Vehicle-machine collaborative delivery method and system based on hybrid fulfillment and heterogeneous coding

Abstract

The application relates to the technical field of unmanned delivery path planning and intelligent optimization algorithm, and discloses a vehicle-machine cooperative delivery method and system based on hybrid performance and heterogeneous coding. The method comprises the following steps: according to the distance and demand weight of customers to a delivery center, the customers are divided into a pure unmanned aerial vehicle direct delivery mode and a vehicle-machine cooperative delivery mode; discrete candidate convergence points are generated by clustering the vehicle-machine cooperative delivery customers; a three-layer heterogeneous coding chromosome is constructed; a genetic algorithm is used for iterative evolution; when the convergence point activation state changes, three-layer coding linkage update is triggered, the customer ownership and the unmanned vehicle path are synchronously adjusted; after consistency, load and voyage and soft time window checking and repair are sequentially performed, an optimal delivery scheme is output. Through hybrid performance prediction and three-layer coding linkage update, the application realizes cooperative optimization of heterogeneous decision variables and maintenance of coding consistency, and reduces the delivery cost and the time window violation degree.

Description

Technical Field

[0001] This invention belongs to the field of unmanned delivery route planning and intelligent optimization algorithm technology, and specifically relates to a vehicle-machine collaborative delivery method and system based on hybrid fulfillment and heterogeneous coding. Background Technology

[0002] In recent years, the collaborative delivery model of unmanned vehicles and drones has become an important technological approach to solve the problems of high costs and poor timeliness in urban last-mile logistics delivery due to its combination of the advantages of unmanned vehicles (high load capacity and long endurance) and drones (high mobility and low-altitude direct delivery). This model, through the coordinated use of air and ground vehicles, can effectively meet the dual requirements of heavy-load, long-distance transportation and flexible, rapid delivery to customers, demonstrating significant economic benefits and application prospects.

[0003] In air-to-ground collaborative delivery systems, meeting point decision-making is a core link connecting unmanned vehicles (UAVs) and drones. Existing meeting point decision-making methods mainly fall into two categories. The first is the continuous spatial dynamic meeting mode, which allows UAVs and UAVs to complete material handover at any location within a two-dimensional plane. However, the dimensionality of decision variables in this type of method increases exponentially with the scale of customers, resulting in excessively high computational complexity, making it difficult to support the real-time solution requirements of medium-sized and larger-scale delivery networks. Continuous spatial dynamic meeting is impractical for scenarios like urban last-mile delivery, which require real-time performance. The second is the fixed docking point mode, which pre-defines the meeting point as a finite number of fixed candidate locations, reducing the computational burden by discretizing and compressing the decision space. However, once the meeting point location is determined, this type of method cannot dynamically adjust with changes in customer spatial distribution, easily leading to excessively long UAV flight distances and a lack of adaptability to customer distribution in the meeting point layout.

[0004] Furthermore, regarding customer fulfillment models, existing research generally assumes that all customers must go through a complete collaborative process of "unmanned vehicle transfer to the rendezvous point, and drone take off from the rendezvous point to complete the last-mile delivery." Current methods uniformly adopt a vehicle-machine collaborative model for delivery to all customers. However, for customers located close to the distribution center and with lighter deliveries, direct delivery from the distribution center by drone is often more economical and efficient, eliminating the need for unmanned vehicle transfer. Forcing the use of the vehicle-machine collaborative process for short-distance, lightly loaded customers will lead to a waste of unmanned vehicle transportation resources and increase the overall operating costs of the system.

[0005] In summary, current technologies lack a delivery technology solution that can simultaneously consider the adaptability of meeting point layout to customer spatial distribution, the endogenous differentiation of delivery modes based on customer characteristics, and the collaborative optimization of heterogeneous decision variables within a unified framework. In particular, when the activation state of a meeting point changes, existing methods typically require recalculating customer affiliation and vehicle routes in stages, or treating these adjustments as separate post-hoc corrective steps. There is a lack of an integrated coding layer linkage mechanism that uses activation state changes as triggering events to synchronously update the meeting point activation layer, customer affiliation layer, and autonomous vehicle route layer.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] To address or at least alleviate one or more of the above problems, a vehicle-machine collaborative delivery method and system based on hybrid fulfillment and heterogeneous coding is provided. This method not only enables adaptive adjustment of the meeting layout to customer distribution through the dynamic activation of discrete candidate meeting points, but also endogenously integrates pure drone delivery and vehicle-machine collaborative delivery into the same optimization framework through a hybrid fulfillment pre-division mechanism. Furthermore, it achieves the co-evolution of three types of variables—meeting point activation, customer assignment, and vehicle route—through three-layer heterogeneous coding. By establishing a three-layer coding linkage update mechanism triggered by changes in the activation state of meeting points, it achieves synchronous linkage updates and consistency maintenance of the activation layer, the attribution layer, and the route layer, ultimately outputting an optimal delivery solution that balances economy and timeliness.

[0008] To achieve the above objectives, according to a first aspect of the present invention, a vehicle-machine collaborative delivery method based on hybrid fulfillment and heterogeneous coding is provided, comprising the following steps:

[0009] S1. Obtain delivery network data, perform mixed fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode.

[0010] S2. Cluster the coordinates of customers in the vehicle-machine collaborative delivery mode to generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points.

[0011] S3. Construct a three-layer heterogeneous coding chromosome to uniformly represent the activation status of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which unmanned vehicles visit the rendezvous point.

[0012] S4. A genetic algorithm is used to iteratively evolve the population composed of the three-layer heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer affiliation coding layer and the unmanned vehicle path coding layer are performed respectively.

[0013] S5. After the crossover or mutation operation, check whether the activation status of each candidate merging point in the merging point activation coding layer has changed; when a candidate merging point changes from an active state to an inactive state, all vehicle-machine collaborative delivery customers currently belonging to that merging point are reassigned to other activated merging points that meet the load and range constraints, and the merging point is deleted from the unmanned vehicle path coding layer; when a candidate merging point changes from an inactive state to an active state, select the customer closest to that merging point from the vehicle-machine collaborative delivery customers currently belonging to other activated merging points, update their affiliation to the new activated merging point, and insert that merging point into the unmanned vehicle path coding layer according to the nearest neighbor principle; perform synchronous write-back on the merging point activation coding layer, customer affiliation coding layer, and unmanned vehicle path coding layer to form the updated complete chromosome;

[0014] S6. Perform the following checks on the updated chromosome in sequence: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Repair the chromosome accordingly based on the verification results and calculate the fitness of the repaired chromosome.

[0015] S7. Iterate with the weighted sum of delivery time, operating cost and time window penalty as the optimization objective. When the preset convergence condition is met, decode the optimal chromosome and output the final delivery plan.

[0016] Furthermore, the delivery network data includes the location of the delivery center, the location and weight of the goods, the customer's soft time window, the location of the candidate meeting point, the speed and unit operating cost of the unmanned vehicle and drone, the maximum payload and maximum range of the drone, and the battery capacity and safety reserve power of the unmanned vehicle.

[0017] Furthermore, the pre-partitioning of hybrid performance includes:

[0018] Calculate the Euclidean distance from each customer to the distribution center; when the Euclidean distance from the customer to the distribution center is less than a preset distance threshold and the customer's required weight is less than a preset weight threshold, mark the customer as a pure drone direct delivery mode; otherwise, mark it as a vehicle-machine collaborative delivery mode.

[0019] For customers marked as using the pure drone delivery mode, the drone takes off directly from the distribution center to complete the delivery and return, without occupying the resources of the unmanned vehicle and the rendezvous point; for customers marked as using the vehicle-machine collaborative delivery mode, the unmanned vehicle transports the delivery goods to the corresponding rendezvous point, and then the drone takes off from the rendezvous point to complete the last delivery.

[0020] The pure drone delivery mode and the vehicle-machine collaborative delivery mode coexist throughout the optimization process, and the mode label of each customer remains unchanged after the pre-division is completed.

[0021] Furthermore, the three-layered heterogeneous coding chromosome specifically includes:

[0022] A merge point activation encoding layer is used to represent the activation state of each candidate merge point using binary encoding. Its length is equal to the total number of candidate merge points. Each bit in the string corresponds to a candidate merge point. A value of 1 indicates that the candidate merge point is in an active state, and a value of 0 indicates that the candidate merge point is in an inactive state.

[0023] The customer affiliation coding layer uses integer encoding to represent the rendezvous point number to which each vehicle-machine collaborative delivery customer belongs. Its length is equal to the total number of customers in the vehicle-machine collaborative delivery mode. The value of each element in the array represents the rendezvous point number to which the corresponding customer is assigned. Customers can only be assigned to rendezvous points that are in an active state in the rendezvous point activation coding layer.

[0024] An autonomous vehicle path coding layer is used to represent the order in which autonomous vehicles visit activated meeting points. Its length is equal to the number of meeting points in the activation coding layer that are active. The elements in the sequence represent the order in which autonomous vehicles visit activated meeting points and only include meeting points with a value of 1 in the activation coding layer.

[0025] The three coding layers together form a complete chromosome, share the same fitness function for evaluation, and the lengths of the customer home coding layer and the autonomous vehicle path coding layer are dynamically adjusted according to the changes in the activation state in the merging point activation coding layer.

[0026] Furthermore, discrete candidate meeting points and initial attribution relationships are generated, including:

[0027] K-means clustering is performed on the two-dimensional spatial coordinates of all customers in the vehicle-machine collaborative delivery mode. Each cluster center is used as the initial discrete candidate merging point, and customers belonging to the same cluster are initially assigned to the corresponding candidate merging point.

[0028] Starting from the distribution center, all activated meeting points are grouped into an unvisited set. Each time, the meeting point closest to the current location is selected from the unvisited set, added to the unmanned vehicle path, and the meeting point is used as the new current location. This process is repeated until all activated meeting points are visited. Finally, the vehicle returns to the distribution center to obtain the initial closed-loop path of the unmanned vehicle.

[0029] By repeatedly performing customer clustering, initial affiliation assignment, and nearest neighbor path generation operations, a preset number of initial individuals are generated, and these initial individuals form the initial population of the genetic algorithm.

[0030] Furthermore, when a candidate merging point changes from an active state to an inactive state, it includes:

[0031] Traverse the customer affiliation coding layer to find all vehicle-machine collaborative delivery customers currently belonging to the closure rendezvous point and form a set of customers to be reassigned.

[0032] For each customer in the set of customers to be reassigned, calculate the distance from the customer to each of the other activated rendezvous points, prioritize the nearest activated rendezvous point as the candidate target rendezvous point, and check whether the drone delivery task corresponding to the candidate target rendezvous point meets the maximum load and maximum range requirements after the customer is transferred to the candidate target rendezvous point. If not, select the next closest activated rendezvous point in sequence.

[0033] After completing the attribution update for all customers to be reassigned, update the rendezvous point number of the corresponding customer in the customer attribution coding layer to the new target rendezvous point number, and delete the elements corresponding to the closed rendezvous points from the autonomous vehicle path coding layer.

[0034] Furthermore, the specific handling method when a candidate merging point changes from an inactive state to an active state is as follows:

[0035] Calculate the distance from each vehicle-machine collaborative delivery customer to the newly activated rendezvous point. From the customers currently belonging to other activated rendezvous points, select the customer closest to the newly activated rendezvous point and update their affiliation in the customer affiliation coding layer to the newly activated rendezvous point.

[0036] The newly activated merging point is added to the set of activated merging points. Starting from the distribution center, all the updated activated merging points are combined into an unvisited set. Each time, the unvisited merging point closest to the current location is selected and added to the autonomous vehicle path, which is then used as the new current location. This process is repeated until all activated merging points have been visited. Finally, the system returns to the distribution center to obtain the updated autonomous vehicle path coding layer.

[0037] Furthermore, the crossover and mutation operations adapted to the characteristics of each coding layer are as follows:

[0038] The co-activation coding layer at the confluence point employs two-point crossover and bit-flip mutation. In the two-point crossover, two crossover positions are randomly selected and the binary segments between the two parent individuals are swapped. In the bit-flip mutation, a gene bit is randomly selected and its binary value is inverted.

[0039] The customer affiliation coding layer employs single-point crossover and single-gene random replacement mutations. In the single-point crossover, a crossover position is randomly selected and the integer segments following the crossover position of the two parent individuals are swapped. In the single-gene random replacement mutation, a gene position is randomly selected and replaced with the number of another activated confluence point.

[0040] The path coding layer of the unmanned vehicle employs partial matching crossover and exchange mutation. In partial matching crossover, two crossover positions are randomly selected, and the permutation segments of the two parent individuals between the two crossover positions are exchanged. Repeated elements are repaired by partial mapping. In exchange mutation, two gene positions are randomly selected and their positions are exchanged.

[0041] Furthermore, multi-level constraint verification and repair include:

[0042] First, a consistency check is performed between the meeting point, customer assignment, and autonomous vehicle route: This check whether each customer in the customer assignment coding layer has been assigned to an active meeting point in the meeting point activation coding layer; whether the autonomous vehicle route coding layer contains inactive meeting points; and whether all active meeting points undertaking customer delivery tasks are included in the autonomous vehicle route coding layer. If any inconsistency exists, the customer is reassigned or the autonomous vehicle route is corrected.

[0043] After the consistency verification is passed, the drone load and range verification is performed: the total weight of goods and the total flight distance of all drone delivery tasks corresponding to each activated rendezvous point are calculated to determine whether they exceed the maximum load and maximum range limits of the drone; if the limits are exceeded, the customers who exceed the limits will be transferred to other nearby activated rendezvous points that do not exceed the limits, and the load and range verification will be performed again.

[0044] After the load and range verification are passed, customer soft time window verification is performed: check whether the actual delivery completion time of each customer is later than the preset latest delivery time; for customers whose lateness exceeds the preset tolerance range, the customer is reassigned to an activated rendezvous point that the unmanned vehicle can reach earlier, or the access order of the customer's corresponding rendezvous point in the unmanned vehicle path coding layer is adjusted to reduce the degree of time window default.

[0045] To achieve the above objectives, according to a second aspect of the present invention, a vehicle-machine collaborative delivery system based on hybrid fulfillment and heterogeneous coding is provided, comprising:

[0046] The hybrid fulfillment pre-segmentation module is used to acquire delivery network data, perform hybrid fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode;

[0047] The discrete meeting point management module is used to cluster the coordinates of customers in the vehicle-machine collaborative delivery mode, generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points.

[0048] The three-layer co-evolution module is used to construct a three-layer heterogeneous coding chromosome to uniformly represent the activation state of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which the unmanned vehicle visits the rendezvous point.

[0049] A genetic algorithm is used to iteratively evolve the population composed of the three heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer affiliation coding layer and the autonomous vehicle path coding layer are performed respectively.

[0050] After the crossover or mutation operation, the activation status of each candidate merging point in the merging point activation coding layer is checked to see if it has changed. When a candidate merging point changes from an active state to an inactive state, all vehicle-to-machine cooperative delivery customers currently belonging to that merging point are reassigned to other active merging points that meet the load and range constraints, and the merging point is deleted from the autonomous vehicle path coding layer. When a candidate merging point changes from an inactive state to an active state, the customer closest to that merging point is selected from the vehicle-to-machine cooperative delivery customers currently belonging to other active merging points, their affiliation is updated to the new active merging point, and the merging point is inserted into the autonomous vehicle path coding layer according to the nearest neighbor principle. Synchronous write-back is performed on the merging point activation coding layer, customer affiliation coding layer, and autonomous vehicle path coding layer to form the updated complete chromosome.

[0051] The multi-constraint verification module is used to sequentially perform the following checks on the updated chromosome: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Based on the verification results, the chromosome is repaired accordingly, and the fitness of the repaired chromosome is calculated.

[0052] The delivery solution output module iterates with the optimization objective of minimizing the weighted sum of delivery time, operating costs, and time window penalties. When the preset convergence condition is met, it decodes the optimal chromosome and outputs the final delivery solution.

[0053] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0054] This invention compresses the continuous decision space into a finite set of candidate points through dynamic activation of discrete candidate meeting points and a hybrid pre-partitioning mechanism for fulfillment. It also pre-distributes pure drone delivery orders based on customer distance and material weight, thereby reducing computational complexity and avoiding wasted capacity. Simultaneously, it employs a three-layer heterogeneous coding and linkage update mechanism, enabling meeting point activation, customer assignment, and path order to co-evolve within the same chromosome, ensuring decision consistency. Combined with multi-level constraint verification and a weighted single objective function, it improves solution quality and engineering feasibility while closely aligning with actual timeliness and cost trade-offs. All modules are decoupled through standardized interfaces, exhibiting good scalability and maintainability.

[0055] Specifically, this invention compresses the merging decision space from an infinite continuous domain to a finite set of discrete candidate points through a dynamic activation mechanism for discrete candidate merging points, significantly reducing computational complexity. At the same time, by dynamically optimizing the activation bits of merging points in the encoding with iteration, the merging layout can be adaptively adjusted according to the customer spatial distribution, demonstrating clear engineering practicality and scenario adaptability.

[0056] Specifically, this invention uses a hybrid pre-segmentation mechanism to classify customers into two categories—pure drone direct delivery and vehicle-machine collaborative delivery—before coding is generated, based on the distance of the customer from the distribution center and the weight of the goods. Pure drone direct delivery customers do not need to be transferred by unmanned vehicles and can be delivered directly from the distribution center by drones. This avoids the problem of wasted transportation capacity for short-distance, lightly loaded customers caused by using the vehicle-machine collaborative process for all customers, and effectively reduces the total operating cost of the system.

[0057] Specifically, this invention uses a three-layer heterogeneous coding mechanism to encode three types of heterogeneous decision variables—the activation state of the meeting point, the drone assignment relationship from the customer to the meeting point, and the order in which the unmanned vehicle visits the meeting point—into the same chromosome. The three-layer coding shares the same fitness function for co-evolution during the evolution process, avoiding the problems of fragmented variable coupling relationships and local optima caused by staged optimization or independent coding, thus improving the solution quality.

[0058] Specifically, this invention uses a three-layer coding linkage update mechanism triggered by changes in the activation status of a rendezvous point. When a rendezvous point changes from active to inactive, customers are automatically reassigned and the rendezvous point is removed from the path. When a rendezvous point changes from inactive to active, nearby customers are automatically assigned to that rendezvous point and inserted into the path using the nearest neighbor algorithm. This ensures that the updates of the activation layer, the assignment layer, and the path layer are completed synchronously, structurally avoiding inconsistencies between the three layers of coding.

[0059] Specifically, this invention employs a multi-level constraint processing mechanism that sequentially executes consistency checks, load and range checks, and soft time window checks to perform targeted checks and repairs on the consistency and physical constraints among the three types of decision variables. This avoids the inherent defects of the traditional penalty function method, which requires repeated parameter tuning and cannot guarantee a feasible solution, thereby improving the engineering practicality and convergence efficiency of the algorithm.

[0060] Specifically, this invention incorporates delivery time, operating costs, and time window penalties into a unified weighted single objective function, and applies a linear penalty only to late customers in the time window processing, while early customers do not incur penalty costs, making the objective function more in line with the comprehensive trade-off between timeliness and economy in actual operating scenarios.

[0061] Specifically, this invention divides the discrete convergence point management module, the three-layer collaborative evolution module, and the multi-constraint verification module into independent interactive functional units. Data is transmitted between the modules through standardized input and output interfaces, giving the system good scalability and maintainability.

[0062] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0063] The accompanying drawings, as part of this invention, are provided to further illustrate the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation thereof. Clearly, the drawings described below are merely some embodiments; those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0064] In the attached diagram:

[0065] Figure 1 This is a flowchart illustrating the vehicle-machine collaborative delivery method based on hybrid fulfillment and heterogeneous coding in this specific embodiment;

[0066] Figure 2 This is a flowchart illustrating the hybrid performance pre-division mechanism in this specific embodiment;

[0067] Figure 3 This is a schematic diagram of the structure of the three-layer heterogeneous coding chromosome in this specific embodiment;

[0068] Figure 4 This is a schematic diagram of the overall process of the improved genetic algorithm in this specific embodiment;

[0069] Figure 5 This is a comparative diagram of the spatial layout of the meeting point before and after optimization in this specific embodiment, where (a) is the initial layout before optimization and (b) is the layout after optimization.

[0070] Figure 6 This is a schematic diagram comparing the effects of customer soft time window default repair before and after in this specific embodiment, wherein (a) is a schematic diagram of the effect before customer soft time window default repair, and (b) is a schematic diagram of the effect after customer soft time window default repair.

[0071] Figure 7 This is a schematic diagram of the vehicle-machine collaborative delivery system based on hybrid fulfillment and heterogeneous coding in this specific embodiment. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0073] In the following description, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0074] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0075] Please see Figure 1 This invention provides a vehicle-machine collaborative delivery method based on hybrid fulfillment and heterogeneous coding, comprising the following steps:

[0076] S1. Obtain delivery network data, perform mixed fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode.

[0077] S2. Cluster the coordinates of customers in the vehicle-machine collaborative delivery mode to generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points.

[0078] S3. Construct a three-layer heterogeneous coding chromosome to uniformly represent the activation state of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which unmanned vehicles visit the rendezvous point.

[0079] S4. A genetic algorithm is used to iteratively evolve the population composed of the three-layer heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer home coding layer and the unmanned vehicle path coding layer are performed respectively.

[0080] S5. After the crossover or mutation operation, check whether the activation status of each candidate merging point in the merging point activation coding layer has changed; when a candidate merging point changes from an active state to an inactive state, all vehicle-machine collaborative delivery customers currently belonging to that merging point are reassigned to other activated merging points that meet the load and range constraints, and the merging point is deleted from the unmanned vehicle path coding layer; when a candidate merging point changes from an inactive state to an active state, select the customer closest to that merging point from the vehicle-machine collaborative delivery customers currently belonging to other activated merging points, update their affiliation to the new activated merging point, and insert that merging point into the unmanned vehicle path coding layer according to the nearest neighbor principle; perform synchronous write-back on the merging point activation coding layer, customer affiliation coding layer, and unmanned vehicle path coding layer to form the updated complete chromosome;

[0081] S6. Perform the following checks on the updated chromosome in sequence: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Repair the chromosome accordingly based on the verification results and calculate the fitness of the repaired chromosome.

[0082] S7. Iterate with the weighted sum of delivery time, operating costs and time window penalties as the optimization objective. When the preset convergence condition is met, decode the optimal chromosome and output the final delivery plan.

[0083] Without contradiction, the technical features described in this embodiment can be combined with each other. In this embodiment, the "meeting point" refers to a discrete candidate location where the unmanned vehicle stops and completes the handover of materials with the drone.

[0084] Please see Figure 1 The unmanned vehicle-drone collaborative delivery scenario involved in this embodiment includes a distribution center, multiple customer locations, a finite number of discrete candidate meeting points, at least one unmanned vehicle, and multiple drones. Customers are divided into pure drone direct delivery customers and vehicle-drone collaborative delivery customers based on their distance to the distribution center and the weight of their goods. Figure 1 In this diagram, the distribution center is denoted as N0, candidate rendezvous points are denoted as M1 and M2, and example customer points are denoted as C1, C2, C3, and C4. M1 is an inactive candidate rendezvous point, and M2 is an active candidate rendezvous point. C1, C2, and C3 are customers using vehicle-to-machine (V2M) collaborative delivery; the delivery goods are transported by an autonomous vehicle to the active rendezvous point M2, and then the last-mile delivery is completed by a drone. C4 is a customer using pure drone delivery; the drone takes off directly from distribution center N0 to complete the delivery and return. The inactive candidate rendezvous point M1 does not undertake customer delivery tasks and is not included in the autonomous vehicle's access path.

[0085] This embodiment selects a standard delivery case including one distribution center, 40 customer points, 6 discrete candidate meeting points, and 4 drones for verification. The coordinates of the distribution center are set to (125, 125). The spatial coordinates of the 40 customer points are randomly and evenly distributed within the plane area [0, 250] × [0, 250]. The weight of the goods to be delivered to each customer is randomly generated within the range of 1 to 10 kg. The customer service time window is uniformly set within the service period of 8:00 to 16:00, and the available service window for each customer varies from 1 to 3 hours.

[0086] Key operational and technical parameters are set as follows: Cost per unit distance traveled by the autonomous vehicle C g =2.5 yuan / km, cost of drone flight per unit distance C a =1.8 yuan / km; unmanned vehicle speed v g =40km / h, drone flight speed v a =60km / h; Maximum payload of the drone W d =30kg, maximum single flight range R d =200km; Time window default penalty coefficient C t =10 yuan / minute. Genetic algorithm parameters: population size P=100, maximum number of iterations G=200, crossover probability P c =0.8, mutation probability P m =0.05, number of cluster centers K=6.

[0087] Hybrid performance pre-partition

[0088] In some embodiments, hybrid performance pre-division includes:

[0089] Calculate the Euclidean distance from each customer to the distribution center; when the Euclidean distance from the customer to the distribution center is less than a preset distance threshold and the customer's required weight is less than a preset weight threshold, mark the customer as a pure drone direct delivery mode; otherwise, mark it as a vehicle-machine collaborative delivery mode.

[0090] For customers marked as using the pure drone delivery mode, the drone takes off directly from the distribution center to complete the delivery and return, without occupying the resources of the unmanned vehicle and the rendezvous point; for customers marked as using the vehicle-machine collaborative delivery mode, the unmanned vehicle transports the delivery goods to the corresponding rendezvous point, and then the drone takes off from the rendezvous point to complete the last delivery.

[0091] The pure drone delivery mode and the vehicle-machine collaborative delivery mode coexist throughout the optimization process, and the mode label of each customer remains unchanged after the pre-division is completed.

[0092] Specifically, please see Figure 2First, the delivery network data is acquired. This data includes the location of the delivery center, the location of the customer, the weight of the customer's demand, the customer's soft time window, the location of the candidate meeting point, and parameters such as the operating speed of unmanned vehicles and drones, the unit operating cost, the maximum payload and maximum range of drones.

[0093] Based on the distance from the customer to the distribution center and their required weight, all customers are pre-divided into mixed fulfillment groups. Let the coordinates of the distribution center be (x0, y0), and the coordinates of the i-th customer be (x0, y0). i ,y i Its required weight is First, calculate the Euclidean distance from the customer to the distribution center. :

[0094] ;

[0095] Where (x0, y0) are the coordinates of the distribution center, (x i ,y i () represents the coordinates of the i-th customer.

[0096] When the distance from the customer to the distribution center Less than the preset distance threshold And the customer requires weight Less than the preset weight threshold At that time, the customer is marked as a pure drone delivery mode; other customers are marked as a vehicle-machine collaborative delivery mode.

[0097] In this embodiment, the distance threshold Set to 60 km, weight threshold Set to 3 kg. For customers who receive direct delivery by drone, the drone takes off directly from the distribution center to complete the delivery and return, without occupying the resources of unmanned vehicles and rendezvous points; for customers who receive vehicle-to-drone collaborative delivery, the unmanned vehicle transports the goods to the corresponding rendezvous point, and then the drone takes off from the rendezvous point to complete the last-mile delivery.

[0098] After the above pre-classification, 16 of the 40 customers adopted the pure drone direct delivery mode, with an average distance of 42.3 km and an average demand weight of 1.6 kg; the remaining 24 customers adopted the vehicle-machine collaborative delivery mode, with an average distance of 108.7 km and an average demand weight of 5.3 kg.

[0099] Both delivery modes coexist in the subsequent optimization process, and the customer's fulfillment mode label remains unchanged. By pre-separating lightly loaded customers in close proximity, the number of customers who need to participate in the rendezvous point assignment and autonomous vehicle path optimization can be reduced, thereby reducing the scale of subsequent coding and solution.

[0100] Candidate meeting points and initial population generation

[0101] In some embodiments, generating discrete candidate meeting points and initial attribution relationships includes:

[0102] K-means clustering is performed on the two-dimensional spatial coordinates of all customers in the vehicle-machine collaborative delivery mode. Each cluster center is used as the initial discrete candidate merging point, and customers belonging to the same cluster are initially assigned to the corresponding candidate merging point.

[0103] Starting from the distribution center, all activated meeting points are grouped into an unvisited set. Each time, the meeting point closest to the current location is selected from the unvisited set, added to the unmanned vehicle path, and the meeting point is used as the new current location. This process is repeated until all activated meeting points are visited. Finally, the vehicle returns to the distribution center to obtain the initial closed-loop path of the unmanned vehicle.

[0104] By repeatedly performing customer clustering, initial affiliation assignment, and nearest neighbor path generation operations, a preset number of initial individuals are generated, and these initial individuals form the initial population of the three-layer heterogeneous encoded chromosome of the genetic algorithm.

[0105] Specifically, K-means clustering was performed on the two-dimensional coordinates of 24 vehicle-to-machine collaborative delivery customers, and the cluster centers were used as initial discrete candidate merging points. The number of clusters was set to 6, thus generating 6 initial candidate merging points, and customers within the same cluster were initially assigned to the corresponding merging points.

[0106] After determining the initial set of activated meeting points, the nearest neighbor algorithm is used to generate the initial closed-loop path for the autonomous vehicle. Specifically, the distribution center is used as the initial current location, and all the initial activated meeting points are combined into an unvisited set. Each time, the meeting point closest to the current location is selected from the unvisited set, added to the autonomous vehicle's path, and set as the new current location. This process is repeated until all activated meeting points have been visited, and finally, the autonomous vehicle returns to the distribution center.

[0107] By repeatedly performing customer clustering, initial affiliation assignment, and nearest neighbor path generation operations, a preset number of initial individuals are formed, and these initial individuals constitute the initial population of the genetic algorithm.

[0108] Construction of a three-layer heterogeneous coding chromosome

[0109] In some embodiments, the three-layer heterogeneous coding chromosome specifically includes:

[0110] A merge point activation encoding layer is used to represent the activation state of each candidate merge point using binary encoding. Its length is equal to the total number of candidate merge points. Each bit in the string corresponds to a candidate merge point. A value of 1 indicates that the candidate merge point is in an active state, and a value of 0 indicates that the candidate merge point is in an inactive state.

[0111] The customer affiliation coding layer uses integer encoding to represent the rendezvous point number to which each vehicle-machine collaborative delivery customer belongs. Its length is equal to the total number of customers in the vehicle-machine collaborative delivery mode. The value of each element in the array represents the rendezvous point number to which the corresponding customer is assigned. Customers can only be assigned to rendezvous points that are in an active state in the rendezvous point activation coding layer.

[0112] An autonomous vehicle path coding layer is used to represent the order in which autonomous vehicles visit activated meeting points. Its length is equal to the number of meeting points in the activation coding layer that are active. The elements in the sequence represent the order in which autonomous vehicles visit activated meeting points and only include meeting points with a value of 1 in the activation coding layer.

[0113] The three coding layers together constitute a complete chromosome, sharing the same fitness function for evaluation. Furthermore, the lengths of the customer home coding layer and the autonomous vehicle path coding layer are dynamically adjusted according to the changes in the activation state in the merging point activation coding layer.

[0114] Specifically, please see Figure 3 In this embodiment, a three-layer heterogeneous coding chromosome is used to uniformly represent the activation status of the meeting point, the customer affiliation relationship, and the unmanned vehicle access path.

[0115] For ease of explanation Figure 3 The correspondence between the three layers of codes is shown in this embodiment. The six candidate rendezvous points are marked as rendezvous point 1, rendezvous point 2, rendezvous point 3, rendezvous point 4, rendezvous point 5 and rendezvous point 6 in sequence, and customer 1, customer 2, customer 3 and customer 4 are used as example marks for vehicle-machine collaborative delivery customers. Figure 3 The above numbers are only used to illustrate the correspondence between each coded element and the candidate meeting point and customer, and do not constitute a limitation on the number of candidate meeting points and the number of customers.

[0116] The first layer is the merge point activation encoding layer, which uses binary strings to indicate whether each candidate merge point is activated. A value of 1 indicates that the corresponding candidate merge point is in an active state, and a value of 0 indicates that the corresponding candidate merge point is in an inactive state.

[0117] For example, when the activation code for a merge point is [1, 0, 1, 0, 0, 1], it means that merge point 1, merge point 3, and merge point 6 are activated, while the remaining candidate merge points are not activated.

[0118] The second layer is the customer attribution coding layer, which uses an integer array to represent the rendezvous point number to which each vehicle-to-everything (V2X) delivery customer currently belongs. The length of the customer attribution coding layer is the same as the number of V2X delivery customers, and each coding element corresponds to one customer.

[0119] The third layer is the autonomous vehicle path coding layer, which uses natural numbers to represent the order in which the autonomous vehicle visits each active rendezvous point. For example, when the autonomous vehicle path coding is [1,3,6], it means that after the autonomous vehicle departs from the distribution center, it visits rendezvous point 1, rendezvous point 3, and rendezvous point 6 in sequence, and finally returns to the distribution center.

[0120] Specifically, Figure 3 The activation codes for the first-level convergence points are shown as [1,0,1,0,0,1]. Each coded bit corresponds to convergence points 1 through 6 from left to right. A code value of 1 indicates that the corresponding convergence point is active, and a code value of 0 indicates that the corresponding convergence point is inactive. Therefore, convergence points 1, 3, and 6 are active, while convergence points 2, 4, and 5 are inactive. The second-level customer affiliation codes are shown as [1,3,1,6]. Each coded element corresponds to customer 1, customer 2, customer 3, and customer 4 from left to right. The element value represents the convergence point number to which the corresponding customer belongs; that is, customer 1 belongs to convergence point 1, customer 2 belongs to convergence point 3, customer 3 belongs to convergence point 1, and customer 4 belongs to convergence point 6. The third-layer autonomous vehicle path code is [1,3,6], indicating that after departing from the distribution center, the autonomous vehicle visits rendezvous point 1 for the first time, rendezvous point 3 for the second time, and rendezvous point 6 for the third time, before returning to the distribution center. Therefore, Figure 3 The meeting point number, customer number, customer affiliation code value, and autonomous vehicle access order in the code correspond one-to-one with the corresponding decision variables in the three-layer heterogeneous coding.

[0121] The three coding layers together form a complete chromosome and share the same fitness function. Specifically, customers in the customer home coding layer can only be assigned to active merging points, and the autonomous vehicle path coding layer can only contain active merging points. The lengths of both the customer home coding layer and the autonomous vehicle path coding layer are dynamically adjusted based on the activation status of merging points in the merging point activation coding layer. When a merging point is not activated, no customer is assigned to that merging point, and it is not included in the third-layer path coding.

[0122] Genetic Algorithm Iterative Process

[0123] In some embodiments, the crossover and mutation operations adapted to the characteristics of each coding layer are specifically as follows:

[0124] The co-activation coding layer at the confluence point employs two-point crossover and bit-flip mutation. In the two-point crossover, two crossover positions are randomly selected and the binary segments between the two parent individuals are swapped. In the bit-flip mutation, a gene bit is randomly selected and its binary value is inverted.

[0125] The customer affiliation coding layer employs single-point crossover and single-gene random replacement mutations. In the single-point crossover, a crossover position is randomly selected and the integer segments following the crossover position of the two parent individuals are swapped. In the single-gene random replacement mutation, a gene position is randomly selected and replaced with the number of another activated confluence point.

[0126] The path coding layer of the unmanned vehicle employs partial matching crossover and exchange mutation. In partial matching crossover, two crossover positions are randomly selected, and the permutation segments of the two parent individuals between the two crossover positions are exchanged. Repeated elements are repaired by partial mapping. In exchange mutation, two gene loci are randomly selected and their positions are exchanged.

[0127] Specifically, please see Figure 4 After the initial population is generated, the fitness of each chromosome is calculated, and the genetic algorithm iterative process begins. Each generation sequentially performs selection, crossover, mutation, three-layer coding linkage update, constraint verification and repair, and fitness calculation.

[0128] The selection process combines roulette wheel selection with elite retention. Roulette wheel selection determines the probability of selection based on the fitness of each chromosome, while elite retention directly preserves individuals with higher fitness in the current population to the next generation, reducing the loss of superior individuals during crossover and mutation.

[0129] To address the different characteristics of the three-layer coding structure, appropriate crossover and mutation operators are employed. Specifically, the merge point activation coding layer uses two-point crossover and bit-flip mutation: two-point crossover randomly selects two crossover positions and swaps the binary segments of the two parent individuals between the two crossover positions; bit-flip mutation randomly selects a gene bit and inverts its binary value. The customer affiliation coding layer uses single-point crossover and single-gene random replacement mutation: single-point crossover randomly selects a crossover position and swaps the integer segments of the two parent individuals after the crossover position; single-gene random replacement mutation randomly selects a gene bit and replaces it with the number of another activated merge point. The autonomous vehicle path coding layer uses partial matching crossover and exchange mutation: partial matching crossover randomly selects two crossover positions, swaps the permutation segments of the two parent individuals between the two crossover positions, and repairs duplicate elements through partial mapping; exchange mutation randomly selects two gene bits and swaps their positions.

[0130] After the three layers of coding complete the corresponding crossover and mutation operations, they are recombined into a complete chromosome and evaluated using the same fitness function, thereby enabling the three types of variables—meeting point activation, customer affiliation, and autonomous vehicle path—to be collaboratively optimized in a unified evolutionary process.

[0131] Three-layer coding linkage update

[0132] Specifically, after each crossover or mutation operation is completed, it is checked whether the activation state of each candidate rendezvous point in the rendezvous point activation coding layer has changed. When the activation code of a certain candidate rendezvous point changes from 1 to 0, it indicates that the rendezvous point changes from an activated state to an inactivated state; when the activation code changes from 0 to 1, it indicates that the rendezvous point changes from an inactivated state to an activated state. If the activation state of the rendezvous point does not change, the current customer affiliation relationship and the unmanned vehicle path are maintained, and the process directly proceeds to the subsequent constraint verification process; if a change in the activation state of the rendezvous point is detected, three-layer coding linkage update is initiated.

[0133] In some embodiments, when a candidate rendezvous point changes from an activated state to an inactivated state, the process includes:

[0134] Traversing the customer affiliation coding layer to find all vehicle-machine collaborative delivery customers currently affiliated with the to-be-closed rendezvous point, so as to form a to-be-reallocated customer set;

[0135] For each customer in the to-be-reallocated customer set, calculating the distance from the customer to each of the remaining activated rendezvous points, preferentially selecting the activated rendezvous point with the shortest distance as the candidate target rendezvous point, and checking whether the drone delivery task corresponding to the candidate target rendezvous point meets the maximum load and maximum range requirements after the customer is transferred to the candidate target rendezvous point, and if not, sequentially selecting the activated rendezvous point with the next shortest distance;

[0136] After the affiliation update of all to-be-reallocated customers is completed, the rendezvous point number of the corresponding customer in the customer affiliation coding layer is updated to the new target rendezvous point number, and the element corresponding to the closed rendezvous point is deleted from the unmanned vehicle path coding layer.

[0137] Specifically, in the linkage update when a rendezvous point is closed, when the m-th candidate rendezvous point changes from an activated state to an inactivated state, the customer affiliation coding layer is first traversed to find all vehicle-machine collaborative delivery customers originally assigned to the m-th rendezvous point, and a to-be-reallocated customer set is formed. For each customer in the to-be-reallocated customer set, the distance from the customer to the remaining activated rendezvous points is calculated, and the activated rendezvous point with a relatively short distance is preferentially selected as the new target rendezvous point. In the reallocation process, it shall also be checked whether the delivery task after customer transfer meets the maximum load and maximum range requirements of the drone, so as to avoid reallocating customers to rendezvous points that have already exceeded the limit. After the customer reallocation is completed, the rendezvous point number of the corresponding customer in the customer affiliation coding layer is updated, and the closed m-th rendezvous point is deleted from the unmanned vehicle path coding layer, so that the inactivated rendezvous point is no longer included in the unmanned vehicle path.

[0138] In some embodiments, the specific processing manner when a candidate rendezvous point changes from an inactivated state to an activated state is as follows:

[0139] Calculate the distance from each vehicle-machine collaborative delivery customer to the newly activated rendezvous point. From the customers currently belonging to other activated rendezvous points, select the customer closest to the newly activated rendezvous point and update their affiliation in the customer affiliation coding layer to the newly activated rendezvous point.

[0140] The newly activated merging point is added to the set of activated merging points. Starting from the distribution center, all the updated activated merging points are combined into an unvisited set. Each time, the unvisited merging point closest to the current location is selected and added to the autonomous vehicle path, which is then used as the new current location. This process is repeated until all activated merging points have been visited. Finally, the system returns to the distribution center to obtain the updated autonomous vehicle path coding layer.

[0141] For example, when rendezvous point 2 changes from active to inactive, customers previously served by rendezvous point 2 are reassigned to other active rendezvous points that are closer and meet load and range constraints. Simultaneously, rendezvous point 2 is removed from the autonomous vehicle's path. Through this process, the impact of rendezvous point closure is synchronously transmitted to both the customer affiliation coding layer and the autonomous vehicle path coding layer, preventing situations where customers still point to closed rendezvous points or autonomous vehicles continue to access closed rendezvous points.

[0142] During the synchronized update process when a rendezvous point is activated, when the m-th candidate rendezvous point changes from an inactive state to an active state, the distance from each vehicle-machine collaborative delivery customer to the m-th rendezvous point is calculated. Then, from the customers currently assigned to other active rendezvous points, customers closest to the m-th rendezvous point are selected and reassigned to that rendezvous point. After the customer assignment adjustment is completed, the m-th rendezvous point is added to the set of active rendezvous points. Based on the updated set of active rendezvous points, the nearest neighbor algorithm is used to regenerate the autonomous vehicle's closed-loop path. Specifically, the delivery center is used as the initial current location, and all updated active rendezvous points are grouped into an unvisited set. Each time, the unvisited rendezvous point closest to the current location is selected and added to the autonomous vehicle's path, becoming the new current location. This process is repeated until all active rendezvous points have been visited, and finally, the vehicle returns to the delivery center.

[0143] For example, when the second meeting point changes from an inactive state to an active state, a vehicle-to-machine collaborative delivery customer that is closer to the second meeting point is selected and reassigned to the second meeting point. The nearest neighbor algorithm is then used to generate a closed-loop path for the autonomous vehicle that includes the second meeting point. Through this process, the newly activated meeting point can simultaneously enter both the customer affiliation and the autonomous vehicle access path, thus maintaining consistency among the three layers of coding.

[0144] After completing the adjustments to the rendezvous point activation status, customer affiliation, and autonomous vehicle path updates, the updated three-layer coding is written back to the same chromosome. The updated customer affiliation coding layer only contains the rendezvous point numbers that are currently active, and the autonomous vehicle path coding layer also only contains the rendezvous points that are currently active. Through the above synchronous updates, the rendezvous point activation status, customer affiliation, and autonomous vehicle access path can maintain their corresponding relationship.

[0145] The linkage update is only executed when the activation status of the meeting point changes; if the activation status does not change, the customer affiliation and the autonomous vehicle path are not readjusted, but the process directly proceeds to the subsequent constraint verification process.

[0146] Multi-level constraint verification and repair

[0147] In some embodiments, multi-level constraint verification and repair includes:

[0148] First, a consistency check is performed between the meeting point, customer assignment, and autonomous vehicle route: This check whether each customer in the customer assignment coding layer has been assigned to an active meeting point in the meeting point activation coding layer; whether the autonomous vehicle route coding layer contains inactive meeting points; and whether all active meeting points undertaking customer delivery tasks are included in the autonomous vehicle route coding layer. If any inconsistency exists, the customer is reassigned or the autonomous vehicle route is corrected.

[0149] After the consistency verification is passed, the drone load and range verification is performed: the total weight of goods and the total flight distance of all drone delivery tasks corresponding to each activated rendezvous point are calculated to determine whether they exceed the maximum load and maximum range limits of the drone; if the limits are exceeded, the customers who exceed the limits will be transferred to other nearby activated rendezvous points that do not exceed the limits, and the load and range verification will be performed again.

[0150] After the load and range verification are passed, customer soft time window verification is performed: check whether the actual delivery completion time of each customer is later than the preset latest delivery time; for customers whose lateness exceeds the preset tolerance range, the customer is reassigned to an activated rendezvous point that the unmanned vehicle can reach earlier, or the access order of the customer's corresponding rendezvous point in the unmanned vehicle path coding layer is adjusted to reduce the degree of time window default.

[0151] Specifically, after completing the three-layer coding linkage update, the chromosome is sequentially subjected to the following checks: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and soft time window verification.

[0152] The consistency check primarily verifies whether a customer has been assigned to an active rendezvous point, whether the autonomous vehicle's path contains inactive rendezvous points, and whether the active rendezvous point responsible for the customer's delivery is included in the autonomous vehicle's path. If a customer is assigned to an inactive rendezvous point, the customer is reassigned to a nearby active rendezvous point; if the autonomous vehicle's path contains an inactive rendezvous point, that rendezvous point is removed from the path; if the active rendezvous point responsible for the customer's delivery is not present in the autonomous vehicle's path, the autonomous vehicle's path is corrected.

[0153] The load and range verification is used to check whether the drone delivery missions corresponding to each active rendezvous point exceed the drone's maximum load and maximum range. If the load or range exceeds the limit, the customers exceeding the limit will be transferred to a nearby active rendezvous point that does not exceed the limit, and the load and range verification will be performed again.

[0154] Soft time window verification checks whether the customer's actual delivery completion time is later than the latest delivery time of the soft time window. No time window penalty cost is incurred when the customer is not late; when the customer is late, a penalty cost is calculated based on the lateness time. For customers whose lateness exceeds a preset tolerance range, they can be reassigned to an active rendezvous point that the autonomous vehicle can reach earlier, or the access order of the corresponding rendezvous point in the autonomous vehicle's path can be adjusted to reduce the degree of time window default.

[0155] The three types of checks are performed sequentially, proceeding to the next level of check only after the previous level passes. The repaired chromosome undergoes the corresponding checks again, and participates in subsequent genetic iterations after meeting the constraints.

[0156] Fitness calculation and population update

[0157] This embodiment uses the weighted sum of total delivery time, total operating cost, and soft time window penalty cost as the optimization objective. The total operating cost includes the cost of the autonomous vehicle's operation and the drone's flight. The soft time window penalty cost is calculated only for late deliveries where the actual delivery completion time is later than the latest delivery time; early deliveries incur no penalty cost. The objective function is defined as follows: in, Represents the total delivery time. This represents the total operating cost (including the cost of driving the autonomous vehicle and the cost of flying the drone). The cost of soft time window penalties incurred due to customer lateness; and These are the weighted coefficients for delivery time and cost / penalty, respectively, satisfying... .

[0158] The total operating cost and the soft time window penalty cost are calculated as follows:

[0159] ; ;

[0160] in, and Indicate whether the driverless car and the drone have traversed the path. , For nodes To the node European distance, and The unit distance operating costs for autonomous vehicles and drones are respectively ( Yuan / km (yuan / km) The default penalty coefficient per unit time window ( (yuan / minute) For customers The actual completion time For customers The latest delivery time.

[0161] When calculating fitness, the fitness of each chromosome is calculated based on the objective function value. Since the objective function seeks to minimize the value, the fitness of this algorithm... Defined as the reciprocal of the objective function value: The smaller the objective function value, the better the overall performance of the corresponding delivery solution in terms of delivery time, operating costs, and time window defaults, and the higher its chromosome fitness. After each crossover and mutation, a three-layer encoding linkage update is first performed based on the change in the activation state of the convergence point, followed by constraint verification and repair. Finally, the fitness of the repaired chromosome is calculated, and the population is updated based on the fitness results. When the genetic algorithm reaches the maximum number of iterations or meets the preset convergence condition, the iteration stops, and the chromosome with the highest fitness in the current population is taken as the optimal chromosome.

[0162] Delivery solution output

[0163] Decoding the optimal chromosome yields the final unmanned vehicle-drone collaborative delivery scheme. The delivery scheme includes the activation status of each candidate meeting point, the affiliation between the vehicle-drone collaborative delivery customer and the activated meeting point, the closed-loop path for the unmanned vehicle to access the activated meeting point, the final delivery path for each drone, and the delivery path for pure drone direct delivery to the customer.

[0164] Implementation Results and Experimental Verification

[0165] Please see Figure 5 and Figure 6After 200 generations of genetic iteration, the total flight range of the drone decreased from 1013.9km to 861.2km, a reduction of 15.1%; the total operating cost decreased from 2534.8 yuan to 2211.6 yuan, a reduction of 12.8%. The number of customers who defaulted on their time windows decreased from 23 to 7, the cumulative late time decreased from 357 minutes to 40 minutes, and the time window penalty cost decreased from 1867 yuan to 203 yuan.

[0166] like Figure 5 The diagram shows a comparison of the spatial layout before and after optimization of the meeting point in this embodiment. (a) represents the initial layout before optimization, and (b) represents the layout after optimization, where the meeting point is adjusted to the customer area. In the diagram, solid squares represent distribution centers, black dots represent customer points, hollow triangles (M1-M6) represent the initial candidate meeting points before optimization, solid rhombuses represent meeting points dynamically adjusted to the customer gathering area after optimization, and arrows indicate the direction of displacement of the meeting point. Figure 6 The diagram shows a comparison of the effects before and after the time window default repair. The horizontal axis represents the customer number, and the vertical axis represents the default time (unit: minutes).

[0167] Finally, the unmanned vehicles visited the corresponding rendezvous points in the order of M1, M3, M5, M6, M4, and M2, and returned to the distribution center after completing all visits, with a total travel distance of 486.3 km. The number of customers served at each rendezvous point was 5, 4, 4, 4, 3, and 4, respectively, with corresponding drone flight distances of 143.8 km, 189.2 km, 92.7 km, 151.3 km, 189.8 km, and 94.4 km; the total flight distance for the 16 purely drone-delivered customers was 412.5 km. The algorithm's objective function value decreased from 8500 to 5200 in the first 50 generations, and gradually stabilized after the 120th generation, with a final objective function value of 4865.2. The difference between the population mean and the optimal value was 8.3%, indicating that the algorithm reached stable convergence within 200 generations.

[0168] The above results demonstrate that the method of this invention reduces the number of customers requiring vehicle-machine collaborative optimization through hybrid fulfillment pre-partitioning, compresses the rendezvous location decision space using a finite number of discrete candidate rendezvous points, and unifies the rendezvous point activation state, customer affiliation, and autonomous vehicle access path through a three-layer heterogeneous coding system. When the rendezvous point activation state changes, the customer affiliation coding layer and the autonomous vehicle path coding layer can be updated accordingly, thereby reducing conflicts between the three coding layers. Combined with multi-level constraint verification and repair after genetic iteration, a delivery solution that meets the requirements of drone payload, range, and customer soft time windows is ultimately formed.

[0169] The method of this invention can be used in various application scenarios such as urban last-mile logistics distribution, emergency material delivery, express delivery coverage in remote areas, fresh food cold chain distribution, and emergency logistics supply.

[0170] Please see Figure 7 Based on the same inventive concept, the present invention also provides a vehicle-machine collaborative delivery system based on hybrid fulfillment and heterogeneous coding, comprising:

[0171] The hybrid fulfillment pre-segmentation module is used to acquire delivery network data, perform hybrid fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode;

[0172] The discrete meeting point management module is used to cluster the coordinates of customers in the vehicle-machine collaborative delivery mode, generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points.

[0173] The three-layer co-evolution module is used to construct a three-layer heterogeneous coding chromosome to uniformly represent the activation status of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which the unmanned vehicle visits the rendezvous point.

[0174] A genetic algorithm is used to iteratively evolve the population composed of the three heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer affiliation coding layer and the autonomous vehicle path coding layer are performed respectively.

[0175] After the crossover or mutation operation, the activation status of each candidate merging point in the merging point activation coding layer is checked to see if it has changed. When a candidate merging point changes from an active state to an inactive state, all vehicle-to-machine cooperative delivery customers currently belonging to that merging point are reassigned to other active merging points that meet the load and range constraints, and the merging point is deleted from the autonomous vehicle path coding layer. When a candidate merging point changes from an inactive state to an active state, the customer closest to that merging point is selected from the vehicle-to-machine cooperative delivery customers currently belonging to other active merging points, their affiliation is updated to the new active merging point, and the merging point is inserted into the autonomous vehicle path coding layer according to the nearest neighbor principle. Synchronous write-back is performed on the merging point activation coding layer, customer affiliation coding layer, and autonomous vehicle path coding layer to form the updated complete chromosome.

[0176] The multi-constraint verification module is used to sequentially perform the following checks on the updated chromosome: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Based on the verification results, the chromosome is repaired accordingly, and the fitness of the repaired chromosome is calculated.

[0177] The delivery solution output module is used to iterate with the optimization objective of minimizing the weighted sum of delivery time, operating costs, and time window penalties. When the preset convergence condition is met, the optimal chromosome is decoded and the final delivery solution is output.

[0178] Unless otherwise defined, the technical or scientific terms used in this invention shall have the general meaning understood by one of ordinary skill in the art to which this invention pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this invention do not indicate quantitative limitations and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this invention are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “A plurality” used in this invention refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. In this invention, the terms "first," "second," "third," etc., are used only to distinguish similar objects and do not represent a specific order of objects.

[0179] Obviously, the accompanying drawings are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, certain design, manufacturing, or production modifications made according to the technical content disclosed in the present invention are merely conventional technical means and should not be considered as insufficient disclosure of the present invention.

[0180] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-described technical content to create equivalent embodiments without departing from the scope of the present invention. The implementation schemes in the above embodiments can also be further combined or replaced. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A vehicle-machine collaborative delivery method based on hybrid fulfillment and heterogeneous coding, characterized in that, Includes the following steps: S1. Obtain delivery network data, perform mixed fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode. S2. Cluster the coordinates of customers in the vehicle-machine collaborative delivery mode to generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points. S3. Construct a three-layer heterogeneous coding chromosome to uniformly represent the activation state of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which unmanned vehicles visit the rendezvous point. S4. A genetic algorithm is used to iteratively evolve the population composed of the three-layer heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer affiliation coding layer and the unmanned vehicle path coding layer are performed respectively. S5. After the crossover or mutation operation, check whether the activation state of each candidate merging point in the merging point activation coding layer has changed. When a candidate meeting point changes from an active state to an inactive state, all vehicle-to-machine collaborative delivery customers currently belonging to that meeting point are reassigned to other active meeting points that meet the load and range constraints, and the meeting point is removed from the autonomous vehicle path coding layer. When a candidate meeting point changes from an inactive state to an active state, the customer closest to that meeting point is selected from the vehicle-to-machine collaborative delivery customers currently belonging to other active meeting points, and their affiliation is updated to the newly active meeting point. The meeting point is then inserted into the autonomous vehicle path coding layer according to the nearest neighbor principle. A synchronous write-back is performed on the meeting point activation coding layer, the customer affiliation coding layer, and the autonomous vehicle path coding layer to form an updated complete chromosome. S6. Perform the following checks on the updated chromosome in sequence: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Repair the chromosome accordingly based on the verification results and calculate the fitness of the repaired chromosome. S7. Iterate with the weighted sum of delivery time, operating costs and time window penalties as the optimization objective. When the preset convergence condition is met, decode the optimal chromosome and output the final delivery plan.

2. The method according to claim 1, characterized in that, The delivery network data includes the location of the delivery center, the location and weight of the goods, the customer's soft time window, the location of the candidate meeting point, the speed and unit operating cost of unmanned vehicles and drones, the maximum payload and maximum range of drones, and the battery capacity and safety reserve power of unmanned vehicles.

3. The method according to claim 1, characterized in that, Hybrid performance pre-partitioning includes: Calculate the Euclidean distance from each customer to the distribution center; when the Euclidean distance from the customer to the distribution center is less than a preset distance threshold and the customer's required weight is less than a preset weight threshold, mark the customer as a pure drone direct delivery mode; otherwise, mark it as a vehicle-machine collaborative delivery mode. For customers marked as using the pure drone delivery mode, the drone takes off directly from the distribution center to complete the delivery and return, without occupying the resources of the unmanned vehicle and the rendezvous point; for customers marked as using the vehicle-machine collaborative delivery mode, the unmanned vehicle transports the delivery goods to the corresponding rendezvous point, and then the drone takes off from the rendezvous point to complete the last delivery. The pure drone delivery mode and the vehicle-machine collaborative delivery mode coexist throughout the optimization process, and the mode label of each customer remains unchanged after the pre-division is completed.

4. The method according to claim 1, characterized in that, The three-layered heterogeneous coding chromosome specifically includes: A merge point activation encoding layer is used to represent the activation state of each candidate merge point using binary encoding. Its length is equal to the total number of candidate merge points. Each bit in the string corresponds to a candidate merge point. A value of 1 indicates that the candidate merge point is in an active state, and a value of 0 indicates that the candidate merge point is in an inactive state. The customer affiliation coding layer uses integer encoding to represent the rendezvous point number to which each vehicle-machine collaborative delivery customer belongs. Its length is equal to the total number of customers in the vehicle-machine collaborative delivery mode. The value of each element in the array represents the rendezvous point number to which the corresponding customer is assigned. Customers can only be assigned to rendezvous points that are in an active state in the rendezvous point activation coding layer. An autonomous vehicle path coding layer is used to represent the order in which autonomous vehicles visit activated meeting points. Its length is equal to the number of meeting points in the activation coding layer that are active. The elements in the sequence represent the order in which autonomous vehicles visit activated meeting points and only include meeting points with a value of 1 in the activation coding layer. The three coding layers together form a complete chromosome, share the same fitness function for evaluation, and the lengths of the customer home coding layer and the autonomous vehicle path coding layer are dynamically adjusted according to the changes in the activation state in the merging point activation coding layer.

5. The method according to claim 1, characterized in that, Generate discrete candidate meeting points and initial attribution relationships, including: K-means clustering is performed on the two-dimensional spatial coordinates of all customers in the vehicle-machine collaborative delivery mode. Each cluster center is used as the initial discrete candidate merging point, and customers belonging to the same cluster are initially assigned to the corresponding candidate merging point. Starting from the distribution center, all activated meeting points are grouped into an unvisited set. Each time, the meeting point closest to the current location is selected from the unvisited set, added to the unmanned vehicle path, and the meeting point is used as the new current location. This process is repeated until all activated meeting points are visited. Finally, the vehicle returns to the distribution center to obtain the initial closed-loop path of the unmanned vehicle. By repeatedly performing customer clustering, initial affiliation assignment, and nearest neighbor path generation operations, a preset number of initial individuals are generated, and these initial individuals form the initial population of the genetic algorithm.

6. The method according to claim 1, characterized in that, When a candidate merging point changes from an active state to an inactive state, it includes: Traverse the customer affiliation coding layer to find all vehicle-machine collaborative delivery customers currently belonging to the closure rendezvous point and form a set of customers to be reassigned. For each customer in the set of customers to be reassigned, calculate the distance from the customer to each of the other activated rendezvous points, prioritize the nearest activated rendezvous point as the candidate target rendezvous point, and check whether the drone delivery task corresponding to the candidate target rendezvous point meets the maximum load and maximum range requirements after the customer is transferred to the candidate target rendezvous point. If not, select the next closest activated rendezvous point in sequence. After completing the attribution update for all customers to be reassigned, update the rendezvous point number of the corresponding customer in the customer attribution coding layer to the new target rendezvous point number, and delete the elements corresponding to the closed rendezvous points from the autonomous vehicle path coding layer.

7. The method according to claim 1, characterized in that, The specific handling method when a candidate merging point changes from an inactive state to an active state is as follows: Calculate the distance from each vehicle-machine collaborative delivery customer to the newly activated rendezvous point. From the customers currently belonging to other activated rendezvous points, select the customer closest to the newly activated rendezvous point and update their affiliation in the customer affiliation coding layer to the newly activated rendezvous point. The newly activated merging point is added to the set of activated merging points. Starting from the distribution center, all the updated activated merging points are combined into an unvisited set. Each time, the unvisited merging point closest to the current location is selected and added to the autonomous vehicle path, which is then used as the new current location. This process is repeated until all activated merging points have been visited. Finally, the system returns to the distribution center to obtain the updated autonomous vehicle path coding layer.

8. The method according to claim 3, characterized in that, The crossover and mutation operations adapted to the characteristics of each coding layer are as follows: The co-activation coding layer at the confluence point employs two-point crossover and bit-flip mutation. In the two-point crossover, two crossover positions are randomly selected and the binary segments between the two parent individuals are swapped. In the bit-flip mutation, a gene bit is randomly selected and its binary value is inverted. The customer affiliation coding layer employs single-point crossover and single-gene random replacement mutations. In the single-point crossover, a crossover position is randomly selected and the integer segments following the crossover position of the two parent individuals are swapped. In the single-gene random replacement mutation, a gene position is randomly selected and replaced with the number of another activated confluence point. The path coding layer of the unmanned vehicle employs partial matching crossover and exchange mutation. In partial matching crossover, two crossover positions are randomly selected, and the permutation segments of the two parent individuals between the two crossover positions are exchanged. Repeated elements are repaired by partial mapping. In exchange mutation, two gene positions are randomly selected and their positions are exchanged.

9. The method according to claim 1, characterized in that, Multi-level constraint verification and repair includes: First, a consistency check is performed between the meeting point, customer assignment, and autonomous vehicle route: This check whether each customer in the customer assignment coding layer has been assigned to an active meeting point in the meeting point activation coding layer; whether the autonomous vehicle route coding layer contains inactive meeting points; and whether all active meeting points undertaking customer delivery tasks are included in the autonomous vehicle route coding layer. If any inconsistency exists, the customer is reassigned or the autonomous vehicle route is corrected. After the consistency verification is passed, the drone load and range verification is performed: the total weight of goods and the total flight distance of all drone delivery tasks corresponding to each activated rendezvous point are calculated to determine whether they exceed the maximum load and maximum range limits of the drone; if the limits are exceeded, the customers who exceed the limits will be transferred to other nearby activated rendezvous points that do not exceed the limits, and the load and range verification will be performed again. After the load and range verification are passed, customer soft time window verification is performed: check whether the actual delivery completion time of each customer is later than the preset latest delivery time; for customers whose lateness exceeds the preset tolerance range, the customer is reassigned to an activated rendezvous point that the unmanned vehicle can reach earlier, or the access order of the customer's corresponding rendezvous point in the unmanned vehicle path coding layer is adjusted to reduce the degree of time window default.

10. A vehicle-machine collaborative delivery system based on hybrid fulfillment and heterogeneous coding, characterized in that, include: The hybrid fulfillment pre-segmentation module is used to acquire delivery network data, perform hybrid fulfillment pre-segmentation for all customers, and mark customers as either pure drone direct delivery mode or vehicle-machine collaborative delivery mode; The discrete meeting point management module is used to cluster the coordinates of customers in the vehicle-machine collaborative delivery mode, generate a finite number of discrete candidate meeting points, and establish the initial affiliation relationship between each vehicle-machine collaborative delivery customer and the candidate meeting points. The three-layer co-evolution module is used to construct a three-layer heterogeneous coding chromosome to uniformly represent the activation state of the rendezvous point, the drone assignment relationship from the customer to the rendezvous point, and the order in which the unmanned vehicle visits the rendezvous point. A genetic algorithm is used to iteratively evolve the population composed of the three heterogeneous coding chromosomes. During the iteration process, crossover and mutation operations adapted to the coding characteristics of the convergence point activation coding layer, the customer affiliation coding layer and the autonomous vehicle path coding layer are performed respectively. After a crossover or mutation operation, check whether the activation state of each candidate merging point in the merging point activation coding layer has changed. When a candidate meeting point changes from an active state to an inactive state, all vehicle-to-machine collaborative delivery customers currently belonging to that meeting point are reassigned to other active meeting points that meet the load and range constraints, and the meeting point is removed from the autonomous vehicle path coding layer. When a candidate meeting point changes from an inactive state to an active state, the customer closest to that meeting point is selected from the vehicle-to-machine collaborative delivery customers currently belonging to other active meeting points, and their affiliation is updated to the newly active meeting point. The meeting point is then inserted into the autonomous vehicle path coding layer according to the nearest neighbor principle. A synchronous write-back is performed on the meeting point activation coding layer, the customer affiliation coding layer, and the autonomous vehicle path coding layer to form an updated complete chromosome. The multi-constraint verification module is used to sequentially perform the following checks on the updated chromosome: convergence point-customer affiliation-unmanned vehicle path consistency verification, UAV payload and range verification, and customer soft time window verification. Based on the verification results, the chromosome is repaired accordingly, and the fitness of the repaired chromosome is calculated. The delivery solution output module iterates with the optimization objective of minimizing the weighted sum of delivery time, operating costs, and time window penalties. When the preset convergence condition is met, it decodes the optimal chromosome and outputs the final delivery solution.

Images