectly assess the potential changes in benefits caused by altering the drone's recovery location. Evaluation Method 1.1 applies to recovery location removal operations.

[0206] Specifically, for drone i, evaluation method 1.1 uses the following formula to score it. Where, The overall score for drone i The scoring items for drone i are recovery timeout duration, idle time, and vehicle waiting time, respectively, and are calculated using the following formula. That is the corresponding coefficient.

[0207]

[0208] Evaluation methods for candidate schemes of insertion operation

[0209] An evaluation method for candidate insertion operations is used to evaluate the set of drones to be inserted. The evaluation method prioritizes the recovery sequence number of the drone in the disrupted solution Slt″, or selects a new recovery location for the drone to be inserted. Based on the complete recovery results of the disrupted solution Slt″ before insertion and the partial recovery results of the disrupted solution Slt″ after insertion, the method scores the candidate insertion positions and candidate recovery locations of the drone in the Slt″ recovery sequence. Higher scores suggest a smaller impact on the recovery results and objective function value of the planned solution after insertion. The main evaluation indicators include: the number of drones requiring vehicle waiting and the number of drones with recovery timeouts after the candidate insertion positions in the recovery sequence before insertion; the recovery timeout duration and idle time of the inserted drone after insertion; and the delay in the rendezvous time of the next drone after the insertion position.

[0210] For heuristic insertion operations, four evaluation methods were used: evaluation method 2 and its variants, evaluation methods 2.1, 2.2, and 2.3.

[0211] Evaluation Method 2

[0212] Evaluation Method 2 evaluates the scheme of inserting the drone i to be inserted into the disrupted solution Slt″ during the insertion operation, i.e., the combination g = [I,P] of "recovery order insertion position - recovery location number". Evaluation Method 2 comprehensively evaluates the recovery results of the planned solution after inserting drone i and the characteristics of the recovery results of drone i itself, thereby simplifying and indirectly assessing the potential increase in the objective function value of the planned solution caused by inserting drone i according to the g insertion scheme of Slt″. Evaluation Method 2 applies to adjustable drone optimization insertion operations.

[0213] Specifically, for drone i, evaluation method 2 uses the following formula to score it. Wherein, The total score for the insertion scheme of inserting drone i into Slt″ according to g=[I,P] is... Each of the scoring items is represented separately. is a coefficient.

[0214]

[0215] The scoring criteria comprehensively consider the idle time of drone i after inserting it into Slt″. The increase in the recovery completion time of the (I+2)th UAV in the post-insertion recovery sequence. The following formula is used for calculation. After insertion, the longer the idle time of drone i is and the smaller the impact on the rendezvous time of the next drone after the insertion position, the higher the score for this item.

[0216]

[0217] In the formula, n″ represents the number of drones in Slt″ before insertion. If the insertion position I is at the end of Slt″, the idle time of drone i after insertion needs to be considered. Otherwise, we also need to consider the increase in the recovery time of the (I+2)th rendezvous drone after insertion. According to the above formula, the insertion scheme with a longer idle time and a smaller increase in the recovery time of the I+2 rendezvous drones scores higher.

[0218] The scoring criteria primarily consider the number of drones that require vehicle waiting before drone i is inserted and after position i is inserted in Slt″. The following formula is used for calculation. The more drones that require the vehicle to wait after insertion position I, the higher the score for this item.

[0219]

[0220] The scoring criteria primarily consider the number of drones that timed out before insertion of drone i, after insertion of position i in Slt″, and during which timeouts were recovered. The scoring criteria are calculated using the following formula: The more drones that time out after being inserted into the location, the higher the score for this criterion, but the lower the total score.

[0221]

[0222] In summary, the main purpose of evaluation method 2 is to select an insertion scheme for the drone i to be inserted such that before it is inserted into Slt″, there are more drones that need to wait for the vehicle after its insertion position and fewer drones that will time out during retrieval; after it is inserted into Slt″, its idle time is long and its impact on the retrieval completion time of drones after the insertion position is small.

[0223] Evaluation Method 2.1

[0224] Evaluation Method 2.1 is derived from a simplified evaluation method 2, considering only the impact of inserting UAV i into Slt″ on the idle time of UAV i, the recovery timeout duration, and the recovery completion time of UAV i after insertion. Evaluation Method 2.1 applies to adjustable UAV preferred insertion operations.

[0225] Specifically, for UAV i, evaluation method 2.1 uses the following formula to score it. The meaning and solution method of each scoring item and parameter item in the formula are consistent with evaluation method 2.

[0226]

[0227] Evaluation Method 2.2

[0228] Evaluation Method 2.2 is derived from a simplified version of Evaluation Method 2. While it also considers the idle time of UAV i after insertion into Slt″, the recovery timeout duration, and the impact on the recovery completion time of UAVs after insertion as primary evaluation indicators, Evaluation Method 2.2 differs significantly from Evaluation Methods 2 and 2.1: if the insertion position I is at the end of Slt″, then the idle time of UAV i after insertion is taken into account. The larger the better; otherwise, consider the increase in the recovery completion time of the (I+2)th recovery drone after insertion. The smaller the better. Evaluation method 2.2 applies to the adjustment method of adjustable UAV preferred insertion operation, the adjustment strategy of single UAV insertion operation, and adjustable UAV random insertion operation.

[0229] Specifically, for UAV i, evaluation method 2.2 uses the following formula to score it. The meaning and solution method of each scoring item in the formula are consistent with evaluation method 2.

[0230]

[0231]

[0232] Evaluation Method 2.3

[0233] Evaluation Method 2.3 is derived from a modification of Evaluation Method 2. It primarily addresses the scenario where UAV i has already been inserted into Slt″ and its recovery sequence number and recovery location within Slt″ need adjustment. It evaluates the adjustment scheme for the "recovery sequence insertion position - recovery location number" of UAV i within Slt″. Similar to Evaluation Method 2.2, Evaluation Method 2.3 also considers the UAV's idle time after insertion into Slt″ and its impact on the recovery completion time of UAVs after insertion as key performance indicators. However, unlike Evaluation Method 2.2, which focuses on the recovery timeout after insertion, Evaluation Method 2.3 focuses on whether the recovery timeout of UAV i changes after adjustment. Evaluation Method 2.3 applies to the recovery location insertion operation.

[0234] Specifically, for UAV i, evaluation method 2.3 uses the following formula to score it. The meaning and solution method of each scoring item in the formula are consistent with evaluation method 2.

[0235]

[0236]

[0237] Example

[0238] This embodiment uses, as follows Figure 4 The modeling and solution process of the route planning problem for vehicle recovery of multiple drones (15 drones) under certain road network constraints is shown as an example to illustrate the specific application process of the route planning method for vehicle recovery of multiple drones under road network constraints of the present invention.

[0239] The process of modeling and solving the route planning problem for vehicle-to-drone recovery under road network constraints using this invention is divided into two stages: preparation and solution. In the preparation stage, the user, referring to the problem definition in the aforementioned mathematical model, identifies the elements of the problem to be planned and analyzes whether the problem is suitable for being solved using this invention. If it is, the user prepares the required input parameters based on the mathematical model. In the solution stage, the user inputs the parameters prepared in the preparation stage into the aforementioned vehicle-to-drone recovery route planning program, which automatically solves for a feasible solution to the route planning problem for vehicle-to-drone recovery under road network constraints.

[0240] The specific process of modeling and solving the route planning problem for vehicle recovery of multiple UAVs under road network constraints using this invention is as follows:

[0241] Step (1): This is the preparation stage. Check whether the scenario of vehicle retrieval of multiple drones under road network constraints, and its retrieval route planning problem, completely matches the problem definition of the mathematical model above. If they completely match, then it is the retrieval route planning problem for the scenario of vehicle retrieval of multiple drones under road network constraints, which can be modeled and solved using this invention.

[0242] The scenario of vehicle recovery of multiple drones under road network constraints, and the problem of recovery route planning, are completely consistent with the problem definition of the above mathematical model and need to meet the following requirements:

[0243] 1) The scenario characteristics and initial conditions for vehicle recovery of multiple drones include all the scenario characteristics and initial conditions defined in the above problem;

[0244] 2) The vehicle recovery scenario involving multiple drones can be subject to all the simplified assumptions defined in the above problem definition;

[0245] 3) The planning objective and solution for the vehicle recovery route planning problem in a multi-drone scenario are consistent with the planning objective and solution defined in the above problem.

[0246] 4) The known conditions defined in the above problem definition are all known and remain unchanged in the scenario of vehicle recovery of multiple drones under road network constraints.

[0247] Step (2): This is the preparation stage. If the problem of route planning for vehicle recovery of multiple drones under road network constraints can be solved using this invention, then based on the parameter definitions in the mathematical modeling of the above mathematical model, prepare the parameters required for using this invention. The parameters that need to be prepared are shown in Table 3. Users can refer to the assignment examples in the table to design data structures and assign values ​​to each parameter.

[0248] Table 3

[0249]

[0250]

[0251] Step (3): The solution stage.

[0252] The user inputs the parameters prepared in step (2) into the above-mentioned multi-UAV route planning program for vehicle recovery under road network constraints. The program automatically solves the route planning solution for multi-UAV recovery under road network constraints (including vehicle routes and UAV routes), as well as the recovery results of each UAV corresponding to the planning solution, such as... Figure 5 As shown.

[0253] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A route planning method for recovering multiple drones using a vehicle with road network constraints, characterized in that, This method is applied to situations where the original route fails, including: S1. Acquire road network data, the current location and speed of vehicles, the current location of each unrecovered drone, the recovery time window constraints and flight speed, and construct a route planning mathematical model by integrating all data; S2. Based on the improved adaptive large-scale neighborhood search algorithm, solve the route planning mathematical model to obtain a set of recovery routes that enable all flying UAVs to safely return to the vehicle; In each iteration, the improved adaptive large-scale neighborhood search algorithm obtains a new solution by first randomly selecting and performing a heuristic removal operation, and then randomly selecting and performing a heuristic insertion operation. The heuristic removal operation is any of the following: removal of the worst-rated drone, removal of the worst-rated crew, removal of a single drone, or removal from the recycling site; The heuristic insertion operation is any of the following: adjustable drone preferred insertion, recovery location insertion, single drone insertion, or adjustable drone random insertion; The main steps for removing the worst-performing drone from the list of operable drones are: First, analyze the current solution... Results of the recycling Then, based on All drones were scored using an evaluation method for candidate removal operations; random selection was then performed. The high-scoring drones are used to form a set of drones to be inserted. ; to be inserted into the drone and its recycling locations from the current solution Remove from the middle to obtain the destroyed solution. , , Number of drones; The main steps for removing multiple operational drones from the worst-performing group are: First, analyze the current solution. Recovery results Then, based on All drones were scored using an evaluation method for candidate removal operations; the drone with the highest score was selected. Add it to the collection of drones to be inserted. and its recycling sites from the current solution Remove from; calculate The remaining drones relative to The similarity was then randomly selected. Add a highly similar drone to the list. ; and they and their respective recycling locations from Remove from the middle to obtain the destroyed solution. ; The single-drone removal method can only operate one drone, and its main steps are: First, analyze the current solution. Recovery results If no drone recovery timeout occurs, a drone with a short idle time is randomly selected. If a drone recovery timeout occurs, a drone with a long recovery timeout is randomly selected. Then, the selected drone is added to the set of drones to be inserted. and its recycling sites from the current solution Remove from the middle to obtain the destroyed solution. ; The removal of the recovery site can handle multiple drones, but can only operate on their recovery sites. Its main steps are: First, analyze the current solution... Recovery results Then, based on The evaluation method for candidate removal operation schemes was used to score all drone recovery locations; random selection was then performed. A high-scoring drone, in The node number of the selected drone's recovery location is set to 0, thus obtaining the destroyed solution. ; The adjustable UAV preferred insertion operation mainly involves the following steps: from In the iteration, drones with small recovery time window widths are selected and inserted. until a complete new solution is obtained. In each iteration, first analyze the current... Corresponding recovery results Then, according to An evaluation method for candidate insertion operations is used to score all candidate combinations of "recovery sequence insertion position - recovery location" for selected UAVs with small recovery time windows. Then, a set of high-scoring combinations is randomly selected, and corresponding new elements of "UAV number - recovery location node number" are generated and inserted into the system. In the middle; finally, as needed, insert drones with small selected recovery time windows. The subsequent recycling sequence numbers and recycling locations will be further optimized. The main steps of the recycling location insertion operation are as follows: First, find... For all drones whose recovery location numbers have expired, the road network node closest to their initial location will be selected as the new recovery location; for the aforementioned drones, based on their location... The recovery locations are adjusted iteratively in descending order of the recovery sequence number; the specific method for adjusting the recovery locations is as follows: for each drone, first analyze... Recovery results Then according to An evaluation method for candidate insertion operation schemes was used to score all candidate recovery locations for the UAV; the location with the best score was selected as the new recovery location for the UAV, and modifications were made accordingly. ; The single-drone insertion operation processes only one drone, and its main steps are as follows: First, insert the drone to be inserted into the damaged device. In the middle, restore to the current solution Then, analyze the current solution. Recovery results Next, according to In the process of handling the timeout of drone retrieval that requires insertion processing, the appropriate adjustment method is selected to address the drone requiring insertion processing in the current solution. The recycling sequence number is adjusted, and the recycling locations of the drones requiring processing are optimized to form a new solution. ; The adjustable random insertion operation of the drone mainly involves the following steps: first, cyclically starting from... Select the drone with the narrowest time window and randomly insert the corrupted solution. Then, an evaluation method for candidate solutions to the insertion operation is used to re-optimize the drone recovery location, forming a new solution. Finally, in the new interpretation In the process, we attempted to adjust the "drone number - recovery location node number" of all inserted drones.

2. The method as described in claim 1, characterized in that, The evaluation method for candidate solutions to the removal operation is based on the current solution. The recovery results, for The drones in the process are scored, and the higher the score of the drone, the more likely the candidate removal operation is to improve the recovery result of the route planning solution. The main evaluation indicators include: the drone's timeout duration, idle time, vehicle waiting time, distance between the initial position and the vehicle's initial position, the end time of the recovery time window, the distance between its recovery location and the recovery locations of adjacent drones in the recovery sequence, and the similarity with other drones that do not time out. Specifically, drones with the following characteristics score highly: 1) long timeout duration or short idle duration; 2) long waiting time for vehicles; 3) far initial position from vehicle initial position; 4) short recovery time window deadline; 5) long distance from adjacent drones in the recovery sequence at the recovery location; 6) low similarity to other drones that do not time out. The evaluation method for candidate solutions for insertion operations is based on the complete destructive solution before insertion. The recovery results and the partial destruction of the inserted solution. The recovery results, regarding the insertion of the drone in The candidate insertion positions and candidate recycling locations in the recycling sequence are scored. The higher the score, the less likely the insertion according to this scheme will affect the recycling results and objective function value of the planning solution. The main evaluation indicators include: the number of drones that require vehicles to wait and the number of drones that time out of recovery after the candidate insertion position in the recovery sequence before insertion; the recovery timeout duration and idle time of the inserted drone after insertion; and the delay in the rendezvous time of the next drone after the insertion position caused by insertion. Specifically, candidate insertion positions and candidate recycling locations with the following characteristics score highly: 1) Destroyed solution before insertion In the process, there are a large number of drones that require vehicles to wait after the candidate insertion position in the recovery sequence; 2) Destroyed solution before insertion In the process, the number of drones that timed out after being inserted into the candidate recovery sequence position was small; 3) Insert the UAV into the damaged solution at the candidate insertion position and candidate recovery location according to the recovery order. Afterwards, the inserted drone does not exceed the timeout limit and has a large idle time; 4) Insert the UAV into the damaged solution at the candidate insertion position and candidate recovery location according to the recovery order. Afterwards, the rendezvous time of the next drone after the recovery sequence insertion position is delayed by a small amount.

3. The method as described in claim 1, characterized in that, In each iteration, the improved adaptive large-scale neighborhood search algorithm obtains a new solution. If the new solution meets the acceptance criteria, it updates the current solution or the optimal solution and records the updated current solution. The new solution satisfies the following acceptance conditions: when the new solution... It was not recorded, and the objective function value is better than the optimal solution. Or the current solution When the objective function value is reached, the new solution will be accepted; while when the new solution... The objective function value is worse than the current solution. When determining the objective function value, the possibility function of simulated annealing is used. To control the acceptance of new solutions , It is a random number. Let be the objective function. As an auxiliary variable, where, If a new solution The objective function value is greater than or equal to 1 and the current solution If the objective function value is less than 1, then , ; If a new solution and the current solution If the objective function values ​​are all greater than or equal to 1, then , ; If a new solution and the current solution If the objective function values ​​are all less than 1, then , ; To control the attenuation coefficient, 0 < <1, For the number of iterations, The acceptable percentage of drones that have exceeded their recovery time limit relative to the total number of drones. For the number of drones, To provide the weight parameters of each metric in the objective function for the timeout period, For an acceptable average timeout duration, The maximum acceptable timeout duration, is a coefficient.

4. The method according to any one of claims 1 to 3, characterized in that, The route planning mathematical model uses the recovery sequence number and recovery location selection of each UAV as decision variables, and vehicle driving constraints, UAV flight constraints, and recovery time window constraints of each UAV as constraints. If all UAVs do not exceed the recovery time limit, the goal is to shorten the time required to complete the recovery of all UAVs and increase the idle time of each UAV. If there are UAVs that exceed the recovery time limit, the goal is to reduce the number of UAVs that exceed the time limit, shorten the average time limit of UAVs, and minimize the maximum time limit of UAVs.

5. The method as described in claim 4, characterized in that, The objective function of the route planning mathematical model is as follows: in, and To switch the objective function calculation method between two scenarios: when all drones have not been recovered and the timeout has occurred, and when some drones have experienced a recovery timeout. These are the weight parameters for each indicator in the objective function. The moment when the vehicle completes the recovery of all drones, For the number The deadline for the drone's recovery time window. Number the drone. A set of drone ID numbers, For the number of drones, For the first The time when the drone rendezvous with and is recovered from the vehicle, and the moment when the drone completes its recovery. For the first The opening time of the recovery window for the recovered drone. For the first The deadline for the recovery window of the recovered drone. For the recycling serial number, To recover the number of drones that have exceeded their time limit, For the first The duration of the recovery timeout when the drone is successfully recovered.

6. The method as described in claim 4, characterized in that, The constraints of the route planning mathematical model include: in, For the number of drones, The number of road network nodes. Number the drone. Numbering road network nodes For the recycling serial number, For decision variables, a value of 1 indicates that the first... The drone recovered by the vehicle was numbered as follows The road network node corresponding to its recycling location is numbered as follows: , A set of drone ID numbers, For the first The serial number of the drone that was recovered after being rendezvoused with the vehicle. For the first The road network node number of the drone recovery location where the drone is retrieved after meeting with the vehicle. The parameters agreed upon to ensure the integrity of the mathematical model. The moment when the vehicle completes the recovery of all drones, For the first The time when the drone rendezvous with and is recovered from the vehicle, and the moment when the drone completes its recovery. For the first When the drone rendezvous with and is retrieved from the vehicle, the time when the vehicle arrives at the retrieval location... For the vehicle from the The drone was taken from the recovery location and arrived at the recovery point. The journey to the recovery site for the recovered drone takes time. For the first When the drone rendezvous with the vehicle and is recovered, the time when the drone arrives at the recovery location... For the first The flight time required for a recovered drone to travel from its current location to its recovery point. For the first The opening time of the recovery window for the recovered drone. For the first The time when the drone rendezvous with and is recovered from the vehicle, and the moment when the drone completes its recovery. For the first The duration of the recovery timeout when the drone completes its recovery. For the first The deadline for the recovery window of the recovered drone. For the first The idle time between the drone's completion of recovery and the end of its time window. For the first When a drone rendezvous with and is retrieved from a vehicle, the duration the vehicle waits for the drone at the retrieval location is as follows: To recover the number of drones that have exceeded their time limit, For the number The drone's recovery serial number, For the number The road network node number where the drone's recovery location is located. Complete the process for vehicle numbered The timing of the drone recovery mission For the number The timeout duration for drone recovery For the number The idle time between the drone's completion of recovery and the end of its time window. For the number When the drone is successfully recovered, the vehicle waits at the recovery site for a certain period of time. This refers to the time required for the drone to land and be recovered when the vehicle and the drone meet.

7. A route planning system for recovering multiple drones using a road network-restricted order vehicle, characterized in that, include: Processor and memory; The memory is used to store computer-executed instructions; The processor is configured to execute the computer execution instructions, causing the method described in any one of claims 1 to 6 to be executed.

Images