Highway intelligent inspection path planning method based on cooperation of inspection vehicle and multiple unmanned aerial vehicles

Abstract

The invention relates to the technical field of intelligent traffic inspection, in particular to an intelligent expressway inspection path planning method based on inspection vehicle-multi-unmanned aerial vehicle coordination, which realizes path optimization through data acquisition, model construction, improved ALNS algorithm solution and scheme execution whole process based on an inspection vehicle-multi-unmanned aerial vehicle coordination architecture. The method comprises the following steps: firstly, collecting parameters of a road section, an inspection point and equipment, constructing a joint optimization model with minimum total operation completion time as a target, and embedding inspection point assignment, endurance, a time window and vehicle-machine cooperative constraints; an improved ALNS algorithm is adopted, solving is carried out through a dynamic penalty coefficient, a multi-type damage / repair operator and a self-adaptive mechanism, and an in-situ / off-site take-off and landing mode of the unmanned aerial vehicle is supported. Through a complete research chain from problem modeling, algorithm design to experimental verification, related models, algorithms and conclusions have important theoretical significance and application reference value for promoting the development of an intelligent inspection technology.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic inspection technology, specifically to a method for intelligent highway inspection path planning based on the collaboration of inspection vehicles and multiple unmanned aerial vehicles. Background Technology

[0002] With the continuous expansion of the expressway network and the high traffic volume over the years, expressway infrastructure as a whole has entered a stage of intensive maintenance and high incidence of defects. Safety risks such as aging roadside facilities, slope instability, and bridge structural defects are characterized by numerous locations, long routes, wide distribution, strong concealment, and rapid evolution. On the one hand, early defects often manifest as small cracks, local deformation, and minor erosion, which are not easily identified through macroscopic observation during driving. On the other hand, expressways have high vehicle speeds and dense traffic, and inspection personnel are subject to strict safety controls when working on the road. Traditional inspection methods, which mainly rely on manual visual inspection and sampling surveys, have significant shortcomings in terms of operation frequency, spatial coverage, and precision, resulting in low inspection efficiency, high operational risks, and persistently high labor and traffic organization costs, making it difficult to detect and address early hidden dangers in a timely manner.

[0003] To improve inspection accuracy and work efficiency, UAV technology has been gradually introduced into highway operation and maintenance scenarios in recent years. However, based on existing engineering practices and related research, such applications still have the following prominent problems: (1) The scale of UAV operations does not match the task requirements. Most systems only support the inspection of a single or a small number of UAVs on local road sections and small-scale facilities, mainly targeting single targets such as bridges, tunnel portals or typical slopes. It is difficult to coordinate and plan the inspection tasks of multiple types and multiple locations on long-distance highway sections in a single task. (2) The UAV has limited endurance and is not capable of continuous inspection work under long-distance and complex road network conditions. Due to the constraints of battery capacity, payload capacity, and environmental wind field, the range that a single UAV can cover in one take-off and landing cycle is limited. If it relies only on fixed take-off and landing points or a single base station, it must frequently return to resupply, which increases the number of take-offs and landings and the difficulty of organization, and also significantly limits the effective length of the inspectable road sections. (3) UAV scheduling under complex constraints still relies on experience or simple rules. In real-world highway operation environments, drones are simultaneously subject to multiple constraints, including endurance, mission time windows, safe airspace for operations, and no-fly or altitude-restricted areas. These constraints exhibit significant spatiotemporal coupling characteristics. Existing scheduling methods largely rely on operator experience or simple heuristic rules for manual route planning and task allocation, lacking system optimization models and algorithms to address complex constraints. Consequently, it is difficult to guarantee near-optimal or suboptimal collaborative operation solutions from a global perspective. Summary of the Invention

[0004] The purpose of this invention is to provide a method for intelligent highway inspection path planning based on the collaboration of inspection vehicles and multiple drones, so as to solve the problems of low efficiency and poor adaptability to multiple constraints in existing highway inspections mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent highway inspection path planning based on patrol vehicle-multi-UAV collaboration includes the following steps: S1. Obtain the set of alternative stop points for the target section of the highway. Inspection point assembly and core device parameters, including the maximum flight range of the drone. ; S2. Construct a joint optimization model coupling the selection of inspection vehicle docking points and driving routes, and the allocation of multiple UAV tasks and flight paths, with the total operation completion time as the starting point. With minimization as the objective, and incorporating constraints such as UAV endurance, inspection point time windows, and vehicle-machine collaboration timing, the core objective function of the joint optimization model is: ; in, , , For drones, For the inspection vehicle to reach the alternative parking point At that moment, For inspection vehicles or drones at alternative parking points The time spent at the place, For drones The time when you return to the parking spot after completing the task; S3. The improved adaptive large neighborhood search ALNS algorithm is used to solve the joint optimization model to obtain the collaborative inspection path scheme of the inspection vehicle and multiple UAVs; S4. Output the collaborative inspection path plan to guide the inspection vehicle and multiple drones to complete the highway inspection.

[0006] Preferably, in step S1, the information of the target road segment also includes the starting mileage marker and the ending mileage marker; the set of alternative stop points The set of inspection points is a discrete set of locations along the highway that meet safety, spatial, and communication requirements; This is a collection of key inspection facilities or potential risk points along the highway. Each inspection point has precise spatial coordinates, work type, estimated work duration, and acceptable time window. The core parameters of the equipment also include the drone's flight speed and the inspection vehicle's travel speed.

[0007] Preferably, the set of alternative stops This includes dedicated work areas near highway service areas, ramp triangle areas or ramp merging / departure areas, emergency parking lanes, operation and management sub-centers, maintenance work area entrances, and temporary work areas designated according to regulations.

[0008] Preferably, in step S3, the UAV's flight path supports both in-situ take-off and landing mode and out-of-situ take-off and landing mode: the in-situ take-off and landing mode allows the UAV to land at the same alternative docking point. Complete takeoff, inspection, and landing tasks; the remote takeoff and landing mode refers to the UAV landing at an alternative docking point. After takeoff and completion of the inspection mission, the vehicle will proceed to the designated alternative docking point already in place. landing, .

[0009] Preferably, the constraints of the joint optimization model include inspection point assignment constraints, inspection vehicle flow balance constraints, and UAV flow balance constraints, as detailed below: (1) Constraints on the assignment of inspection points: ,in Indicates drone From the inspection point Fly to The value is 1 if it is true, and 0 otherwise. (2) Inspection vehicle flow balance constraints: ; ; ,in This indicates that the inspection vehicle will depart from the alternative parking point. Drive to the alternative parking spot The value is 1 if it is true, and 0 otherwise. To remove the set of alternative parking spots for the destination, To remove the set of alternative parking spots from the starting point, To remove the set of alternative parking spots from the origin and destination; (3) Unmanned aerial vehicle (UAV) flow balance constraints: .

[0010] Preferably, the UAV endurance constraint expression of the joint optimization model is: ,in Indicates drone Arrival at the inspection point The cumulative flight distance over time, and , For point and The Euclidean distance between them The constant is sufficiently large; the time window constraint expression for the inspection point is: ,in For drone inspection The starting moment of point.

[0011] Preferably, the vehicle-machine cooperative constraints of the joint optimization model are as follows: (1) Starting point constraints for UAVs: ; ; (2) UAV endpoint constraints: ; ; (3) Time transit constraints: ; ; ; ; ; in, For drones travel time of the segment For drones at inspection points The time required for inspection For the inspection vehicle The travel time of the segment.

[0012] Preferably, in step S3, the improved ALNS algorithm handles the range constraint and time window constraint through a constraint relaxation and penalty function mechanism. The relaxed objective function is: in, This is a penalty item for power constraints. The time window constraint penalty term, and This is a dynamic penalty coefficient; The power constraint penalty item ; The time window constraint penalty item ; The dynamic penalty coefficient is updated according to an exponential decay strategy: , ,in As the attenuation factor, , .

[0013] Preferably, in step S3, the destruction operator of the improved ALNS algorithm includes at least one of the following: maximum flight distance contribution point removal operator, time window urgency-guided removal operator, high-load UAV path overall removal operator, and random removal operator. The improved ALNS algorithm's repair operators include at least one of the following: minimum cost greedy insertion operator, regret-first insertion operator, and time window urgency-first insertion operator.

[0014] Preferably, in step S3, the adaptive process of the improved ALNS algorithm includes a multimodal scoring mechanism, dynamic weight update, and simulated annealing acceptance criterion. The improved ALNS algorithm generates its initial solution by: distributing inspection points evenly among multiple drones based on the number of drones and the number of candidate docking points, while ensuring maximum endurance. Under the premise of mapping to adjacent alternative docking point intervals, the internal inspection points of each UAV are randomly connected to form a path, and combined to obtain a feasible initial solution; the operation content types of the inspection points include taking pictures, video acquisition, and laser scanning; the time window of the inspection points It is determined by traffic environment, lighting conditions, operation and management requirements, and safety regulations.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) The ALNS algorithm constructed in this invention forms a logically rigorous and structurally complete heuristic optimization framework through the organic combination of constraint relaxation and penalty mechanisms, multi-strategy operator design, adaptive weight adjustment, and solution acceptance mechanism. This framework not only effectively addresses the special challenges of the collaborative inspection path planning problem between inspection vehicles and UAVs, but also demonstrates good convergence performance and solution quality, providing valuable solution ideas for collaborative task planning problems in complex environments. This invention systematically studies the path planning problem under the collaborative inspection mode of inspection vehicles and UAVs. First, based on an in-depth analysis of the actual needs and operation scenarios of highway inspection, the necessity and advantages of the collaborative mode are clarified, namely, using inspection vehicles as mobile bases to overcome the bottleneck of UAV endurance, and using UAVs to reach inspection points that are difficult for humans to reach. Through mathematical modeling and comparative analysis of the two operation modes of UAVs, namely, taking off and landing in place and taking off and landing in different places, the superiority of the taking off and landing mode in minimizing the total operation time is theoretically proven, thus laying the core decision foundation for subsequent model construction.

[0016] (2) This invention constructs a hybrid integer programming model integrating inspection vehicle point selection and UAV path planning. The model takes minimizing the total completion time as its core objective, while integrating the maximum endurance of the UAV, the time window constraint of the inspection point, the flow balance between the inspection vehicle and the UAV, and the complex constraints of the complex spatiotemporal coupling relationship. However, given the NP-hard nature of this model, directly using an exact algorithm to solve it faces a huge challenge. This invention designs and implements an improved adaptive large neighborhood search algorithm. The innovation of this algorithm lies in the introduction of constraint relaxation and dynamic penalty function mechanisms, which transform the hard constraints in the original problem into penalty terms in the objective function, effectively enhancing the algorithm's search capability and convergence stability in complex solution spaces. The core of the algorithm includes a set of destruction operators and repair operators designed for the characteristics of the problem, and dynamically selects operators through an adaptive weight adjustment mechanism based on historical performance feedback, thereby intelligently balancing global exploration and local development capabilities. In addition, the embedded simulated annealing acceptance criterion provides an effective way for the algorithm to escape local optima.

[0017] In summary, this invention, through a complete research chain from problem modeling and algorithm design to experimental verification, has successfully established a set of theoretical methods and practical solutions for solving the collaborative path planning problem of inspection vehicles and drones. The relevant models, algorithms, and conclusions have important theoretical significance and application reference value for promoting the development of intelligent inspection technology. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are explained in detail together with the embodiments of the invention, but do not constitute a limitation thereof.

[0019] Figure 1 This is a flowchart illustrating the core architecture of the inspection path planning method based on the collaboration of a single inspection vehicle and multiple drones in this invention. Figure 2 This is a schematic diagram of the initialization process of the present invention; Figure 3 This is a framework diagram of the ALNS algorithm of this invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention proposes an inspection path planning method based on the collaboration of a single inspection vehicle and multiple UAVs. This method constructs a joint optimization model and employs an improved Adaptive Large Neighborhood Search (ALNS) algorithm to minimize the overall inspection task completion time. Under complex conditions such as long linear distances on highways, densely distributed inspection points, limited UAV endurance, and strictly limited parking spots, this invention can achieve global optimization and efficient execution of the task.

[0022] like Figure 1 As shown, the entire scheme involves constructing a highway inspection model and then using a heuristic algorithm to optimize and solve the path scheme for the inspection vehicle and the drone.

[0023] This invention is achieved through the following technical solution. In long-distance highway inspection scenarios, this invention uses a highway inspection vehicle as a ground mobile command center and a UAV take-off and landing supply platform, and multiple UAVs as aerial inspection execution units. Through integrated joint optimization of the inspection vehicle's stopping points and travel routes, the allocation of inspection tasks and flight paths for multiple UAVs, and the vehicle-UAV collaborative timing relationship, the total completion time of the entire highway section inspection task is minimized while meeting multiple constraints such as UAV endurance, inspection point time windows, and safe operation standards. Specifically, the target highway section extends from the starting kilometer marker to the ending kilometer marker. Alternative stopping points where the inspection vehicle is allowed to make brief stops include service areas, toll stations, management stations, emergency parking areas near ramps, and emergency operation areas approved by the maintenance unit. During system operation, the inspection vehicle travels at a safe speed along the main highway or emergency lane, only briefly stopping at a few alternative stopping points selected by the optimization model and activating them as actual stopping points. At these stopping points, the drones complete take-off, landing, charging / battery swapping, and mission data transmission. Multiple drones use the inspection vehicle as a mobile base, taking off from a certain actual stopping point and sequentially visiting several inspection points assigned to each drone according to the planned path. After completing tasks such as taking photos, videos, or laser scanning, they can choose to return to the original stopping point to land, or they can meet up with the inspection vehicle at another actual stopping point ahead and land. Thus, two collaborative operation modes, namely in-situ take-off and landing and remote take-off and landing, are naturally formed in the model. If the drone arrives at the designated docking point before the inspection vehicle, it will hover in a safe airspace above or near the docking point until the inspection vehicle arrives. If the inspection vehicle arrives first, it must wait for all drones associated with that docking point to complete their operations and return before it can set off again. The entire process will continuously cycle along the linear section of the highway until all inspection points have been completed and all drones have been safely recovered, at which point the inspection mission will be completed.

[0024] Given the numerous variables, complex constraints, and NP-hard nature of the joint path planning problem, directly using a general mixed-integer programming solver is insufficient to obtain high-quality solutions on an engineering scale. Therefore, this invention proposes an adaptive large-scale neighborhood search (ALNS) solution method adapted to highway inspection scenarios at the algorithm level. The overall process is as follows: Figure 3 As shown, the algorithm first generates a feasible initial solution based on the scale and equipment configuration of the actual highway inspection task. This initial solution can be quickly obtained by a solver within a given time, or it can be generated by a random construction strategy based on the number of drones and the number of candidate stopping points. Specifically, the inspection points are roughly evenly distributed among multiple drones according to their order or spatial location. Under the premise of satisfying the maximum inspection distance of each drone, these task segments are mapped to adjacent candidate stopping point intervals, thus obtaining a set of points where the inspection vehicle may stop. Then, the inspection points within each drone are randomly connected to form a complete path. The combination of the inspection vehicle path and the multiple drone paths yields a feasible joint operation scheme. Although this initial scheme may be relatively coarse in terms of objective value, it has the advantages of being able to generate quickly at any scale and strictly satisfying basic hard constraints, providing a starting point for subsequent large neighborhood search.

[0025] During the iterative optimization process, the ALNS algorithm iteratively executes a process of destruction, repair, acceptance, and adaptive update around the current solution. In the destruction phase, the algorithm utilizes various destruction operators designed specifically for the characteristics of highway inspection to remove some inspection point assignments or cancel some inspection vehicle parking points from the current joint operation plan. For example, the maximum flight distance contribution point removal operator calculates the flight distance contribution value of each inspection point in the UAV path and prioritizes the removal of far-end nodes that cause significant additional flight costs, thereby alleviating the endurance bottleneck; the time window urgency-guided removal operator calculates the urgency based on the time window margin of each inspection point and prioritizes the removal of nodes with higher time window risks, creating time leeway for subsequent rescheduling; the high-load UAV path overall removal operator identifies the UAV with the longest task completion time and removes all inspection points in its path to release the bottleneck task affecting the overall completion time; and the random removal operator randomly removes some inspection points or parking points to increase search diversity. After the destruction operation, an incomplete partial solution is obtained. The repair phase then begins, utilizing different styles of insertion operators to re-insert the removed inspection points and docking points into the inspection vehicle and drone paths: the minimum-cost greedy insertion operator selects the optimal insertion position for each node based on minimizing the objective function increment, quickly generating a high-quality feasible solution; the regret-priority insertion operator calculates regret values ​​based on evaluating multiple candidate insertion positions, prioritizing critical nodes whose failure to insert in time would significantly increase future costs, thus overcoming the shortsightedness of a pure greedy strategy; the time-window urgency-priority insertion operator sorts the nodes according to their time window urgency, prioritizing the placement of inspection points with the smallest time window margin, and minimizing overall cost while ensuring time window feasibility, which is particularly important for improving the feasibility and robustness of the solution. The destruction and repair operators are selected in each iteration using a roulette wheel approach, with the selection probability dynamically adjusted based on the operator's historical performance, allowing the algorithm to adaptively favor more effective operator combinations for the current stage.

[0026] To further improve convergence stability under complex constraints, this invention introduces a constraint relaxation and penalty function mechanism into the optimization model for the UAV endurance constraint and the inspection point time window constraint. This transforms the original hard constraints into soft constraint penalty terms in the objective function, allowing for slight violations in flight distance or time window in intermediate iterative solutions. However, by superimposing a power penalty term and a time window penalty term based on the squared excess penalty in the objective function, a cost is incurred for exceeding the limits. The more the limits are exceeded, the greater the penalty, thereby expanding the connectivity of the explorable solution space in the early stages of the search and avoiding search breaks caused by hard constraints. Then, driven by the penalty, the algorithm gradually approaches the feasible region. The penalty coefficient is adaptively adjusted according to an exponential decay strategy as the iteration progresses. A larger penalty coefficient is used in the early stages of the algorithm to strengthen the suppression of severely infeasible solutions; as the number of iterations increases, the penalty coefficient gradually decreases, allowing the algorithm to briefly explore solutions that slightly violate the constraints while maintaining the overall feasible trend, in order to escape local optima. Meanwhile, this invention introduces simulated annealing based on the Metropolis criterion into the solution acceptance mechanism. Degraded solutions are accepted probabilistically; the smaller the difference in target values ​​between new and old solutions and the higher the current temperature, the greater the probability of acceptance for a degraded solution. This maintains sufficient global exploration capability in the early stages of the algorithm, and shifts to local fine-tuning search as the temperature gradually decreases in the later stages, achieving a balance between global exploration and local convergence. A graded scoring mechanism based on solution quality quantifies and scores the performance of each operator combination in each iteration. The highest reward is given for generating the globally optimal solution, a medium reward for generating a feasible solution better than the current solution, and a basic reward for generating a slightly inferior but accepted solution. Based on this, the weights of each operator are updated through an exponential smoothing model, ensuring that operators with good recent performance are called more frequently in subsequent iterations, forming an operator selection mechanism with memory and self-adjustment capabilities.

[0027] The main parameters and variable definitions of the model are shown in the table below. It is assumed that there are several alternative parking points along the inspection route. All alternative parking spots constitute a set. The collection of homogeneous drones carried on the inspection vehicle is denoted as... The set of drone inspection task points is denoted as... For a single drone, its maximum single-charge endurance is denoted as... Based on the parameter definitions, the problem is described as follows: The inspection vehicle passes through the selected parking points 1 to 2 along a predetermined track. The drone will be deployed at certain alternative parking locations. If the patrol vehicle continues to operate at the parking location... and parking spots If docked, If the testing point In the interval The drones released from within are to be inspected. If testing is performed, then After the drone is released within the designated area, it must be inspected along the specified path. If the drone... Successive nodes and nodes To access, .

[0028] The optimization objective of this model is to minimize the total task completion time. The task completion time is defined as the maximum value of the completion time of the inspection vehicle and the drone.

[0029] (1) Constraints: (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (twenty one) (twenty two) (twenty three) (twenty four) (25) Equations (1), (2), and (3) indicate that the optimization objective is to minimize the total flight time of the UAV, i.e., to minimize the total operation cycle. Firstly, there are assignment constraints, mainly including the assignment of each inspection point to a specific UAV and the distance each inspection point should be inspected along. Equation (4) indicates that all off-line inspection points are inspected by one and only one UAV. Equations (5), (6), and (7) collectively represent the flow balance of the inspection vehicle: Equation (5) indicates that after stopping at an intermediate stop, the inspection vehicle will inevitably travel to the next stop; Equation (6) indicates that the inspection vehicle needs to start from the starting station; Equation (7) indicates that the inspection vehicle needs to return to the destination station. Equation (8) indicates that after inspecting a certain off-line inspection point, the UAV will inevitably fly to the next point. Equations (9) and (10) collectively indicate that if the inspection vehicle... If you tap to stop, the drone can then choose to fly from... The point of takeoff is the constraint of the drone's starting point for completing the inspection task. Equations (11) and (12) together indicate that if the inspection vehicle is in If the drone stops, it can choose to land after completing its inspection mission. The point is the endpoint constraint of the UAV. (13) indicates that any UAV will only inspect once within the time period and return to the inspection vehicle to charge after completing the task. Equation (14) indicates that in The specific calculation formula for the flight distance of the point. Equation (15) indicates that the flight distance of the UAV performing one inspection mission shall not exceed the maximum range of the UAV after a single full charge. Coupling constraint: Equation (16) indicates that if the rail inspection vehicle stops at a certain alternative parking point, it shall be selected as the parking point and the UAV may take off at this point. Equation (17) indicates that if a UAV lands at an alternative parking point, that point shall be selected by the inspection vehicle for parking. Equations (18) and (19) together indicate the coupling relationship between variables y and z, that is, if the UAV takes off or lands at a certain point of the alternative parking point, that alternative parking point shall be selected as the parking point. Equation (20) indicates that each inspection point Service hours must be within the time window Equations (21), (22), and (23) represent the time transfer for UAV takeoff, between inspection points, and landing, respectively. That is, the time of arrival at the later point is the time of arrival at the earlier point plus the flight time between the two points and the inspection time at the earlier point. Equation (24) represents the path time transfer for the inspection vehicle. That is, the time of arrival at the later point is the time of arrival at the earlier point plus the travel time between the two points and the waiting time at the earlier point. Equation (25) represents the calculation method for the waiting time of the inspection vehicle waiting for the UAV at the parking point.

[0030] This invention addresses an NP-hard problem, which can be viewed as a variant of the Traveling Salesman Problem with drones, adding an additional synchronization constraint between the patrol vehicle and the drone. While commercial solvers like Gurobi can handle small-scale examples, solving large-scale examples becomes computationally difficult. Therefore, it is necessary to develop an efficient heuristic algorithm capable of providing high-quality solutions in a reasonable time. To this end, an Adaptive Large Neighborhood Search (ALNS) algorithm is proposed to handle large-scale instances. After designing and formulating appropriate destruction and repair operators, the algorithm converges quickly and can escape local optima, achieving excellent results in a finite number of iterations.

[0031] In the specific scenario of collaborative patrol vehicle and drone operation, a disruption operator and a repair operator are specifically designed. The traditional ALNS algorithm is typically used to solve classic vehicle pathing problems. However, in the scenario of this invention, the alternative parking points of the patrol vehicle do not need to be completely traversed, and the patrol vehicle travels along a defined track, making the traditional algorithm unsuitable for direct application. Therefore, this invention constructs an ALNS algorithm specifically for this scenario, capable of simultaneously optimizing the parking point selection of the patrol vehicle and the patrol path of the drone, thereby achieving efficient solutions for the collaborative patrol system.

[0032] This model involves many variables, and no cost function is set when the constraints are defined. This can lead to non-convergence when using heuristic algorithms to solve the problem, because hard constraints may cause the search space to break. To solve this problem, this section will relax the model to allow some solutions that do not meet the constraints to appear in the solution space, thereby avoiding the dilemma caused by hard constraints. The target value of the relaxed solution may become larger, but this method will help the ALNS algorithm to better learn the improvement of the solution, thereby accelerating convergence and finding a better solution. Specifically, this invention will constrain the flight distance of the UAV performing a single inspection mission as represented by constraint (15) to not exceed the maximum range of the UAV after a single full charge, and constrain the flight distance of each inspection point as represented by constraint (20) to not exceed the maximum range of the UAV after a single full charge. The service time must be relaxed within the time window. These two constraints will no longer be strict hard constraints, but will be transformed into soft constraints, thus allowing some solutions that violate the constraints to appear. At the same time, these violations will be penalized by the penalty term in the objective function, prompting the search process to gradually approach a feasible solution.

[0033] To address the issues of numerous model variables, complex constraints, and the problem of solution space fragmentation and algorithm convergence difficulties caused by strict hard constraints, this invention introduces a constraint relaxation and penalty function mechanism to relax the original optimization model, thereby improving the search efficiency and solution quality of the ALNS algorithm.

[0034] The relaxed objective function is expanded to the form of equation (26): (26) in, This represents the total completion time of the original problem. and These are the penalty terms for power constraints and time window constraints, respectively. and This is a dynamic penalty coefficient that adaptively adjusts during the iteration process. (Energy penalty term) Defined as: (27) This penalty applies to all drones that have exceeded the maximum permissible flight distance at each inspection point due to excessive flight distance caused by battery power consumption. The sum of squares. The use of square terms is intended to impose a nonlinear penalty on out-of-limit behavior. Solutions that severely exceed the limits will receive significantly higher penalty values, thereby effectively guiding the search toward the energy-feasible region.

[0035] Time window penalty Defined as: (28) This item applies to services earlier than the earliest service time. and later than the latest service time Violations are penalized. A quadratic penalty is also used, ensuring that nodes severely deviating from the time window have a significant negative impact on the objective function, thus prompting the algorithm to prioritize processing inspection points with obvious time window conflicts during optimization.

[0036] To balance exploration and exploitation, a penalty coefficient is applied. and The algorithm is designed for dynamic adjustment. A relatively large initial value is set in the early stages (e.g., ...). This strengthens the suppression of infeasible solutions; with each iteration, updates are performed according to an exponential decay strategy: (29) in As a decay factor, the penalty coefficient gradually decreases after a certain number of iterations, thus allowing a certain degree of exploration of infeasible regions in the later stages of the algorithm, enhancing the algorithm's ability to escape local optima.

[0037] Through the relaxation and penalty mechanisms described above, the ALNS algorithm can perform more flexible and continuous searches in a space containing slightly infeasible solutions, maintaining the guiding role of the original constraints while significantly improving the connectivity of the solution space and the convergence stability of the algorithm.

[0038] Based on the characteristics of patrol vehicle and drone patrol scenarios, this section designs a random initial solution based on the number of drones. Specifically, by analyzing the number of drones and the number of alternative parking points, patrol points are evenly and sequentially allocated to the drones. Then, the drones are allocated to alternative points according to battery-capable distance limits. This process generates a set of patrol vehicle parking points, and the drones are then randomly assigned to these points for patrol, resulting in a feasible solution for multiple drones. These solutions are combined to obtain a single feasible solution for this model, defined as an ordered sequence. The routes for the inspection vehicle and each drone are detailed below.

[0039] The patrol vehicle It consists of point pairs arranged in sequence of alternative parking spots, among which The solution for the drone is R = This represents the set of paths for all drones. Each drone's path... This indicates that the path for each drone consists of multiple sequentially arranged pairs of points, where... .

[0040] The specific method for generating the initial solution is as follows: Figure 2 As shown, the advantage of this method is that it can obtain a feasible solution regardless of the size of the data. The disadvantage is that while obtaining a feasible solution, the target value may be too large, causing the initial solution to be too far from the optimal solution. However, this problem can be solved by large-scale destruction and repair in the ALNS algorithm. Figure 2 The initialization process is shown.

[0041] After obtaining the initial solution, the algorithm performs a destruction operator to delete checkpoints and alternative points based on the generated initial problem solution. The selection of the destruction operator is determined by roulette wheel selection. This results in a partial solution that has undergone the destruction operation. Including deleted ones The set of inspection points that have been removed as alternative inspection vehicle locations. And these partial solutions This requires subsequent repair operations to reconstruct a new solution and enter the loop.

[0042] (1) Maximum flight distance contribution point removal operator This operator aims to optimize the energy efficiency and balance of task scheduling for drones. For each inspection point along a drone path... Its flight distance contribution value is defined as: (30) in and Representing points respectively In the path, the predecessor and successor nodes, if If there is no node before the point =0, The same applies to subsequent points. This value reflects the additional flight cost incurred by that point on the current path. The algorithm applies this value to all inspection points. Sort tasks in descending order and remove high-cost nodes. This operation helps to remove scheduling bottlenecks caused by distant nodes, making it possible to reallocate tasks to more suitable drones or take-off and landing points in subsequent repair phases, thereby promoting overall energy consumption reduction and time coordination optimization. (2) Random removal operator This operator aims to maintain search diversity by randomly selecting removal targets from all assigned inspection points or inspection vehicle candidate points. When selecting an inspection point to remove, the number of points to be removed is set according to a fixed proportion or absolute value, and each point has an equal probability of being selected. When selecting a candidate point to remove, the removed candidate points are stored in a new set. This is intended for use during repair. Its design philosophy aims to offset potential search biases introduced by heuristic operators: when an algorithm relies excessively on problem-feature-driven destruction strategies, it is prone to getting trapped in local optima. By introducing this stochastic strategy, the search can be encouraged to move beyond the current region and explore potentially high-quality solutions with greater structural differences in the solution space, thus providing a fundamental guarantee for global convergence.

[0043] (3) Time window urgency guides the removal operator This operator is specifically designed to handle the scheduling challenges posed by time window constraints. For each assigned inspection point... Its urgency index is defined as: (31) in Start the inspection with drones The time of the hour, and These represent the earliest and latest start times, respectively. The closer this value is to 1, the smaller the time window margin for the node. The algorithm is based on... All points are sorted in descending order, and points with high urgency are removed. This strategy proactively identifies and prioritizes resolving high-risk time window conflicts, creating opportunities for the remediation phase to reschedule these critical nodes with a more lenient time margin, thus improving the feasibility and robustness of the solution.

[0044] (4) High-load UAV path removal operator This operator belongs to a structural disruption strategy that operates along the entire drone path. The specific execution process is as follows: First, each drone is evaluated... The workload can be quantified by its task completion time. Then select the drone with the highest load. This involves removing all inspection points along the path. The motivation behind this strategy is to identify and eliminate factors affecting the overall completion time. The key bottleneck is the high-load drones, which are often a critical factor limiting the objective function value. By clearing their paths, the algorithm releases these inspection tasks for reallocation, potentially achieving load rebalancing during the repair phase and directly optimizing the maximum completion time. This operator has a high search perturbation strength, making it particularly suitable for structural restart when the algorithm gets stuck in local optima.

[0045] Upon receiving a partial solution that has undergone the destruction operator operation. Afterwards, the algorithm will perform a repair operation to re-insert it into the inspection vehicle path and the drone path. The selection of the insertion operator is also generated by roulette. After the insertion operator operation is completed, a new and complete solution will be obtained. .

[0046] (1) Minimum cost greedy insertion operator This operator employs a local optimum strategy to construct a feasible solution as quickly as possible. Its core mechanism is as follows: for each inspection point or candidate point to be inserted, the algorithm traverses all possible insertion positions (including all possible path segments of all drones) and precisely calculates the increase in the objective function cost after insertion at each position. The algorithm for removing points relative to the maximum flight distance contribution point selects points for each point. Insert at the minimum position. This strategy is computationally efficient and can quickly improve the quality of the solution, but its greedy nature may cause the final solution to get trapped in a local optimum.

[0047] (2) Regret-value-first insertion operator This operator overcomes the short-sightedness of greedy algorithms by simulating forward-looking decision-making. Its execution process consists of two steps: First, for each insertion point... Find the one with the lowest cost. Insertion positions and record their cost values. Next, calculate each point. Regret Value This value quantifies the potential additional cost incurred if the point is not inserted now. The algorithm prioritizes inserting the point with the largest regret value, placing it in the position with the lowest cost. This strategy effectively coordinates the insertion order and typically yields higher-quality solutions than a purely greedy strategy.

[0048] (3) Time window urgency priority insertion operator This operator prioritizes the satisfaction of time window constraints, representing a constraint-oriented repair strategy. It first sorts all inspection points to be inserted according to their time window urgency, which can be defined as: (32) in Start the inspection with drones The time of the point. The smaller the value, the more it represents a point. The smaller the leeway for delayed scheduling, the higher the urgency. The algorithm processes each point in descending order of urgency. When searching for an insertion position for each point, ensuring the feasibility of its time window is the primary objective, while minimizing disturbances to the timing of subsequent nodes. This strategy significantly improves the probability of generating a feasible solution.

[0049] This invention designs an adaptive optimization strategy with feedback adjustment characteristics for the dynamic selection mechanism of operators. Its core lies in constructing an operator performance evaluation system based on historical performance. This adaptive process drives the evolution of operator weights through quantitative evaluation indicators. The specific implementation framework includes the following two key dimensions: (1) Design of multimodal scoring mechanism To avoid the local optimum trap, the algorithm introduces an acceptance probability function based on the Metropolis criterion, the mathematical expression of which is shown in equation (33).

[0050] (33) in This represents the difference in the objective function between the old and new solutions. The annealing temperature parameter is dynamically adjusted. This mechanism allows for the acceptance of suboptimal solutions with a controllable probability under temperature parameter regulation, thereby enhancing the algorithm's escape capability.

[0051] Establish a graded scoring standard based on solution quality to dynamically and quantitatively evaluate operator performance: The highest reward score is assigned when the operator combination generates a new globally optimal solution. If a feasible solution that is better than the current solution is generated, a medium reward score is assigned. For solutions that deteriorate but satisfy the Metropolis acceptance criterion, assign a basic reward score. At the end of each iteration cycle, the performance scores of each operator are accumulated. ,in Indicates the operator within this period The number of times the k-th type of rating event is triggered.

[0052] (2) Dynamic weight update model The weight update function with memory decay characteristics is constructed as shown in equation (34): (34) In the formula, For the t-th periodic operator The weight, Its historical call count, The smoothing coefficient is used to balance historical information with current performance. To prevent the occurrence of abnormally small positive numbers during division by zero, this exponential smoothing model is used. This model preserves the long-term performance characteristics of the operator while also enabling rapid response to recent performance fluctuations.

[0053] Example 1: Medium-sized road section (dominated by off-site take-off and landing mode) Target section: A certain expressway from K100+000 to K150+000, with a total length of 50km, including 35km of straight sections and 15km of curved sections, with a minimum curve radius of 800m and a traffic density of 20 vehicles / km during off-peak hours; Alternative rest stop assembly points N: a total of 8 (N1-N8), including 2 service areas (N2, N6), 4 emergency parking lanes (N1, N3, N7, N8), and 2 maintenance work area entrances (N4, N5), with specific locations as shown in the table below: Inspection Point Assembly I: 30 points in total (I1-I30), including 5 bridges (I1-I5), 10 slopes (I6-I15), 10 guardrails (I16-I25), and 5 signs and markers (I26-I30). The operation time is 1-3 minutes per point, and the time window is divided into a relaxed type ([08:00,17:00]) and a moderately constrained type ([10:00,14:00]). Equipment parameters: Drones: 3 units (U1-U3), zmax=50km, vu=60km / h, operating power consumption 150W / h; Inspection vehicles: vc=80km / h (main line), 40km / h (emergency lane), battery swapping time 5 minutes / unit.

[0054] The implementation process is as follows: Data collection: Data on road sections and inspection points is collected through vehicle-mounted GPS and roadside sensors, and then imported into the algorithm after preprocessing; Model building: Substitute the above parameters to establish a joint optimization model and clarify the objective function and constraints (focusing on vehicle-machine timing matching for take-off and landing in different locations). Algorithm solution: 1000 iterations, α0=100, β0=100, γ=0.95, T initial=500. The destruction operator mainly uses the removal of the maximum flight distance contribution point + random removal. The repair operator adopts the minimum cost greedy insertion + regret value priority insertion. Plan execution: Conduct collaborative inspections according to the output path. U1 and U2 adopt remote take-off and landing mode, while U3 adopts local take-off and landing mode.

[0055] (III) Implementation Results Route plan: Inspection vehicle route: N1→N3→N6→N8, travel time 2.5 hours, stop time at each stop totaling 40 minutes (including battery swapping, drone recovery waiting time); Drone path: U1: N1 (takeoff) → I1 → I6 → I7 → I16 → I17 → N3 (landing), flight distance 48km, operation time 52 minutes; U2: N3 (takeoff) → I2 → I3 → I8 → I9 → I18 → I19 → N6 (landing), flight distance 45km, operation time 63 minutes; U3: N6 (takeoff) → I4 → I5 → I10-I15 → I20-I25 → I26-I30 → N6 (landing), flight distance 42km, operation time 85 minutes; Optimization results: Total task completion time Cmax = 3.5h; Compared with the traditional method (single drone + fixed take-off and landing point): the traditional method has a Cmax of 5.2h, while the present invention optimizes it by 32.7%; the drone's endurance utilization rate is increased from 65% to 92%, and the time window satisfaction rate is 100%.

[0056] Example 2: Large-scale road sections (multi-constraint, high-intensity scenario) Target section: A certain expressway from K200+000 to K300+000, with a total length of 100km, including 8 bridges and 2 tunnels, and a peak traffic density of 40 vehicles / km; Alternative stopping point set up N: 15 in total (N1-N15), including 3 service areas, 6 emergency parking lanes, 4 maintenance work area entrances, and 2 temporary work areas; Inspection Point Assembly I: A total of 80 points (I1-I80), including 4 tunnel portals (I25, I36, I58, I72), 12 high slopes (I10-I21), 20 bridge bearings (I1-I9, I22-I24, I26-I35), 30 guardrails (I37-I66), and 14 signs (I67-I80). The inspection windows for some tunnel portals are extremely tight (e.g., I25: [09:15, 09:25]). Equipment parameters: Drones: 5 units (U1-U5), zmax=40km (range is more critical), vu=55km / h; Inspection vehicle: vc=75km / h (main line), 35km / h (emergency lane), battery swapping time 4 minutes / unit.

[0057] The implementation process is as follows: Data Acquisition: Focus on collecting precise coordinates and time window data of high-risk inspection points such as tunnels and high slopes, and achieve real-time data transmission through 5G communication; Model building: The core constraints focus on the 40km range limit and tight time window, and strengthen the time transfer formula in the vehicle-machine collaboration constraints; Algorithm solution: 2000 iterations, α0=150, β0=150, γ=0.92, T_initial=800. The destruction operator prioritizes removal based on the urgency of the time window and overall removal of high-load drone paths. The repair operator prioritizes insertion based on the urgency of the time window. Plan Implementation: Five drones will be used in a mixed on-site / off-site take-off and landing mode to ensure the timely operation of inspection points within tight time windows.

[0058] The implementation results are as follows: (1) Route plan: Inspection vehicle route: N1→N4→N7→N10→N13→N15, travel time 4.8h, stop time total 75 minutes; (2) Drone routes: U1, U2, and U5 take off and land at different locations (e.g., U2: N4 → N7, U5: N10 → N13), while U3 and U4 take off and land in place (U3: N7, U4: N10). The flight distance of all drones is ≤39.5km (meeting the endurance constraint). (3) Optimization effect: Total task completion time Cmax = 6.2h; Compared with traditional heuristic algorithms (no off-site take-off and landing + simple greedy algorithm): the traditional method has a Cmax of 9.8h, while this invention optimizes it by 36.7%; the time window satisfaction rate is increased from 75% to 100%, the endurance violation rate is reduced from 20% to 0, and the inspection coverage efficiency is improved by 40%.

[0059] The ALNS algorithm constructed in this invention forms a logically rigorous and structurally complete heuristic optimization framework through the organic combination of constraint relaxation and penalty mechanisms, multi-strategy operator design, adaptive weight adjustment, and solution acceptance mechanisms. This framework not only effectively addresses the unique challenges of collaborative inspection path planning between inspection vehicles and UAVs, but also demonstrates good convergence performance and solution quality, providing valuable solution ideas for collaborative task planning problems in complex environments. The main algorithm framework is as follows: Figure 3 As shown.

[0060] This invention systematically studies the path planning problem in a collaborative inspection mode involving inspection vehicles and drones. First, based on an in-depth analysis of the actual needs and operational scenarios of highway inspections, the necessity and advantages of the collaborative mode are clarified: utilizing inspection vehicles as mobile bases to overcome the drone's endurance bottleneck, while simultaneously enabling drones to reach inspection points difficult for humans to access. Through mathematical modeling and comparative analysis of two drone operation modes—stationary take-off and landing and remote take-off and landing—the superiority of the remote take-off and landing mode in minimizing total operation time is theoretically proven, thus laying the core decision-making foundation for subsequent model construction.

[0061] This invention constructs a hybrid integer programming model integrating inspection vehicle point selection and UAV path planning. The model's core objective is to minimize the total completion time, while also incorporating complex constraints such as the maximum UAV range, time window constraints for inspection points, flow balance between the inspection vehicle and the UAV, and intricate spatiotemporal coupling relationships. However, given the NP-hard nature of this model, directly solving it using an exact algorithm presents significant challenges.

[0062] To address this challenge, this invention designs and implements an improved adaptive large neighborhood search algorithm. The algorithm's innovation lies in introducing constraint relaxation and a dynamic penalty function mechanism, transforming the hard constraints in the original problem into penalty terms in the objective function, effectively enhancing the algorithm's search capability and convergence stability in complex solution spaces. The core of the algorithm includes a set of destructive and repair operators designed specifically for the problem characteristics, and dynamically selects operators through an adaptive weight adjustment mechanism based on historical performance feedback, thereby intelligently balancing global exploration and local exploitation capabilities. Furthermore, the embedded simulated annealing acceptance criterion provides an effective way for the algorithm to escape local optima.

[0063] In summary, this invention, through a complete research chain from problem modeling and algorithm design to experimental verification, has successfully established a set of theoretical methods and practical solutions for solving the collaborative path planning problem of inspection vehicles and drones. The relevant models, algorithms, and conclusions have important theoretical significance and application reference value for promoting the development of intelligent inspection technology.

[0064] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent highway inspection path planning based on patrol vehicle-multi-UAV collaboration, characterized in that, Includes the following steps: S1. Obtain the set of alternative stop points for the target section of the highway. Inspection point assembly and core device parameters, including the maximum flight range of the drone. ; S2. Construct a joint optimization model coupling the selection of inspection vehicle docking points and driving routes, and the allocation of multiple UAV tasks and flight paths, with the total operation completion time as the starting point. With minimization as the objective, and incorporating constraints such as UAV endurance, inspection point time windows, and vehicle-machine collaboration timing, the core objective function of the joint optimization model is: ； in, , , For drones, For the inspection vehicle to reach the alternative parking point At that moment, For inspection vehicles or drones at alternative parking points The time spent at the place, For drones The time when you return to the parking spot after completing the task; S3. The improved adaptive large neighborhood search ALNS algorithm is used to solve the joint optimization model to obtain the collaborative inspection path scheme of the inspection vehicle and multiple UAVs; S4. Output the collaborative inspection path plan to guide the inspection vehicle and multiple drones to complete the highway inspection.

2. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, In step S1, the information of the target road segment also includes the starting mileage marker and the ending mileage marker; the set of alternative stop points The set of inspection points is a discrete set of locations along the highway that meet safety, spatial, and communication requirements; This is a collection of key inspection facilities or potential risk points along the highway. Each inspection point has precise spatial coordinates, work type, estimated work duration, and acceptable time window. The core parameters of the equipment also include the drone's flight speed and the inspection vehicle's travel speed.

3. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, The set of alternative stops This includes dedicated work areas near highway service areas, ramp triangle areas or ramp merging / departure areas, emergency parking lanes, operation and management sub-centers, maintenance work area entrances, and temporary work areas designated according to regulations.

4. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, In step S3, the drone's flight path supports both in-situ take-off and landing mode and out-of-situ take-off and landing mode: the in-situ take-off and landing mode is when the drone is at the same alternative docking point. Complete takeoff, inspection, and landing tasks; the remote takeoff and landing mode refers to the UAV landing at an alternative docking point. After takeoff and completion of the inspection mission, the vehicle will proceed to the designated alternative docking point already in place. landing, .

5. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, The constraints of the joint optimization model include inspection point assignment constraints, inspection vehicle flow balance constraints, and UAV flow balance constraints, as detailed below: (1) Constraints on the assignment of inspection points: ,in Indicates drone From the inspection point Fly to The value is 1 if it is true, and 0 otherwise. (2) Inspection vehicle flow balance constraints: ; ; ,in This indicates that the inspection vehicle will depart from the alternative parking point. Drive to the alternative parking spot The value is 1 if it is true, and 0 otherwise. To remove the set of alternative parking spots for the destination, To remove the set of alternative parking spots from the starting point, To remove the set of alternative parking spots from the origin and destination; (3) Unmanned aerial vehicle (UAV) flow balance constraints: .

6. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, The expression for the UAV endurance constraint in the joint optimization model is: ,in Indicates drone Arrival at the inspection point The cumulative flight distance over time, and , For point and The Euclidean distance between them The constant is sufficiently large; the time window constraint expression for the inspection point is: ,in For drone inspection The starting moment of point.

7. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 6, characterized in that, The specific vehicle-machine cooperative constraints of the joint optimization model are as follows: (1) Starting point constraints for UAVs: ; ; (2) UAV endpoint constraints: ; ; (3) Time transit constraints: ； ； ； ； ； in, For drones travel time of the segment For drones at inspection points The time required for inspection For the inspection vehicle The travel time of the segment.

8. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 1, characterized in that, In step S3, the improved ALNS algorithm handles the range constraint and time window constraint through constraint relaxation and penalty function mechanism. The relaxed objective function is: ； in, This is a penalty item for power constraints. The time window constraint penalty term, and This is a dynamic penalty coefficient; The power constraint penalty item ; The time window constraint penalty item ; The dynamic penalty coefficient is updated according to an exponential decay strategy: , ,in As the attenuation factor, , .

9. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 8, characterized in that, In step S3, the destruction operators of the improved ALNS algorithm include at least one of the following: maximum flight distance contribution point removal operator, time window urgency-guided removal operator, high-load UAV path overall removal operator, and random removal operator. The improved ALNS algorithm's repair operators include at least one of the following: minimum cost greedy insertion operator, regret-first insertion operator, and time window urgency-first insertion operator.

10. The intelligent highway inspection path planning method based on inspection vehicle-multi-UAV collaboration as described in claim 9, characterized in that, In step S3, the adaptive process of the improved ALNS algorithm includes a multimodal scoring mechanism, dynamic weight update, and simulated annealing acceptance criterion. The improved ALNS algorithm generates its initial solution by: distributing inspection points evenly among multiple drones based on the number of drones and the number of candidate docking points, while ensuring maximum endurance. Under the premise of mapping to adjacent alternative docking point intervals, the internal inspection points of each UAV are randomly connected to form a path, and combined to obtain a feasible initial solution; the operation content types of the inspection points include taking pictures, video acquisition, and laser scanning; the time window of the inspection points It is determined by traffic environment, lighting conditions, operation and management requirements, and safety regulations.

Images