A large-scale air-ground cooperative task allocation method based on hierarchical divide-and-conquer
By decomposing the air-ground collaborative task into low-dimensional sub-problems using a hierarchical divide-and-conquer approach, and by optimizing through crossover, mutation, and clustering, the problems of poor convergence efficiency and computational stability in air-ground collaborative task allocation are solved. This enables efficient generation of collaborative delivery routes between drones and trucks, thereby improving the performance of urban low-altitude logistics delivery systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies suffer from poor convergence efficiency and computational stability in air-ground collaborative task allocation. In particular, in drone and truck collaborative delivery systems, existing decomposition methods cannot effectively handle the time and space dependencies such as synchronous take-off and recovery between vehicles and drones, resulting in high-dimensional combinatorial characteristics of the solution space, which makes it difficult to achieve large-scale commercial applications.
A hierarchical divide-and-conquer approach is adopted to decompose the air-ground collaborative task into multiple low-dimensional sub-problems. By randomly generating delivery routes, performing crossover and mutation operations, and iteratively updating fitness values, the task grouping is optimized through clustering and pruning to form an efficient truck and drone collaborative delivery route.
It improves the convergence efficiency and computational stability of air-ground collaborative task allocation, can quickly generate high-quality candidate solutions, adapt to changes in delivery demand, and improve the operational performance and reliability of urban low-altitude logistics delivery systems.
Smart Images
Figure CN122155167A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of air-ground collaborative task allocation technology, and specifically to a large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer. Background Technology
[0002] In recent years, drone delivery has emerged as a potential solution to the "last mile" logistics problem in urban areas. However, independently operating drones are limited by insufficient payload capacity, short flight time, and restricted airspace (no-fly zones), hindering large-scale commercial application. The "air-ground collaborative delivery system" effectively overcomes these limitations by utilizing drones and trucks to operate collaboratively within a unified logistics network. In this system, trucks act as mobile supply stations, effectively extending the drones' operating radius; the drones, in turn, possess flexible and rapid last-mile delivery capabilities, thereby improving transportation efficiency.
[0003] In the joint optimization of heterogeneous vehicle fleets, the solution space exhibits high-dimensional combinatorial characteristics, leading to problems such as convergence difficulties, infeasibility of solutions, and excessive computation time, posing significant challenges to practical solutions. Therefore, dimensionality reduction through problem decomposition becomes a key means to ensure the scalability of the algorithm. The co-evolutionary framework, by dividing the original problem into multiple low-dimensional subproblems and allowing them to evolve independently, can effectively address large-scale combinatorial optimization problems.
[0004] The core challenge of co-evolutionary optimization lies in designing a reasonable decomposition strategy to maximize intra-group correlation while reducing inter-group interference. Existing spatial clustering-based decomposition methods, while achieving some success in vehicle path optimization, have significant limitations in air-ground cooperative task allocation. In air-ground cooperative task allocation models, there are temporal and spatial dependencies between vehicles and UAVs, such as synchronized takeoff and recovery, leading to overlapping sub-problems. Existing decomposition methods cannot effectively handle these complex relationships. Summary of the Invention
[0005] In view of the above problems, the present invention provides a large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer, which solves the technical problems of poor convergence efficiency and computational stability in the prior art.
[0006] This invention provides a method for large-scale air-ground cooperative task allocation based on hierarchical divide-and-conquer, comprising the following steps: Step S1: Determine the air-ground collaborative task groups, including: each group corresponds to one truck, and determine the customer points of the trucks in each group and the multiple drones carried by the trucks; Multiple delivery routes are randomly generated for each group, and each group corresponds to a subpopulation; the delivery routes are composed of multiple customer points. Step S2: Determine the corresponding megapaths from the current delivery routes of each subpopulation, perform crossover and mutation operations on the megapaths, and update the delivery routes; the updated delivery routes include truck delivery routes and drone delivery routes. Step S3: Evaluate the delivery routes of each subpopulation and obtain their fitness values; update the optimal delivery route based on the fitness values of the delivery routes; return to step S2 until the first maximum number of iterations is reached; Step S4: Obtain an edge set based on the current optimal delivery route, where the endpoints of the edges in the edge set are the customer points; calculate the routing distance between each edge in the edge set to obtain a distance matrix; Step S5: Cluster the distance matrix, update the air-ground collaborative task groups and the delivery routes of each group based on the clustering results; trim the customer points in the delivery routes of each group. Step S6: Return to step S4 until the second maximum number of iterations is reached, then determine the delivery route of each group at this point as the task allocation result.
[0007] Preferably, step S1 specifically includes: Step S1-1: Use the k-means clustering algorithm to group customer points based on their location; Step S1-2: Based on the customer point sequence in each group, randomly shuffle the order of the customer points and add warehouse points at both ends of the sequence to generate multiple delivery routes.
[0008] Preferably, step S2 specifically includes: Step S2-1: Add the customer points of the drone delivery route to the truck delivery route to form a megapath; Step S2-2: Perform a crossover operation on the giant path using a partial mapping crossover operator, and perform a mutation operation on the updated delivery route after the crossover using a hill-climbing method to update the delivery route.
[0009] Preferably, step S2-1 specifically includes: All warehouse points along the truck delivery route are removed, and customer points (excluding the beginning and end) along the drone delivery route are inserted sequentially after the takeoff point in the truck delivery route to form a megapath; the takeoff point is the first customer point in the drone delivery route. Step S2-2 specifically includes: Take any two giant paths as parent giant paths, and use a partial mapping cross method to cross the parent giant paths to generate child giant paths. Based on the drone delivery route of the superior parent megapath, the corresponding drone delivery route is recovered from the child megapath and deleted from the child megapath, and a truck delivery route is constructed based on the remaining customer points; the superior parent megapath refers to the delivery route with a higher fitness value corresponding to the superior parent megapath. The child megapath is subjected to mutation operations, including: truck node swapping, drone node swapping, truck-to-drone node swapping, truck node re-insertion, drone node re-insertion, and truck-to-drone node swapping.
[0010] Preferably, step S3 specifically includes: Step S3-1: Iterate through the customer points in the truck delivery route in sequence until the warehouse point is reached, and obtain the latest arrival time of the truck and drone at the warehouse point as the adaptation value; Step S3-2: Update the optimal delivery route based on the fitness value of the delivery route; Step S3-3: Return to step S2 until the first maximum number of iterations is reached.
[0011] Preferably, step S3-1 specifically includes: For each customer point, calculate the latest arrival time of the vehicle and drone, and use the latest arrival time as the departure time for driving to the next customer point; The expression for the latest arrival time of the vehicle is:
[0012] in, Indicates the latest arrival time. Indicates the number is The vehicle arrives at the customer point At that moment, This indicates taking the maximum value; The latest arrival time of the trucks and drones at the warehouse point obtained through the traversal is used as the fitness value.
[0013] Preferably, step S4 specifically includes: Step S4-1: Obtain adjacent pairs of customer points in the optimal delivery route, determine an edge by the pairs of customer points, and form the edge set by all the edges; Step S4-2: Compare each edge in the edge set pairwise and calculate the routing distance between the edges. The expression is:
[0014] in, express The route distance between them Indicates customer point With customer points The Euclidean distance between them; The distance matrix is formed by calculating the routing distance for all edge pairs in the edge set.
[0015] Preferably, step S5 specifically includes: Step S5-1: Use the k-medoids clustering method to cluster all edges of the distance matrix. After the clustering is completed, extract the user points contained in each edge group to form a new air-ground collaborative task group. Step S5-2: For overlapping nodes in the delivery routes of each current air-ground collaborative task group, calculate the removal savings, sort customer points according to the removal savings, and continuously delete customer points in descending order until each customer point appears only in a single group; the overlapping nodes represent customer points that appear in multiple groups at the same time.
[0016] Preferably, in step S5-2, the expression for the amount of savings removed is:
[0017] in, Indicates the amount of savings removed. and These represent the fitness values of the delivery routes before and after removing overlapping nodes, respectively.
[0018] Compared with the prior art, the present invention has at least the following beneficial effects: (1) This invention first randomly generates multiple delivery routes within each air-ground collaborative task group and forms subpopulations. Then, it uses giant paths to cross and mutate to synchronously update the delivery routes of trucks and drones, and uses fitness values to iteratively update the optimal solution until the first maximum number of iterations is reached. This process maps the coupled decision-making of trucks and multiple drones into an evolutionary path structure, which can quickly generate high-quality candidate solutions in a large search space, thereby improving convergence efficiency.
[0019] (2) This invention extracts the edge set based on the current optimal delivery route and calculates the routing distance matrix. Then, it clusters the distance matrix to update the air-ground collaborative task groups and their delivery routes, and prunes the customer points within the groups. This mechanism uses the evolved structural information (edge set and distance features) to decompose and perturb the problem, enabling rapid hierarchical decoupling of complex decision variables, reducing the problem size and intra-group search difficulty, avoiding blind iteration in the global scope, and improving the solvability and convergence efficiency of large-scale instances.
[0020] (3) By repeatedly clustering and reconstructing task grouping and iterative updating of delivery routes, this invention effectively adapts to the dynamic changes in delivery demand and customer point distribution, enhances the adaptability and stability of the solution, and ensures that the final task allocation result achieves the optimal delivery effect under the condition of satisfying the maximum number of iterations, which greatly improves the actual operation performance and reliability of the urban low-altitude logistics delivery system. Attached Figure Description
[0021] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.
[0022] Figure 1 This is a schematic diagram of the large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer provided by the present invention.
[0023] Figure 2 This is a schematic diagram of urban low-altitude logistics air-ground collaborative delivery provided by the present invention. Detailed Implementation
[0024] To better understand the above-described objectives, features, and advantages of the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other. Furthermore, the present invention can be implemented in other ways different from those described herein; therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.
[0025] like Figure 2 As shown, the air-ground collaborative task of this invention involves allocating routes for multiple trucks and multiple vehicle-mounted drones. In actual delivery, multiple truck fleets and their corresponding drone fleets are responsible for providing services to a group of customer points with predetermined locations and package needs. Each customer must be served only once, and the service can be provided by either a truck or a drone. Trucks can transport drones to locations near the drone's service area. After taking off from the truck, the drone can serve several customers within the constraints of payload capacity and battery energy, and then return to the designated truck for refueling and charging. Drones are only allowed to take off and land at nodes where trucks are located, and trucks and drones can wait for each other and rendezvous synchronously at the same customer point.
[0026] The requirements for the air-ground collaborative mission of this invention are as follows: (1) The objective is to minimize the total completion time for serving all customers; (2) The speeds of the trucks and drones are fixed; (3) The drone can only be launched and recovered at the customer's location where the truck is parked; (4) If the drone arrives at the rendezvous point first, it can temporarily land at the customer's location to wait for the truck without consuming energy; (5) Assume that the truck has sufficient fuel and storage space, while the drone is limited in terms of payload capacity and energy consumption.
[0027] This invention establishes various objective functions and constraints for air-ground collaborative tasks, as shown in Table 1.
[0028] Table 1
[0029]
[0030] The meanings of all symbols in Table 1 are explained below: Point set, including all customer points and one warehouse. , where 0 and This indicates the same warehouse, with 0 representing the originating warehouse. To return to the warehouse.
[0031] Customer point collection, . The set of customer points that includes the departure warehouse; This is a set of customer points that include those returning to the warehouse.
[0032] Truck collection ,in For the number of trucks.
[0033] A collection of drones ,in The number of drones carried by each truck.
[0034] The weight of the drone.
[0035] Maximum payload capacity of drones.
[0036] Maximum energy of the drone battery.
[0037] Any two points The distance between them.
[0038] Truck speed.
[0039] : Flight speed of the drone.
[0040] Customer Point The weight of the package.
[0041] : Decision variable, when truck From point Drive to point (and The value is 1 when the condition is met, and 0 otherwise.
[0042] : Decision variable, when drone From truck Take off, self-pointing Flight Point (and The value is 1 when the condition is met, and 0 otherwise.
[0043] : Decision variable, when drone At the customer point By truck The value is 0 when the device is launched, otherwise it is 1.
[0044] : Decision variable, when drone At the customer point by truck The value is 0 when it is recycled, otherwise it is 1.
[0045] : Decision variable, when drone Equipped in trucks And can be done at the customer's point The value is 1 when launching, and 0 otherwise.
[0046] Drones Equipped in trucks Arrival point The maximum remaining package weight.
[0047] Drones Equipped in trucks Arrival point The remaining energy at that time.
[0048] :truck Arrival at customer point The time.
[0049] :truck From the customer point The time of departure.
[0050] Drones Equipped in trucks Arrive at the customer's location The time.
[0051] The above objective functions and constraints of this invention can be divided into the following parts. First, objective function (1) aims to minimize the total completion time after serving all customers. Second, equations (2)–(6) describe the service constraints of the truck and the drone respectively, ensuring that each customer is visited only once and that the truck path satisfies the departure, return, and flow conservation conditions. Equations (7)–(20) are truck-drone coordination constraints, requiring that the take-off and landing nodes of the drone must be visited by the corresponding truck, thereby ensuring the feasibility of task allocation and launch and recovery operations. Equations (21)–(26) are time constraints, which specify the arrival, departure, and synchronization relationships of the truck and the drone. Subsequently, equations (27)–(29) are drone payload constraints, limiting the weight of the cargo carried by the drone during flight to not exceed the maximum payload capacity. Finally, equations (30)–(32) are drone energy constraints, ensuring that energy consumption during delivery does not exceed the battery capacity, thus ensuring flight sustainability.
[0052] The goal of this invention is to minimize the total completion time for serving all customers, achieve rapid decoupling of complex decision variables through a hierarchical divide-and-conquer strategy, and gradually form an efficient truck and drone collaborative delivery route through an inter-group collaborative optimization mechanism.
[0053] To solve this air-to-ground collaborative task allocation problem, this invention proposes a large-scale air-to-ground collaborative task allocation method based on hierarchical divide-and-conquer. To illustrate the effectiveness of the proposed method, a specific embodiment is provided below for detailed explanation of the above technical solution. A specific embodiment of this invention is as follows: Figure 1 As shown, a method for large-scale air-ground collaborative task allocation based on hierarchical divide-and-conquer is disclosed, and the specific implementation steps are as follows: Step S1: Determine the air-ground collaborative task groups, including: each group corresponds to one truck, and determine the customer points of the trucks in each group and the multiple drones carried by the trucks; For each group, initialize it and randomly generate multiple delivery routes. Each group corresponds to a subpopulation. The delivery routes are composed of multiple customer points, with warehouse points at both ends.
[0054] like Figure 2 As shown, the customer point represents the location where delivery service is required, and is represented by a non-zero number in this invention. The warehouse point represents the start and end points of the delivery, and is represented by the number 0 in this invention.
[0055] (1) Grouping customer locations In this step, the invention groups customer points using a k-means clustering algorithm. It should be noted that the number of groups corresponds to the number of trucks; each subgroup corresponds to one truck, and the customer points within that group are served by that truck and its multiple drones.
[0056] In some embodiments, clustering algorithms can be used to cluster customer points based on their locations, aggregating geographically close customer points into the same group, thereby determining the service area and customer point set for each truck.
[0057] (2) Group initialization In this step, based on the customer point sequence in each group, an initial truck delivery route is generated by randomly shuffling the order of the customer points and adding warehouse point "0" to both ends of the sequence. By repeating this random generation process, multiple non-repeating truck delivery routes are created for each group, and these truck delivery routes together constitute the initial subpopulation of that group.
[0058] The generated delivery route consists of multiple customer points arranged in a specific order, with warehouse points at both ends, representing the complete path of a truck starting from a warehouse, passing through several customer points, and returning to the warehouse.
[0059] The delivery routes of this invention include truck delivery routes and drone delivery routes. During initialization, only truck delivery routes are generated, while drone delivery routes will be generated in subsequent steps.
[0060] This initialization process provides a diverse starting solution space for subsequent route optimization, which is then applied to population optimization.
[0061] Step S2: Determine the corresponding megapaths from the current delivery routes of each subpopulation, perform crossover and mutation operations on the megapaths, and update the delivery routes; the updated delivery routes include truck delivery routes and drone delivery routes. This step specifically includes: Step S2-1: Add the customer points of the drone delivery route to the truck delivery route to form a megapath; In this step, the present invention establishes a megapath for each delivery route of each subpopulation. The specific steps include: removing all warehouse points in the truck delivery route, and sequentially inserting customer points (excluding the beginning and end) in the drone delivery route after the take-off point in the truck delivery route to form a megapath; the take-off point is the first customer point in the drone delivery route.
[0062] Through the above steps, this invention generates a giant path that encodes both truck and drone delivery routes into a single sequence, facilitating genetic operations. The giant path serves as the parent individual, and subsequent steps involve crossover and mutation operations to generate offspring individuals.
[0063] The following example illustrates the steps for constructing a megapath. Given delivery routes for parent 1: truck route (0,1,2,3,4,5,6,7,0), drone 1 route (1,8,4) and (5,9,7), and drone 2 route (1,10,4) and (6,11,12,7). Insert the 8s (excluding the first and last ones) after takeoff point 1 in drone 1 route (1,8,4) into the truck route after 1. Similarly, insert the 9s (excluding the first and last ones) after takeoff point 5 in drone 1 route (5,9,7) into the truck route after 5. Likewise, insert the 11s and 12s (excluding the first and last ones) after takeoff point 6 in drone 2 route (6,11,12,7) into the truck route after 6. Repeat this process for all insertions to obtain the megapath (1,8,10,2,3,4,5,9,6,11,12,7).
[0064] Step S2-2: Perform a crossover operation on the giant path using a partial mapping crossover operator, and perform a mutation operation on the updated delivery route after the crossover using a hill-climbing method to update the delivery route.
[0065] (1) Intersect the megapaths and restore them to truck delivery routes and drone delivery routes. In this step, any two giant paths in the subpopulation are recombined to generate offspring giant paths. Specifically, the partially mapped crossover (PMX) method is used to cross over the giant paths, producing new offspring individuals. The PMX method will not be described in detail in this invention.
[0066] The following example illustrates this: the first parent's giant path is (1,8,10,2,3,4,5,9,6,11,12,7), and the second parent's giant path is (3,9,5,8,12,1,11,7,4,2,6,10). After the two giant paths are partially mapped and cross-processed at the randomly selected cross-index "4", the resulting child giant path is (1,8,10,2,12,3,11,7,4,9,6,5).
[0067] Next, the truck delivery route and the drone delivery route are recovered using the offspring's giant path. Specifically, the superior parent individual among two parent individuals is first obtained. The superior parent individual refers to the parent individual with a higher fitness value, and the method for calculating the fitness value will be described later.
[0068] The recovery process includes: recovering the corresponding drone delivery routes from the offspring megapaths based on the better parent individuals' drone delivery routes and deleting them from the offspring megapaths; and constructing truck delivery routes based on the remaining customer points.
[0069] The following example illustrates the steps for restoring truck and drone delivery routes.
[0070] Parent 1 is a superior parent individual. Parent 1's delivery routes include: truck route (0,1,2,3,4,5,6,7,0), drone 1 routes (1,8,4) and (5,9,7), and drone 2 routes (1,10,4) and (6,11,12,7). Parent 1's megapath is (1,8,10,2,3,4,5,9,6,11,12,7), and the offspring's megapath is (1,8,10,2,12,3,11,7,4,9,6,5).
[0071] Based on the sequence indices (first, second, and sixth) of the parent generation's drone 1 route (1,8,4), the corresponding sequence from the child generation's giant path is (1,8,3), and (1,8,3) is taken as the first route of the child generation's drone 1. Based on the sequence indices (seventh, eighth, and twelfth) of the parent generation's drone 1 route (5,9,7), the corresponding sequence from the child generation's giant path is (11,7,5), and (11,7,5) is taken as the second route of the child generation's drone 1. Similarly, the two routes of the child generation's drone 2 can be obtained as (1,10,3) and (4,9,6,5). Then, remove customer points from the child drone delivery route except for the beginning and end from the child megapath. In this example, the child megapath (1,8,10,2,12,3,11,7,4,9,6,5) has 8,7,10,9,6 removed to get (1,2,12,3,11,4,5). Warehouse point 0 is added at the beginning and end to finally get the truck route (0,1,2,12,3,11,4,5,0).
[0072] Based on the above steps for generating offspring individuals, this invention generates multiple new delivery routes for each subpopulation and updates the current delivery routes.
[0073] (2) Perform a mutation operation on the updated delivery routes after crossover to update the delivery routes. After the crossover operation, the selected offspring are mutated to enhance population diversity and perform a local search. This invention employs a hill-climbing method for local search mutation of offspring, setting up six neighborhood search operators, including: Truck node swapping: Select two customer locations along a truck delivery route and swap them. Drone Node Switching: Select two customer points along the same or two drone delivery routes and switch their access order; Truck-Drone Node Swap: Select a customer point from a truck delivery route and a customer point from a drone delivery route, and swap the two customer points. Truck node re-insertion: Remove a customer point from a truck delivery route and re-insert that customer point into another location on the truck delivery route; Drone node re-insertion: Remove a customer point from the drone delivery route and insert that customer point into another location in the current drone or another drone delivery route; Truck-drone node swapping: Inserting a customer point on a truck delivery route into a drone delivery route; or inserting a drone delivery route back into a truck delivery route.
[0074] By combining these six neighborhood search operators, we can fully explore the solution space and improve the quality of the solution while maintaining the feasibility of the solution.
[0075] Through the crossover and mutation operations described above, the newly generated offspring individuals are used to update the parent population in subsequent iterations. Based on the fitness values of the offspring produced by crossover and mutation, individuals in the parent population are selectively replaced, completing one iteration. This iterative process is repeated continuously, causing the population to gradually evolve towards a better direction, and the quality of the delivery routes to continuously improve. It should be noted that the updated delivery routes include two parts: truck delivery routes and drone delivery routes. The truck delivery routes represent the order in which trucks visit customer points, while the drone delivery routes represent the complete flight path of the drone from the truck, through the customer point, and back to the truck.
[0076] Step S3: Evaluate the delivery routes of each subpopulation and obtain their fitness values; update the optimal delivery route based on the fitness values of the delivery routes; return to step S2 until the first maximum number of iterations is reached; Step S3-1: Iterate through the customer points in the truck delivery route in sequence until the warehouse point is reached, and obtain the latest arrival time of the truck and drone at the warehouse point as the adaptation value.
[0077] For each customer point during the traversal process, the delivery routes for each sub-group need to be evaluated to determine the merits of each route. The evaluation process uses the latest arrival time of the trucks and drones at the warehouse point along the delivery route as the fitness value. This time reflects the total time required to complete all delivery tasks and is a key indicator for measuring delivery efficiency. The specific evaluation steps for truck and drone delivery routes are as follows.
[0078] The process involves sequentially traversing customer points along the truck delivery route until reaching the warehouse point, including: for each customer point, calculating the latest arrival time for the vehicle and drone, and using the latest arrival time as the departure time for the next customer point.
[0079] The expression for the latest arrival time of the vehicle is:
[0080] in, Indicates the latest arrival time. Indicates the number is The vehicle arrives at the customer point At that moment, This indicates taking the maximum value.
[0081] This invention uses the latest arrival time at the warehouse point obtained through traversal as the adaptation value.
[0082] In some embodiments, specifically, the latest arrival time of the drone is expressed as:
[0083] in, The vehicle number is The drone arrived at the landing point At that moment, Indicates the takeoff point The departure time This indicates the drone's flight time.
[0084] Step S3-2: Update the optimal delivery route based on the fitness value of the delivery route; In this step, the invention updates the global optimal solution based on the fitness value of the delivery routes. Specifically, the delivery routes of multiple subproblems are concatenated in a predetermined order to generate the overall solution for the current iteration.
[0085] The overall solution encompasses complete delivery plans for all trucks and drones, covering all customer points requiring service. After generating the overall solution, its fitness value is calculated, which is the maximum delivery time cost across all subgroups. If the fitness value of the overall solution is better than the currently recorded global optimum, the global optimum variable is updated, and the current overall solution is saved as the new global optimum.
[0086] This update mechanism ensures that the algorithm can continuously track and save the optimal delivery solutions discovered during the evolution process.
[0087] Step S3-3: Return to step S2 until the first maximum number of iterations is reached.
[0088] After completing one iteration of evaluation and update, return to step S2 to continue the next round of crossover, mutation, and evaluation operations until the preset maximum number of iterations is reached. Through multiple iterations, a new global optimum is obtained.
[0089] Step S4: Obtain an edge set based on the current optimal delivery route, where the endpoints of the edges in the edge set are the customer points; calculate the routing distance between each edge in the edge set to obtain a distance matrix; In this step, the present invention decomposes the current optimal delivery route into a fine-grained structure according to the path structure, forming a set of basic elements composed of several "edges".
[0090] This invention establishes an edge set for customer points in the optimal delivery route. Specifically, it first obtains adjacent pairs of customer points in the delivery route, determines an edge by the pair of customer points, and then constructs the edge set by all the edges.
[0091] The edge set allows for overlapping nodes, meaning a node can be the endpoint of multiple edges simultaneously, thus preserving potential coupling relationships in the path structure. This decomposition method, which allows node sharing, can fully reflect the connection relationships and access patterns of nodes in the path. After decomposition, a set containing all edge elements is obtained, which serves as the basis for subsequent interactivity detection and grouping.
[0092] After obtaining the edge set, perform pairwise comparisons on all edges in the edge set and calculate the routing distance between the edges. For any two edges in the set... , The expression for calculating the routing distance is:
[0093] in, express The route distance between them , They represent The two vertices, Indicates customer point With customer points The Euclidean distance between them.
[0094] The routing distance is calculated for all edge pairs in the edge set to form a distance matrix. The elements in the distance matrix represent the routing distance between pairs of edges being compared.
[0095] The distance matrix reflects the overall structural characteristics of the edge set, identifies edge groups that are structurally close to each other or have high interaction, and provides a quantitative metric for subsequent clustering and grouping.
[0096] Step S5: Cluster the distance matrix, update the air-ground collaborative task groups and the delivery routes of each group based on the clustering results; trim the customer points in the delivery routes of each group. Based on the distance matrix, sub-problems are reconstructed to dynamically adjust task grouping. Specifically, the k-medoids clustering method is used to regroup all edges.
[0097] In some embodiments, the k-medoids clustering algorithm can be used with the distance matrix as input to aggregate similar edges into the same category.
[0098] After clustering is completed, the nodes contained in each edge group are extracted to form a new air-ground collaborative task group.
[0099] At this point, since there are overlapping nodes between the edges, and multiple edges containing overlapping nodes may be assigned to different clusters during the clustering process, there may be cases where the newly constructed subproblems contain overlapping nodes.
[0100] Some customer points may appear in multiple customer point sets within the same group; these customer points are called overlapping nodes. The existence of overlapping nodes indicates that these customer points are located at the intersection of multiple subproblems in the route structure and have the potential to be served by different trucks.
[0101] Regarding the shared node problem, this invention performs dynamic reallocation of shared nodes, including the following steps: For overlapping nodes in the delivery routes of each current air-to-ground collaborative task group, calculate the removal savings, whereby the removal savings represent the cost savings resulting from removing the node from a subproblem, expressed as:
[0102] in, Indicates the amount of savings removed. and These represent the fitness values of the delivery routes before and after removing overlapping nodes, respectively.
[0103] This invention sorts customer points according to the amount of savings removed, and continuously deletes customer points in descending order until each customer point appears in only a separate group, thus completing the pruning of customer points.
[0104] Step S6: Return to step S4 until the second maximum number of iterations is reached, then determine the delivery route of each group at this point as the task allocation result.
[0105] The present invention returns to step S4, repeats the dynamic decomposition and adjustment of task groups, and ends the optimization when the second maximum number of iterations is reached, and finally determines the latest delivery route as the task allocation result.
[0106] While the specific embodiments of the present invention depict actions or steps in a particular order, this should be understood as requiring such actions or steps to be performed in the shown specific order or sequential order, or requiring all illustrated actions or steps to be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations. The above descriptions are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention.
[0107] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for large-scale air-ground collaborative task allocation based on hierarchical divide-and-conquer, characterized in that, Includes the following steps: Step S1: Determine the air-ground collaborative task groups, including: each group corresponds to one truck, and determine the customer points of the trucks in each group and the multiple drones carried by the trucks; Multiple delivery routes are randomly generated for each group, and each group corresponds to a subpopulation; the delivery routes are composed of multiple customer points. Step S2: Determine the corresponding megapaths from the current delivery routes of each subpopulation, perform crossover and mutation operations on the megapaths, and update the delivery routes; the updated delivery routes include truck delivery routes and drone delivery routes. Step S3: Evaluate the delivery routes of each subpopulation and obtain their fitness values; update the optimal delivery route based on the fitness values of the delivery routes; return to step S2 until the first maximum number of iterations is reached; Step S4: Obtain an edge set based on the current optimal delivery route, where the endpoints of the edges in the edge set are the customer points; calculate the routing distance between each edge in the edge set to obtain a distance matrix; Step S5: Cluster the distance matrix, update the air-ground collaborative task groups and the delivery routes of each group based on the clustering results; trim the customer points in the delivery routes of each group. Step S6: Return to step S4 until the second maximum number of iterations is reached, then determine the delivery route of each group as the task allocation result.
2. The method for large-scale air-ground collaborative task allocation based on hierarchical divide-and-conquer as described in claim 1, characterized in that, Step S1 specifically includes: Step S1-1: Use the k-means clustering algorithm to group customer points based on their location; Step S1-2: Based on the customer point sequence in each group, randomly shuffle the order of the customer points and add warehouse points at both ends of the sequence to generate multiple delivery routes.
3. The large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer as described in claim 2, characterized in that, Step S2 specifically includes: Step S2-1: Add the customer points of the drone delivery route to the truck delivery route to form a megapath; Step S2-2: Perform a crossover operation on the giant path using a partial mapping crossover operator, and perform a mutation operation on the updated delivery route after the crossover using a hill-climbing method to update the delivery route.
4. The large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer as described in claim 3, characterized in that, Step S2-1 specifically includes: All warehouse points along the truck delivery route are removed, and customer points (excluding the beginning and end) along the drone delivery route are inserted sequentially after the takeoff point in the truck delivery route to form a megapath; the takeoff point is the first customer point in the drone delivery route. Step S2-2 specifically includes: Take any two giant paths as parent giant paths, and use a partial mapping cross method to cross the parent giant paths to generate child giant paths. Based on the drone delivery route of the superior parent megapath, the corresponding drone delivery route is recovered from the child megapath and deleted from the child megapath, and a truck delivery route is constructed based on the remaining customer points; the superior parent megapath refers to the delivery route with a higher fitness value corresponding to the superior parent megapath. The child megapath is subjected to mutation operations, including: truck node swapping, drone node swapping, truck-to-drone node swapping, truck node re-insertion, drone node re-insertion, and truck-to-drone node swapping.
5. The large-scale air-ground collaborative task allocation method based on hierarchical divide-and-conquer as described in claim 4, characterized in that, Step S3 specifically includes: Step S3-1: Iterate through the customer points in the truck delivery route in sequence until the warehouse point is reached, and obtain the latest arrival time of the truck and drone at the warehouse point as the adaptation value; Step S3-2: Update the optimal delivery route based on the fitness value of the delivery route; Step S3-3: Return to step S2 until the first maximum number of iterations is reached.
6. The method for large-scale air-ground collaborative task allocation based on hierarchical divide-and-conquer as described in claim 5, characterized in that, Step S3-1 specifically includes: For each customer point, calculate the latest arrival time of the vehicle and drone, and use the latest arrival time as the departure time for driving to the next customer point; The expression for the latest arrival time of the vehicle is: in, Indicates the latest arrival time. Indicates the number is The vehicle arrives at the customer point At that moment, This indicates taking the maximum value; The latest arrival time of the trucks and drones at the warehouse point obtained through the traversal is used as the fitness value.
7. The large-scale air-ground cooperative task allocation method based on hierarchical divide-and-conquer as described in claim 6, characterized in that, Step S4 specifically includes: Step S4-1: Obtain adjacent pairs of customer points in the optimal delivery route, determine an edge by the pairs of customer points, and form the edge set by all the edges; Step S4-2: Compare each edge in the edge set pairwise and calculate the routing distance between the edges. The expression is: in, express The route distance between them Indicates customer point With customer points The Euclidean distance between them; The distance matrix is formed by calculating the routing distance for all edge pairs in the edge set.
8. The large-scale air-ground cooperative task allocation method based on hierarchical divide-and-conquer as described in claim 7, characterized in that, Step S5 specifically includes: Step S5-1: Use the k-medoids clustering method to cluster all edges of the distance matrix. After the clustering is completed, extract the user points contained in each edge group to form a new air-ground collaborative task group. Step S5-2: For overlapping nodes in the delivery routes of each current air-ground collaborative task group, calculate the removal savings, sort customer points according to the removal savings, and continuously delete customer points in descending order until each customer point appears only in a single group; the overlapping nodes represent customer points that appear in multiple groups at the same time.
9. The method for large-scale air-ground collaborative task allocation based on hierarchical divide-and-conquer as described in claim 8, characterized in that, In step S5-2, the expression for the amount of savings removed is: in, Indicates the amount of savings removed. and These represent the fitness values of the delivery routes before and after removing overlapping nodes, respectively.