Highway roadbed-side slope empty vehicle cooperative detection scheduling method

Abstract

The invention relates to a highway subgrade-side slope empty vehicle cooperative detection scheduling method. The method comprises the following steps: acquiring center coordinates, detection time and spatial distribution characteristics of all side slopes; based on the center coordinates, the detection time and the spatial distribution characteristics, dividing all the side slopes by adopting an improved hierarchical clustering algorithm to obtain a plurality of clusters, each cluster representing an operation area; for each operation area, selecting a parking point position by using an optimization model taking minimization of a total task as a target; planning a detection path of the unmanned aerial vehicle for the side slope in each operation area; planning a global optimal vehicle driving path on a road network based on the parking point position; and based on the time cooperative constraint of the vehicle and the unmanned aerial vehicle, in combination with the detection path of the unmanned aerial vehicle and the globally optimal vehicle driving path, cooperative detection is carried out, and a slope detection task is completed. The method has the advantages of improving the detection efficiency and the coverage rate and the like.

Description

Technical Field

[0001] This invention relates to the field of traffic engineering technology, and in particular to a method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope. Background Technology

[0002] Stability testing of highway subgrades and slopes is a crucial aspect of ensuring safe highway operation. Current testing technologies primarily include manual inspection, fixed monitoring equipment, and single-drone inspection. However, these methods generally suffer from low efficiency, incomplete coverage, and poor real-time performance. While vehicle-drone collaborative inspection models theoretically have advantages, and some studies have attempted to combine vehicles and drones for collaborative inspection—including designing vehicle-mounted drone take-off and landing platforms and researching drone swarm collaborative inspection methods—existing methods lack comprehensive consideration of slope spatial distribution characteristics, drone endurance constraints, and highway network topology, making efficient collaborative operations difficult to achieve.

[0003] Existing collaborative detection methods have significant limitations in task allocation, path planning, and temporal scheduling: slope task allocation does not consider the matching of spatial clustering and UAV endurance; vehicle path planning and UAV tasks lack collaborative optimization; UAV detection paths are not adaptively designed for slope terrain features; the time coordination mechanism between vehicles and UAVs is imperfect, leading to excessive waiting time or task interruption; and there is a lack of self-learning optimization capabilities based on historical data. In complex road network environments, balancing the contradiction between "full coverage" and "high efficiency" has become a technical challenge. Traditional methods often adopt simple zoning strategies, such as dividing detection areas by kilometer markers, ignoring the spatial clustering of slopes and the flight characteristics of UAVs, resulting in both energy waste and detection blind spots. Meanwhile, vehicle path planning is usually independent of UAV task allocation, failing to achieve true "air-ground collaboration." Therefore, there is an urgent need for a collaborative detection scheduling method that can comprehensively consider multiple factors to solve the key problems of low detection efficiency, incomplete coverage, low resource utilization, and poor adaptability in existing technologies. Summary of the Invention

[0004] The purpose of this invention is to provide a method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope to improve detection efficiency and coverage.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for collaborative detection and scheduling of empty vehicles on highway subgrade and slope includes the following: Obtain the center coordinates, detection time and spatial distribution characteristics of all slopes; Based on the center coordinates, detection time and spatial distribution characteristics, an improved hierarchical clustering algorithm is used to divide all slopes into multiple clusters, each cluster representing an operating area. For each operating area, the parking point location is selected using an optimization model that aims to minimize the total task. For the slopes within each operating area, plan the detection path for the drone; Based on the location of the parking point, the globally optimal vehicle travel path is planned on the highway network. Based on the time coordination constraints between vehicles and drones, and combining the detection path of the drones with the globally optimal vehicle travel path, collaborative detection is performed to complete the slope detection task.

[0006] Furthermore, the step of obtaining multiple clusters includes: Calculate the Euclidean distance between the center coordinates of any two slopes and construct a distance matrix; Based on the distance matrix and combined with the detection time and spatial distribution characteristics, an improved hierarchical clustering algorithm is used to gradually merge the two clusters with the closest spatial distance from bottom to top. At the same time, spatiotemporal constraints are introduced to obtain multiple initial clusters that satisfy the spatiotemporal constraints. The clustering quality of different initial clusters is evaluated using a combination of silhouette coefficient and intra-class distance, and the optimal number of clusters is determined by the elbow rule.

[0007] Furthermore, the spatiotemporal constraints include: Spatially, the slopes within the cluster must satisfy the requirement of spatial proximity; In terms of time, the total detection time of all slopes within the cluster and the round-trip transfer time of the UAV are calculated in real time during the merging process, and will not exceed the set time. The set time is calculated as the single flight time of the UAV × a set ratio, and the transfer time is estimated based on the spatial distribution characteristics of the slopes within the cluster and the predicted path.

[0008] Furthermore, the step of obtaining multiple clusters also includes: For an isolated slope, if its distance from all cluster centers exceeds the set threshold, it is assigned to the nearest cluster, and the total task time in the cluster is checked again to see if it still meets the spatiotemporal constraints. If the constraints are not met, the slope that is spatially farthest from the isolated point is removed from the cluster and treated as a new isolated point. The assignment and verification process is repeated until all slopes are assigned and all clusters meet the spatiotemporal constraints.

[0009] Furthermore, the optimization model includes an objective function and corresponding constraints, wherein the objective function includes the UAV detection time, the UAV round-trip transfer time, and the vehicle transfer time between parking points, and the expression of the objective function is: , in, Let be the objective function. This indicates the drones in various operating areas. kInspection time, Indicates whether the vehicle has left the parking spot. i Drive to the parking spot j , Indicates parking spot i and j The actual shortest distance between them The average speed of the vehicle during the inspection process; The constraints include: (a) Parking spots must be located at nodes in the highway network to ensure vehicle accessibility; (b) The distance between the parking point and any slope within the operating area shall not be less than the safety threshold D_min; (c) Total task time within the operating area t k satisfy: t k ≤T_max, where T_max is the maximum set task time; (d) The vehicle must visit all parking spots.

[0010] Furthermore, the step of selecting the parking point location includes: An improved K-means algorithm is used, with the center coordinates of each cluster as the initial point. Under the premise of satisfying the constraints, the parking point position is iteratively optimized to minimize the objective function and obtain the initial optimal parking point position for each operating area. The initial optimal parking point location is mapped to the nearest road network node using a projection method to obtain the optimal parking point location.

[0011] Furthermore, the step of planning the detection path of the UAV includes: Based on the surface slope and obstacle distribution of the slope in each operating area, the forward direction of the UAV's zigzag scanning is dynamically adjusted, and the line spacing is adjusted according to the roughness of the slope. The slope surface is discretized into a grid map. The A* algorithm is used to search for the optimal flight trajectory from the starting point to the ending point and covering all zigzag scan segments in the grid map. A start point and an end point are set for each zigzag scan segment, and the transfer path between the zigzag scan segments is optimized to obtain the final detection path of the UAV. The cost function of the A* algorithm is: f(n) = g(n) + h(n), In the formula, f(n) is the total cost function, g(n) is the actual flight distance, and h(n) indicates that the Manhattan distance is used as the heuristic function.

[0012] Furthermore, the step of planning the detection path of the UAV also includes: A steering penalty factor is introduced into the cost function of the A* algorithm. When the drone needs to make a 90-degree turn, an additional cost is added. The cost function with the steering penalty factor is expressed as follows: , , in, As a turning penalty factor, For the redirection penalty item, For the first in the path j The angle of the next turn.

[0013] Furthermore, globally optimal vehicle travel paths are planned for two types of scenarios: a full road segment-slope coverage scenario and a full slope coverage scenario. (1) For the scenario of full coverage of road segment and slope, the minimum overlap Eulerian path algorithm is used for planning. The planning steps include: Abstract the highway network as a graph and check the degree of all nodes in the graph; The Floyd algorithm is used to calculate the shortest Manhattan distance between all pairs of odd-degree nodes. Then, the odd-degree nodes are matched with minimum weight to obtain a set of connections that minimizes the total length of the paths connecting the pairs of odd-degree nodes. The shortest path is then used as a virtual repeating edge to transform the graph into an Eulerian graph. Based on the Eulerian graph, the Hierholzer algorithm is used to solve for the Eulerian circuit. The Eulerian circuit is the path with the shortest total distance for the vehicle to cover all road segments, also known as the globally optimal vehicle travel path. (2) For the scenario of full slope coverage only, the simulated annealing algorithm is used for planning. The planning steps include: Initialization: Set the initial temperature T0 and the final temperature T end Cooling coefficient α, number of iterations at each temperature; Initial solution generation: A greedy algorithm is used to select the nearest unvisited parking point location from the base station to form the initial vehicle travel path. Iterative optimization: Based on the initial vehicle travel path, a neighborhood solution is generated using a 2-opt exchange operation as a new path. According to the new path and the initial vehicle travel path, the Metropolis criterion is used to accept the new path in order to escape the local optimum. This step is repeated multiple times until the iteration termination condition is reached, and the final optimal vehicle access sequence is obtained as the globally optimal vehicle travel path.

[0014] Furthermore, the steps for performing collaborative detection include: Establish the time coordination constraints between the vehicle and the drone as follows: , In the formula, For the vehicle to arrive at the i The time at each parking location Let be the time when the drone takes off from the i-th parking point. The total flight time required for the UAV to complete slope detection within the i-th operating area. Let be the time when the vehicle leaves from the i-th parking point; Construct a multi-objective optimization function: , in, , Indicates the first i There are parking spots, and the downtime while waiting for the drone to take off after the vehicle arrives. This is used to balance the two objectives of total task time and the tightness of vehicle-machine collaboration; An improved genetic algorithm is used to solve the multi-objective optimization function under the time coordination constraint of the vehicle and the UAV, so as to complete the detection task in the shortest time. At the same time, an elastic time window mechanism is introduced to dynamically adjust the planned arrival time of subsequent parking points when the task times out at a certain parking point, so as to finally complete the slope detection task.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) Significantly improved efficiency and coverage: This invention significantly expands the coverage of a single task by matching and optimizing the spatial clustering of slopes with the endurance of UAVs, thereby comprehensively improving detection efficiency and coverage and effectively enhancing the accuracy of slope disease identification.

[0016] (2) Resource synergy optimization: This invention achieves precise matching of vehicle and drone working time through parking point location optimization, significantly reducing vehicle waiting time, improving drone energy utilization efficiency, and reducing overall testing costs.

[0017] (3) Strong adaptability: The present invention adopts a hybrid path optimization strategy for different scenarios, which can flexibly adapt to various road network topologies, ensure a high task completion rate, and effectively cope with complex and ever-changing highway environments. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a road network topology diagram of Embodiment 1 of the present invention; Figure 3 The vehicle trajectory and parking point location calculated in Embodiment 1 of the present invention; Figure 4 This is a road network topology diagram of Embodiment 2 of the present invention; Figure 5 The vehicle trajectory and parking point location are calculated in Embodiment 2 of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0020] Example 1 This embodiment provides a collaborative detection and scheduling method for highway subgrade and slope with empty vehicles to address technical problems in existing highway slope detection technologies, such as low efficiency of vehicle-UAV collaboration, incomplete slope coverage, insufficient resource utilization, and poor adaptability. Specifically, this method achieves an intelligent highway detection system with reduced detection time and improved coverage efficiency through coupled design of slope clustering and parking point optimization, collaborative planning of UAV scanning paths and highway network topology, and a closed-loop feedback adaptive scheduling mechanism.

[0021] This embodiment uses a simulated 3km × 3km grid-like road network as an example. This network contains 16 nodes and 24 road segments, with 44 slopes to be tested randomly distributed along the road. Figure 2 As shown in the figure. This method consists of a detection vehicle and a drone.

[0022] like Figure 1 As shown, the method includes the following steps: S1: Obtain the center coordinates, detection time and spatial distribution characteristics of all slopes to be detected; combine slope distance and UAV endurance, and use an improved hierarchical clustering algorithm to divide the operation area.

[0023] First, this embodiment obtains the center coordinates (c) of all 44 slopes. i ) and its estimated detection time ( t i The detection time was calculated based on the slope length and the "Z"-shaped scanning parameters of the UAV.

[0024] Next, an improved hierarchical clustering algorithm is used to partition the operational region. The specific steps are as follows: S11: Calculate the Euclidean distance between any two slope center points and construct a 44×44 distance matrix.

[0025] S12: A bottom-up hierarchical clustering algorithm is adopted, starting with the slope as the initial cluster, and gradually merging the two clusters with the closest spatial distance.

[0026] S13: Introducing spatiotemporal constraints. Spatially, slopes within a cluster must satisfy spatial proximity; temporally, during merging, the total detection time of all slopes within the new cluster and the round-trip transfer time of the UAV are calculated in real time to ensure that it does not exceed a set time. The set time = UAV single-trip endurance time (T_max = 40 minutes) × set ratio (85%), i.e., the set time is 34 minutes. The transfer time here is estimated based on the distribution of slopes within the cluster and the predicted path.

[0027] S14: The clustering quality under different numbers of clusters (k) is evaluated using a combination of silhouette coefficient and intra-cluster distance. The optimal number of clusters, k=14, is determined by the "elbow rule," which achieves the highest intra-cluster compactness while satisfying the time constraint.

[0028] S15: For isolated slopes that are far from all cluster centers, assign them to the nearest cluster and check again whether the total task time of the cluster still meets the constraints. If the constraints are not met, remove the slope that is farthest from the isolated point from this cluster and treat it as a new isolated point. Re-execute the assignment and verification process until all slopes are assigned and all clusters meet the spatiotemporal constraints.

[0029] Ultimately, the 44 slopes were divided into 14 operational areas (clusters), and the number of slopes in each area and the estimated task duration are shown in Table 1.

[0030] Table 1. Slope clustering results and parking point coordinates S2: The weighted sum of the round-trip time of the drone and the transfer time of the vehicle determines the minimum target parking point, which meets the minimum safe distance from the slope in the area.

[0031] For the 14 divided operating areas, an optimal parking point (Pk) is determined for each area.

[0032] S21: Establish a mechanism to minimize the total task detection time ( t k The optimization model is defined by the objective function, which includes the drone detection time, the drone round-trip transfer time, and the vehicle transfer time between parking points, as shown in the following formula: , in, Let be the objective function. Indicates the drone's location in each operating area. k Inspection time, Indicates whether the vehicle has left the parking spot. i Drive to the parking spot j , Indicates parking spot i and jThe actual shortest distance between them This represents the average speed of the vehicle during the inspection process.

[0033] S22: Set constraints: (a) The parking point must be located on a road network node to ensure vehicle accessibility; (b) The distance between the parking point and any slope within the area must not be less than the safety threshold D_min (set to 50 meters in this embodiment); (c) The total task time within the area. t k ≤T_max, where T_max is the maximum task time set; (d) The detection vehicle must visit all parking points.

[0034] S23: Using the improved K-means algorithm, with the coordinates of each cluster center obtained in step S1 as the initial point, the parking point position is iteratively optimized under the premise of satisfying the safety distance constraint, so as to minimize the objective function.

[0035] S24: By using the projection method, the coordinates of the optimal parking point obtained from the theoretical calculation are mapped to the nearest road network node to ensure vehicle accessibility and form the final 14 vehicle-accessible parking points.

[0036] Step S25: For adjacent clustered regions, set up a location conflict detection mechanism and adjust the parking point positions to avoid spatial overlap.

[0037] S3: The path is planned by scanning in a "Z" shape and the trajectory is optimized by combining the A* algorithm to ensure full coverage of the slope surface without repetition.

[0038] This step involves planning a refined detection path for the drone on the slopes within each operating area.

[0039] S31: Based on the surface slope and obstacle distribution of the slope, dynamically adjust the direction of the "Z"-shaped scan to make it follow the contour line direction as much as possible, and adjust the scan line spacing (Dw) according to the slope roughness.

[0040] S32: Discretize the slope surface into a grid map, and apply the A* algorithm to search for the optimal flight trajectory from the starting point to the ending point in the grid map, which covers all scan lines. In the cost function f(n) = g(n) + h(n) of the A* algorithm, f(n) is the total cost function, g(n) is the actual flight distance, and h(n) uses the Manhattan distance as a heuristic function.

[0041] S33: Introduce a turning penalty factor into the cost function. When the drone needs to make a 90-degree turn, an additional cost is added, thereby planning a smoother, lower-energy-consumption path. (Introduce turning penalty factor) The cost function is: , , in, As a turning penalty factor, For the redirection penalty item, For the first in the path j The angle of the second turn. S35: Set up high-density scanning sub-regions for key areas of the slope (such as historical disease points and high-risk areas) to improve detection accuracy.

[0042] The final detection path for the drone was planned.

[0043] S4: Based on the road network and parking point distribution, calculate the optimal vehicle path for different scenarios: 1. Full coverage of road segment and slope: Minimum overlap Eulerian path algorithm 2. Full coverage of slope only: Simulated annealing algorithm.

[0044] This embodiment is for the "slope coverage only" scenario, where the vehicle is responsible for transporting the drone to each parking point and conducting roadbed detection during the journey, without needing to cover all road sections.

[0045] The simulated annealing (SA) algorithm is used to find the optimal sequence for vehicles to visit all 14 parking points (including origin / endpoint base stations).

[0046] Initialization: Set the initial temperature T0 = 120, and the final temperature T end =1, cooling coefficient α=0.99, 1000 iterations at each temperature.

[0047] Initial solution generation: A greedy algorithm is used to start from the base station (operation and maintenance center) and select the nearest unvisited parking point each time to form the initial path.

[0048] Iterative optimization: A 2-opt swap operation is used to generate a neighborhood solution (new solution). The new solution is accepted according to the Metropolis criterion to escape local optima. The final optimal vehicle access sequence, as the globally optimal vehicle travel path, is: Base Station → P6 → P13 → P15 → P14 → P9 → P8 → P11 → P2 → P3 → P12 → P10 → P4 → P5 → P7 → Base Station. The planned path is as follows: Figure 3 As shown.

[0049] S5: Establish coordinated constraints between vehicle arrival and drone takeoff to ensure that detection is completed during the parking period with the shortest total time.

[0050] This step mainly involves establishing time-coordinated constraints between the vehicle and the drone, under which mechanical energy is detected collaboratively.

[0051] S51: For each parking point Pk, establish time coordination constraints between the vehicle and the drone: , In the formula, For the vehicle to arrive at the i The time at each parking location For drones from the first i The departure time from each parking spot To complete the first for drones i Total flight time required for all slope inspections within the operating area For vehicles from the first i The time to leave from each parking spot.

[0052] S52: Design a multi-objective optimization function to simultaneously minimize the total task completion time, vehicle waiting time, and UAV energy consumption. The multi-objective optimization function is expressed as: , in, , Indicates the first i There are parking spots, and the downtime while waiting for the drone to take off after the vehicle arrives. This is used to balance the two objectives of total mission time and vehicle-machine collaboration.

[0053] S53: An improved genetic algorithm is used for cooperative scheduling optimization to complete the detection task in the shortest possible time. Chromosome encoding consists of two parts: the order in which vehicles visit parking spots (from S4) and the task sequence of the drone at each parking spot (from S3). The fitness function simultaneously minimizes the total task completion time, the total vehicle waiting time, and the total drone energy consumption.

[0054] S54: Introducing a flexible time window mechanism. If a certain area (such as area 3) times out due to some reason, the planned arrival time of all subsequent parking points will be dynamically adjusted to ensure that the task chain is not broken.

[0055] Step S55: Set priority rules to allocate longer detection times and more flexible time windows to high-risk slope areas.

[0056] S6: Collect detection data and execution time, dynamically update clustering, parking point and path parameters to form a continuous optimization system.

[0057] In this simulation, calculations show that the total detection time using the collaborative detection method of this invention is 265.01 minutes. Compared to independent, non-cooperative operation of vehicles and drones, the detection time is reduced by 32%. Compared to using a greedy algorithm (GA) for vehicle path planning, the simulated annealing algorithm (SA) used in this invention reduces the vehicle travel distance by 13%, verifying the effectiveness of this method.

[0058] Example 2 This embodiment provides a collaborative detection and scheduling method for empty vehicles along a highway subgrade and slope. This method is based on a real-world road segment-slope full-coverage scenario and specifically includes the following: This embodiment uses a real highway network in a certain province as an example, such as... Figure 4 As shown, the application is highlighted in scenarios where "full coverage of road sections and slopes" is required.

[0059] S1 to S3: Similar to Example 1, slope information is first obtained based on GIS data, the operation area is divided and the parking point is determined by hierarchical clustering, and then the drone plans a "Z"-shaped scanning path for each area.

[0060] S4: Global Vehicle Path Planning (Full Road Coverage Scenario) This scenario requires the inspection vehicle to perform at least one inspection on all road sections, using the Minimum Overlap Eulerian Path (MOE) algorithm.

[0061] S41: First, abstract the road network as a graph and check the degree of all nodes. Since road networks usually do not satisfy the Eulerian graph condition (i.e., all nodes have an even degree), there are multiple nodes with odd degrees.

[0062] S42: Calculate the shortest path (Manhattan distance) between all pairs of odd-degree nodes using the Floyd algorithm. Then, perform minimum-weight matching on these odd-degree nodes, finding a set of connections that minimizes the total length of the paths connecting these pairs. By adding these shortest paths as "virtual repeating edges," the original graph is transformed into an Eulerian graph.

[0063] S43: On the transformed Eulerian graph, the Hierholzer algorithm is used to solve for the Eulerian circuit, which is the path with the shortest total distance and covers all road segments. The planning result is as follows: Figure 5 As shown.

[0064] S5 and S6: The collaborative scheduling method is the same as in Example 1. Final experimental results show that in this real-world case, compared to the baseline method, the method of this invention reduces vehicle detection time by 12.48%, while achieving 100% detection coverage of the road network through the MOE algorithm, increasing coverage by 37.98%.

[0065] In summary, through the above specific implementation steps, this invention effectively solves the problem of time and air temperature in the collaborative detection of highway subgrade and slope, and significantly improves detection efficiency and coverage.

[0066] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for collaborative detection and scheduling of empty vehicles along highway subgrade and slope, characterized in that, Including the following: Obtain the center coordinates, detection time and spatial distribution characteristics of all slopes; Based on the center coordinates, detection time and spatial distribution characteristics, an improved hierarchical clustering algorithm is used to divide all slopes into multiple clusters, each cluster representing an operating area. For each operating area, the parking point location is selected using an optimization model that aims to minimize the total task. For the slopes within each operating area, plan the detection path for the UAV; Based on the parking point locations, the globally optimal vehicle travel path is planned on the highway network. Based on the time coordination constraints between vehicles and drones, and combining the detection path of the drones with the globally optimal vehicle travel path, collaborative detection is performed to complete the slope detection task.

2. The method for coordinated detection and scheduling of empty vehicles along a highway subgrade and slope as described in claim 1, characterized in that, The steps for obtaining multiple clusters include: Calculate the Euclidean distance between the center coordinates of any two slopes and construct a distance matrix; Based on the distance matrix and combined with the detection time and spatial distribution characteristics, an improved hierarchical clustering algorithm is used to gradually merge the two clusters with the closest spatial distance from bottom to top. At the same time, spatiotemporal constraints are introduced to obtain multiple initial clusters that satisfy the spatiotemporal constraints. The clustering quality of different initial clusters is evaluated using a combination of silhouette coefficient and intra-class distance, and the optimal number of clusters is determined by the elbow rule.

3. The method for coordinated detection and scheduling of empty vehicles along a highway subgrade and slope according to claim 2, characterized in that, The spatiotemporal constraints include: Spatially, the slopes within the cluster must satisfy the requirement of spatial proximity; In terms of time, the total detection time of all slopes within the cluster and the round-trip transfer time of the UAV are calculated in real time during the merging process, and will not exceed the set time. The set time is calculated as the single flight time of the UAV × a set ratio, and the transfer time is estimated based on the spatial distribution characteristics of the slopes within the cluster and the predicted path.

4. The method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope according to claim 2, characterized in that, The step of obtaining multiple clusters also includes: For an isolated slope, if its distance from all cluster centers exceeds the set threshold, it is assigned to the nearest cluster, and the total task time in the cluster is checked again to see if it still meets the spatiotemporal constraints. If the constraints are not met, the slope that is spatially farthest from the isolated point is removed from the cluster and treated as a new isolated point. The assignment and verification process is repeated until all slopes are assigned and all clusters meet the spatiotemporal constraints.

5. The method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope according to claim 1, characterized in that, The optimization model includes an objective function and corresponding constraints. The objective function includes the UAV detection time, the UAV round-trip transfer time, and the vehicle transfer time between parking points. The expression of the objective function is: ， in, Let be the objective function. Indicates the drone's location in each operating area. k Inspection time, Indicates whether the vehicle has left the parking spot. i Drive to the parking spot j , Indicates parking spot i and j The actual shortest distance between them The average speed of the vehicle during the inspection process; The constraints include: (a) Parking spots must be located at nodes in the highway network to ensure vehicle accessibility; (b) The distance between the parking point and any slope within the operating area shall not be less than the safety threshold D_min; (c) Total task time within the operating area t k satisfy: t k ≤T_max, where T_max is the maximum set task time; (d) The vehicle must visit all parking spots.

6. The method for coordinated detection and scheduling of empty vehicles along a highway subgrade and slope according to claim 1, characterized in that, The steps for selecting the parking spot location include: An improved K-means algorithm is used, with the center coordinates of each cluster as the initial point. Under the premise of satisfying the constraints, the parking point position is iteratively optimized to minimize the objective function and obtain the initial optimal parking point position for each operating area. The initial optimal parking point location is mapped to the nearest road network node using a projection method to obtain the optimal parking point location.

7. The method for coordinated detection and scheduling of empty vehicles along a highway subgrade and slope according to claim 1, characterized in that, The steps for planning the detection path of the UAV include: Based on the surface slope and obstacle distribution of the slope in each operating area, the forward direction of the UAV's zigzag scanning is dynamically adjusted, and the line spacing is adjusted according to the roughness of the slope. The slope surface is discretized into a grid map. The A* algorithm is used to search for the optimal flight trajectory from the starting point to the ending point and covering all zigzag scan segments in the grid map. A start point and an end point are set for each zigzag scan segment, and the transfer path between the zigzag scan segments is optimized to obtain the final detection path of the UAV. The cost function of the A* algorithm is: f(n) = g(n) + h(n), In the formula, f(n) is the total cost function, g(n) is the actual flight distance, and h(n) indicates that the Manhattan distance is used as the heuristic function.

8. The method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope according to claim 7, characterized in that, The steps for planning the detection path of the UAV also include: A steering penalty factor is introduced into the cost function of the A* algorithm. When the drone needs to make a 90-degree turn, an additional cost is added. The cost function with the steering penalty factor is expressed as follows: ， ， in, As a turning penalty factor, For the reversal penalty item, For the first in the path j The angle of the next turn.

9. The method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope according to claim 1, characterized in that, The globally optimal vehicle travel path is planned for two types of scenarios: a scenario with full coverage of road sections and slopes, and a scenario with full coverage of slopes only. (1) For the scenario of full coverage of road segment and slope, the minimum overlap Eulerian path algorithm is used for planning. The planning steps include: Abstract the highway network as a graph and check the degree of all nodes in the graph; The Floyd algorithm is used to calculate the shortest Manhattan distance between all pairs of odd-degree nodes. Then, the odd-degree nodes are matched with minimum weight to obtain a set of connections that minimizes the total length of the paths connecting the pairs of odd-degree nodes. The shortest path is then used as a virtual repeating edge to transform the graph into an Eulerian graph. Based on the Eulerian graph, the Hierholzer algorithm is used to solve for the Eulerian circuit. The Eulerian circuit is the path with the shortest total distance for the vehicle to cover all road segments, also known as the globally optimal vehicle travel path. (2) For the scenario of full slope coverage only, the simulated annealing algorithm is used for planning. The planning steps include: Initialization: Set the initial temperature T0 and the final temperature T end Cooling coefficient α, number of iterations at each temperature; Initial solution generation: A greedy algorithm is used to select the nearest unvisited parking point location from the base station to form the initial vehicle travel path. Iterative optimization: Based on the initial vehicle travel path, a neighborhood solution is generated using a 2-opt exchange operation as a new path. According to the new path and the initial vehicle travel path, the Metropolis criterion is used to accept the new path in order to escape the local optimum. This step is repeated multiple times until the iteration termination condition is reached, and the final optimal vehicle access sequence is obtained as the globally optimal vehicle travel path.

10. The method for coordinated detection and scheduling of empty vehicles on highway subgrade and slope according to claim 1, characterized in that, The steps for performing collaborative detection include: Establish the time coordination constraints between the vehicle and the drone as follows: ， In the formula, For the vehicle to arrive at the i The time at each parking spot location Let be the time when the drone takes off from the i-th parking point. The total flight time required for the UAV to complete slope detection within the i-th operating area. Let be the time when the vehicle leaves from the i-th parking point; Construct a multi-objective optimization function: ， in, , Indicates the first i There are parking spots, and the downtime while waiting for the drone to take off after the vehicle arrives. This is used to balance the two objectives of total task time and the tightness of vehicle-machine collaboration; An improved genetic algorithm is used to solve the multi-objective optimization function under the time coordination constraint of the vehicle and the UAV, so as to complete the detection task in the shortest time. At the same time, an elastic time window mechanism is introduced to dynamically adjust the planned arrival time of subsequent parking points when the task times out at a certain parking point, so as to finally complete the slope detection task.

Images