Vehicle-heterogeneous unmanned aerial vehicle cooperative distribution method and device

Abstract

The invention discloses a vehicle-heterogeneous unmanned aerial vehicle cooperative distribution method and device, and is applied to the technical field of logistics distribution, and the method comprises the steps: building a path planning model based on a distribution scene; wherein the distribution scene comprises a vehicle, a plurality of unmanned aerial vehicles and a plurality of nodes, the nodes comprise any one of a delivery node, a sending node or a sending and delivery node, the unmanned aerial vehicles comprise any one of a first unmanned aerial vehicle and a second unmanned aerial vehicle, and the second unmanned aerial vehicle is only used for accessing the delivery node. The path planning model aims at minimizing the distribution cost and constrains the distribution behaviors of the vehicle and the unmanned aerial vehicle; performing optimization processing on the path planning model by using an adaptive large neighborhood search algorithm to obtain an optimal distribution scheme; and controlling the vehicle and the plurality of unmanned aerial vehicles to execute corresponding distribution behaviors according to the optimal distribution scheme. According to the embodiment of the invention, by balancing the requirements of delivery, sending and simultaneous sending and delivery, the vehicle-mounted terminal cooperative distribution efficiency can be effectively improved, and the vehicle-mounted terminal cooperative distribution cost is reduced.

Description

Technical Field

[0001] The present application relates to the field of logistics and delivery technology, and in particular to a vehicle-heterogeneous drone collaborative delivery method and device. Background Art

[0002] In logistics and delivery, the Multi-Flying Sidekick Traveling Salesman Problem (mFSTSP) involves the coordinated delivery of multiple drones and a single vehicle. In this problem, the vehicle transports the drones and a portion of the cargo along a primary route. Each drone takes off from the vehicle, provides service to a customer point, and then returns to the vehicle. Related technologies typically use a single type of drone for vehicle-machine collaborative delivery. However, a single type of drone is often used for either delivery or shipping services. This makes these technologies unable to simultaneously handle large volumes of both shipping and delivery requests, resulting in low logistics and delivery efficiency. Summary of the Invention

[0003] The embodiments of the present application provide a vehicle-heterogeneous drone collaborative delivery method and device, which can improve the efficiency of vehicle-machine collaborative delivery and reduce the cost of vehicle-machine collaborative delivery by balancing the needs of delivery, mailing, and simultaneous mailing.

[0004] On the one hand, an embodiment of the present application provides a vehicle-heterogeneous drone collaborative delivery method, comprising the following steps: Constructing a path planning model based on a delivery scenario; wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, wherein the nodes include any one of a delivery node, a consignment node, or a consignment-delivery node; the drone includes any one of a first drone and a second drone, wherein the second drone is used only to access the delivery node; and the path planning model aims to minimize delivery costs and constrains the delivery behaviors of the vehicle and the drone. The path planning model is optimized using an adaptive large neighborhood search algorithm to obtain the best delivery solution; According to the optimal delivery plan, the vehicle and the plurality of drones are controlled to perform corresponding delivery actions.

[0005] On the other hand, an embodiment of the present application provides a vehicle-heterogeneous drone collaborative delivery device, comprising: A first processing module is configured to construct a path planning model based on a delivery scenario; wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, wherein the nodes include any one of a delivery node, a consignment node, or a consignment-delivery node; the drone includes any one of a first drone and a second drone, wherein the second drone is used only to access the delivery node; and the path planning model aims to minimize delivery costs and constrains the delivery behaviors of the vehicle and the drone. The second processing module is used to optimize the path planning model using an adaptive large neighborhood search algorithm to obtain an optimal delivery solution; The third processing module is used to control the vehicle and the plurality of drones to perform corresponding delivery actions according to the optimal delivery plan.

[0006] According to a vehicle-heterogeneous drone collaborative delivery method and device according to an embodiment of the present application, first, a path planning model is constructed based on the delivery scenario; wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, the node includes any one of a delivery node, a shipping node, or a shipping node, the drone includes any one of a first drone or a second drone, and the second drone is only used to access the delivery node, and the path planning model aims to minimize the delivery cost and constrain the delivery behavior of the vehicle and the drone; then, an adaptive large neighborhood search algorithm is used to optimize the path planning model to obtain the best delivery plan; finally, according to the best delivery plan, the vehicle and multiple drones are controlled to perform corresponding delivery behaviors. The embodiment of the present application can effectively improve the efficiency of vehicle-machine collaborative delivery and reduce the cost of vehicle-machine collaborative delivery by balancing the needs of delivery, shipping, and simultaneous shipping. Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be achieved and obtained through the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flow chart of the vehicle-heterogeneous drone collaborative delivery method provided by this application; Figure 2 This is a flow chart for generating the best delivery plan provided by this application; Figure 3 is an example diagram of the destruction operator provided by this application; Figure 4 This is a flowchart of adjusting the adaptive weight of the destruction operator provided by the present application; Figure 5 This is a structural diagram of the vehicle-heterogeneous drone collaborative delivery device provided in this application. DETAILED DESCRIPTION

[0008] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0009] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be considered as limiting the present application. All other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0010] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0011] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0012] The multi-partner traveling salesman problem involves the coordinated delivery of goods by multiple drones and a vehicle. In this problem, the vehicle transports the drones and some of the goods along a main route. Each drone takes off from the vehicle, delivers service to a customer point, and then returns to the vehicle.

[0013] In related technologies, drones equipped with either a single storage box or multiple storage boxes are typically used for vehicle-machine coordinated delivery. However, a single type of drone is often used for either delivery or shipping. For example, drones equipped with a single storage box are often used for both delivery and shipping, but are generally not considered for delivery-only scenarios due to their low delivery efficiency. Drones equipped with multiple storage boxes are often used for delivery due to their high delivery efficiency, but are difficult to adapt to shipping services due to concerns about cargo safety. This makes it difficult for related technologies to simultaneously address the large demand for shipping and delivery, resulting in low efficiency for vehicle-machine coordinated delivery. While some related technologies address the large demand for shipping and delivery by adding a large number of drones of the same type and controlling them separately, this undoubtedly increases the cost of vehicle-machine coordinated delivery. Furthermore, these technologies often struggle to coordinate the different types of drones with shipping and delivery needs, significantly increasing the difficulty of vehicle-machine coordinated delivery. Therefore, related technologies rarely offer vehicle-machine coordinated delivery methods that can simultaneously address the large demand for shipping and delivery.

[0014] In view of this, the embodiments of the present application provide a vehicle-heterogeneous drone collaborative delivery method and device. In response to the needs of delivery, mailing and simultaneous delivery, two different types of drones are introduced, and a path planning model for collaborative delivery of vehicles and heterogeneous drones is established with the goal of minimizing delivery costs. At the same time, an improved adaptive large-area algorithm is used to solve the above model, thereby improving the accuracy and efficiency of vehicle-machine collaborative delivery and reducing the cost of vehicle-machine collaborative delivery.

[0015] It should be noted that the vehicle can be a private car, such as a sedan, a sport utility vehicle (SUV), a multi-purpose vehicle (MPV) or a pickup truck, etc., or an operating vehicle, such as a van, a bus, a small truck or a large trailer, etc. It can also be a gasoline vehicle or a new energy vehicle such as a hybrid or pure electric vehicle.

[0016] The following will describe in detail the implementation steps of a vehicle-heterogeneous drone collaborative delivery method provided by this application with reference to the accompanying drawings.

[0017] The vehicle-heterogeneous drone collaborative delivery method provided in the embodiments of this application can be applied to a terminal or a server, or can be software running on a terminal or server. The terminal can be, but is not limited to, a tablet computer, laptop computer, or desktop computer. The server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Furthermore, the server can be, but is not limited to, a node server in a blockchain network. Blockchain is a novel application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0018] Reference Figure 1 A vehicle-heterogeneous drone collaborative delivery method provided in an embodiment of the present application may include the following steps S101-S103: S101: Build a path planning model based on the delivery scenario.

[0019] It should be noted that the delivery scenario caters to three needs: shipping, delivering, and both shipping and delivering. When assigning nodes to drones, they must be allocated based on their characteristics, and the order in which drones access nodes must meet priority constraints. Therefore, the delivery scenario includes vehicles, multiple drones, and multiple nodes. Among them: The nodes include any one of the delivery nodes, the shipping nodes or the shipping and shipping nodes, that is, the distribution services of the nodes are divided into three categories: delivery service, shipping service and simultaneous shipping and shipping service.

[0020] The drone includes either a first drone or a second drone. The first drone is used to access all nodes, while the second drone is used only to access delivery nodes. Specifically, the first drone is equipped with a large storage box and can access one or two nodes of any type during a single flight. Because the first drone's storage box has a limited capacity, it prioritizes the delivery node over the consignment node when accessing two nodes. The second drone is equipped with multiple smaller storage boxes and can access multiple nodes during a single flight, but the nodes must be delivery nodes.

[0021] The vehicle is used to visit all nodes and is equipped with multiple drones. Any cargo that exceeds the drone's load capacity is delivered by the vehicle.

[0022] It is worth noting that the path planning model aims to minimize the delivery cost and constrain the delivery behavior of vehicles and drones.

[0023] In this step, three different types of nodes are considered: delivery nodes, consignment nodes, and consignment-delivery nodes. Two different types of drones are also considered: a first drone designed to serve nodes of any type, and a second drone designed only to serve delivery nodes. This defines a vehicle-heterogeneous drone-multiple-demand-node delivery scenario. Next, based on the characteristics of the nodes and drones, a new problem model is designed to form a path planning model that minimizes delivery costs while constraining the delivery behaviors of the vehicle and drone. This objective effectively reduces the cost of vehicle-machine collaborative delivery, while constraining the delivery behaviors of the vehicle and drone allows the vehicle and the first drone to provide all types of services, while the second drone provides delivery services, thereby coordinating the roles of the vehicle and drone in vehicle-machine collaborative delivery.

[0024] S102, using an adaptive large neighborhood search algorithm to optimize the path planning model and obtain the best delivery solution.

[0025] It should be noted that the optimal delivery plan includes the optimal delivery path of the vehicle and the optimal delivery paths of multiple drones. Each optimal delivery path includes at least one node, and all optimal delivery paths cover all nodes.

[0026] In this step, the adaptive large neighborhood search algorithm is used as an effective tool to solve complex path planning models. It can solve the optimal delivery plan with the goal of minimizing delivery costs while dealing with the constraints on the delivery behavior of vehicles and drones, thereby helping vehicles and multiple drones plan reasonable delivery paths and ensuring the accuracy of path planning.

[0027] S103, according to the optimal delivery plan, control the vehicle and multiple drones to perform corresponding delivery behaviors.

[0028] In this step, based on the optimal distribution plan, the vehicle and multiple drones are controlled so that the vehicle and multiple drones provide mailing service, delivery service or mailing and delivery service for the corresponding nodes respectively.

[0029] In actual logistics distribution, the nodes to be served are often scattered, and the types of services required by each node are also different. To address this, the present embodiment adds two different types of drones to the vehicle's delivery tasks when addressing the needs of delivering, shipping, and shipping simultaneously. A path planning model is constructed with the goal of minimizing delivery costs, while constraining the delivery behaviors of the vehicle and drones. An adaptive large neighborhood search algorithm is then used to solve the path planning model, resulting in the optimal delivery solution. This solution is then used to control the vehicle and multiple drones to perform the corresponding delivery behaviors.

[0030] It can be seen that the embodiment of the present application introduces two different types of drones for collaborative cooperation, allowing two different types of drones to deliver different types of goods at a time, meeting the actual logistics distribution needs for delivery, mailing and simultaneous delivery, and using the adaptive large neighborhood search algorithm as an effective tool for solving complex path planning models. It can solve the optimal delivery plan with the purpose of minimizing delivery costs while processing the constraints on the delivery behavior of vehicles and drones, thereby helping vehicles and drones to plan reasonable delivery routes. In this way, the embodiment of the present application can balance the needs of delivery, mailing and simultaneous delivery, meet diverse delivery needs, make the drone's delivery path more varied, and at the same time reduce the vehicle's driving distance, thereby effectively improving the efficiency of vehicle-machine collaborative delivery and reducing the cost of vehicle-machine collaborative delivery.

[0031] The above steps will be further described below.

[0032] In some implementations, building a route planning model based on a delivery scenario may include: Based on the attribute information associated with the delivery behavior of vehicles and drones, a fitness function is constructed; wherein the fitness function is used to represent the minimization of delivery costs; Determine the set of constraints for the fitness function; where the set of constraints is used to constrain the delivery behavior of vehicles and drones; Construct a path planning model based on the fitness function and a set of constraints.

[0033] In this implementation, first, attribute information of the delivery scenario is acquired. This attribute information is associated with the delivery behavior of vehicles and drones, which can be defined as shipping, delivering, or both. Then, a fitness function is constructed based on this attribute information. This objective function aims to minimize delivery costs, helping to optimize multi-dimensional metrics for vehicles and drones, such as driving distance, delivery time, and energy consumption. Next, a set of constraints for the fitness function is determined. These constraints are used to constrain the delivery behavior of vehicles and drones, helping to limit unreasonable delivery behavior and improve resource utilization. Finally, a path planning model is formed using the fitness function and its constraint set. In this way, while fully considering the information associated with the delivery behavior of vehicles and drones, a fitness function with the goal of minimizing delivery costs is constructed, and at the same time a set of constraints is introduced to constrain the shipping behavior, delivery behavior and delivery behavior of vehicles and drones when the goal is to minimize delivery costs, thereby forming a path planning model. In this way, the path planning model can be oriented to the diverse needs of delivery, shipping and simultaneous delivery, making it fit the delivery scenario of the embodiment of this application, improving the accuracy of the path planning model, and thus helping to achieve efficient and high-precision vehicle-machine collaborative delivery work.

[0034] In some embodiments, the attribute information may include a node edge set, a drone flight set, a vehicle's unit distance cost, a drone's unit distance cost, a vehicle's driving attribute information, and a drone's driving attribute information. The fitness function constructed based on the attribute information associated with the delivery behavior of the vehicle and drone may include: According to the unit distance cost and driving attribute information of the vehicle, combined with the node edge set, a first cost function is constructed; Based on the UAV’s flight set, unit distance cost and driving attribute information, combined with the node edge set, a second cost function is constructed; A fitness function is constructed based on the first cost function and the second cost function.

[0035] In this implementation, delivery entities are divided into two categories: vehicles and drones. Drones are divided into two categories: first drones and second drones. Based on this, attribute information includes node edge sets, drone flight sets, unit distance costs for vehicles, unit distance costs for drones, vehicle driving attribute information, and drone driving attribute information. Among them: The node edge set is used to store multiple edges used to connect two nodes; the flight set of the drone includes the flight set of the first drone and the flight set of the second drone, and the flight set is used to store the number of flights.

[0036] The unit distance cost of the UAV includes the unit distance cost of the first UAV and the unit distance cost of the second UAV. The unit distance cost of the vehicle is measured by Manhattan distance, and the unit distance cost of the UAV is measured by Euclidean distance. These unit distance costs can be pre-calibrated according to actual conditions.

[0037] A vehicle's driving attribute information includes the vehicle's driving distance relative to any two nodes and a driving binary variable. The driving distance represents the distance the vehicle travels between two adjacent nodes, measured by the Harmanton distance. The driving binary variable indicates whether the vehicle travels from one node to another. If the driving binary variable indicates that the vehicle travels from one node to another, it is 1; otherwise, it is 0.

[0038] Drone driving attribute information includes the drone's flight distance relative to any two nodes, as well as a flight binary variable relative to any flight and any two nodes. The flight distance represents the distance the drone flew between two adjacent nodes, measured using the Euclidean distance. The flight binary variable indicates whether the drone flew from one node to another during any flight. If the flight binary variable indicates that the drone flew from one node to another during any flight, it is 1; otherwise, it is 0.

[0039] Based on this, the fitness function in the path planning model can be divided into two parts: the vehicle cost function and the drone cost function. The vehicle cost function is recorded as the first cost function, and the drone cost function is recorded as the second cost function. Since drones are divided into two categories, the second cost function can be further divided into the first drone cost function and the second drone cost function. The construction of each cost function is as follows: A first cost function is constructed based on the node edge set, the unit distance cost of the vehicle, and the vehicle's driving attribute information. Specifically, the unit distance cost of the vehicle and the driving attribute information are multiplied and summed based on the node edge set to construct the first cost function, which satisfies the following formula (1-1): (1-1); In formula (1-1), represents the first cost function; Represents the node-edge set, which represents the set of edges between any two nodes; Indicates the nodes and The edges of the nodes; represents the unit distance cost of the vehicle; Indicates the vehicle relative to the nodes and The driving distance of the vehicle from the node Drive to the node The driving distance of each node; Indicates the vehicle relative to the nodes and The driving binary variable of the node Drive to the node nodes, then ,on the contrary .

[0040] A second cost function is constructed based on the node edge set, the flight set of the first UAV, the unit distance cost of the first UAV, and the driving attribute information of the first UAV. Specifically, the unit distance cost and the driving attribute information of the first UAV are multiplied and summed based on the node edge set and the flight set of the first UAV to construct a cost function for the first UAV, which satisfies the following formula (1-2): (1-2); In formula (1-2), represents the cost function of the first UAV; represents the flight set of the first UAV; represents the unit distance cost of the first UAV; Indicates the first drone relative to the nodes and The flight distance of the first UAV from the first node Nodes fly to The flight distance of each node; Indicates the first drone relative to the Flight No. nodes and The flight binary variable of the node, if the first UAV is in the From the first flight Nodes fly to nodes, then ,on the contrary .

[0041] A second cost function is constructed based on the node edge set, the flight set of the second UAV, the unit distance cost of the second UAV, and the driving attribute information of the second UAV. Specifically, the unit distance cost and the driving attribute information of the second UAV are multiplied and summed based on the node edge set and the flight set of the second UAV to construct a cost function for the second UAV, which satisfies the following formula (1-3): (1-3); In formula (1-3), represents the cost function of the second UAV; represents the flight set of the second UAV; represents the unit distance cost of the second UAV; Indicates the second drone relative to the first nodes and The flight distance of the second UAV from the first node Nodes fly to The flight distance of each node; Indicates the second drone relative to the first Flight No. nodes and The flight binary variable of the node, if the second UAV is in the From the first flight Nodes fly to nodes, then ,on the contrary .

[0042] Finally, based on the first cost function, the cost function of the first UAV, and the cost function of the second UAV, the fitness function shown in the following formula (1-4) is formed: (1-4); In formula (1-4), represents the fitness function.

[0043] It can be seen that this embodiment fully considers the correlation between factors such as the unit distance cost of the vehicle, whether the vehicle is traveling between any two nodes, and the distance traveled by the vehicle between any two nodes, and the cost of the vehicle in vehicle-machine collaborative delivery. Specifically, under a fixed unit distance cost of the vehicle, if the vehicle is traveling between two nodes, the greater the distance traveled by the vehicle between the two nodes, the higher the cost of the vehicle in vehicle-machine collaborative delivery. By constructing the cost function of the vehicle in vehicle-machine collaborative delivery through the above factors, the actual delivery cost of the vehicle in vehicle-machine collaborative delivery can be accurately measured, thereby improving the accuracy of the vehicle's cost function. Furthermore, this implementation fully considers the correlation between the drone's unit distance cost, whether the drone flies between any two nodes in any given flight, and the flight distance between any two nodes, and the cost of the drone in vehicle-machine collaborative delivery. Specifically, given a fixed unit distance cost, if a drone flies between two nodes in a given flight, the greater the flight distance between the two nodes, the higher the drone's cost in vehicle-machine collaborative delivery. By constructing a cost function for the drone in vehicle-machine collaborative delivery based on these factors, we can accurately measure the actual cost of the drone in vehicle-machine collaborative delivery and improve the accuracy of the drone's cost function. This implementation can effectively improve the accuracy of the path planning model.

[0044] In some embodiments, the above-mentioned constraint set may include: node delivery constraints, overall drone delivery constraints, special delivery constraints of the first drone, special delivery constraints of the second drone, and vehicle delivery constraints.

[0045] In this embodiment, the constraint set includes several constraints, which can be divided into five categories: node delivery constraints, overall drone delivery constraints, special delivery constraints for the first drone, special delivery constraints for the second drone, and vehicle delivery constraints. Node delivery constraints are used to constrain the number of times a node is served, the number of drones that can access a node, and the number of drones that can take off and land at a node. Overall drone delivery constraints are used to constrain the drone's takeoff and landing nodes, and the number of flights. Special delivery constraints for the first drone are used to constrain the first drone's service to the node, as well as the first drone's delivery path, landing conditions, and flight range. Special delivery constraints for the second drone are used to constrain the second drone's service to the node, as well as the second drone's delivery path, payload, and flight range. Vehicle delivery constraints are used to constrain the vehicle's access to a node, the number of drones it carries, and the vehicle's round-trip travel.

[0046] The node's delivery constraints specifically include the following constraints 1 to 8: Constraint 1: Each node is served only once by a drone or vehicle, ensuring that each node is served only once. , No. The sum of the number of times a node is served by a vehicle and the number of times it is served by a drone is one, as shown in the following formula (2-1): (2-1); In formula (2-1), Represents a binary variable of the drone’s visit to any flight and any node. If the drone is in the Visit the first flight nodes, then ,on the contrary ; Represents a binary variable indicating the vehicle's access to any node. nodes, then ,on the contrary ; Represents a collection of nodes; represents the first drone set; Represents the second drone set.

[0047] Constraint 2: Each delivery node and each shipping node is served only once by the first drone or vehicle. This ensures that nodes requiring shipping services and nodes requiring both shipping and shipping services are served only once by the first drone or vehicle, as the service types of these nodes can only be met by these two types of delivery entities, as shown in the following formula (2-2): (2-2); In formula (2-2), Definition and Similarly, Indicates that the first drone is in Visit the first flight nodes; Represents the collection of consignment nodes; Represents a collection of shipping nodes.

[0048] Constraint 3: Each node served by a drone is visited by the drone during its flight, as shown in the following formula (2-3): (2-3); In formula (2-3), the set Contains a collection of nodes , starting warehouse node And the returned warehouse node ; is the number of nodes; Indicates that in the collection Delete the returned warehouse node The remaining node set; Definition and Similarly, When the drone is Visit the first flight nodes.

[0049] Constraint 4: Each node visited by a vehicle is served only by the vehicle, that is, each node visited by a vehicle is directly served by the vehicle, as shown in the following formula (2-4): (2-4).

[0050] Constraint 5: Each node can be visited by a drone at most once, meaning each node can be visited by at most one drone trip. Each drone's flight path is a non-closed loop, with a takeoff node (the node used to launch the drone) and a landing node (the node provided for the drone to land), representing the start and end of the drone's flight, respectively, as shown in the following formulas (2-5)-(2-6): (2-5); (2-6); In formula (2-6), Definition and Similarly, Indicates that the drone is From the first flight Nodes fly to nodes.

[0051] Constraint 6: The number of drones that land at the target node is equal to the number of drones launched from the target node. The target node is any node. This balances the behavior of drones accessing nodes and the flow of drones, as shown in the following formula (2-7): (2-7); In formula (2-7), Represents the landing binary variable of the UAV relative to any flight and any node. The node is the drone in the The landing node in the flight ,on the contrary ; Represents the takeoff binary variable of the UAV relative to any flight and any node. The node is the drone in the The take-off node in the next flight, ,on the contrary .

[0052] Constraint 7: The number of drones taking off from the target node does not exceed the number of drones at the target node, as shown in the following formula (2-8): (2-8); In formula (2-8), Indicates that the vehicle is leaving The number of drones carried by the vehicle at each node; Indicates that the vehicle is leaving The number of drones carried by the vehicle at each node; and Definition and Similarly, Indicates that the drone is From the first flight Nodes fly to nodes, Indicates that the drone is From the first flight Nodes fly to nodes; Represents a sufficiently large positive integer, which can be preset according to actual conditions.

[0053] Constraint 8: This is a special case of constraint 7 and determines the returned warehouse node. The number of drones is as shown in the following formula (2-9): (2-9); In formula (2-9), Indicates the total number of drones; Definition and Similarly, Indicates that the drone is From the first flight Nodes fly to nodes (i.e. the returned warehouse nodes), Definition and The definition of is similar, Indicates that the vehicle is Drive to the node nodes (i.e., the returned warehouse nodes).

[0054] It can be seen that this embodiment constrains and limits the nodes in terms of the number of times the nodes serve, the access conditions of the delivery entities to the nodes, and the number of take-off and landing drones at the nodes. In this way, the constraints on the nodes in the path planning model are realized while fully considering the characteristics of the nodes. It can accurately constrain the delivery services corresponding to the nodes in the vehicle-machine collaborative delivery, thereby helping to improve the accuracy of the path planning model.

[0055] The overall delivery constraints of drones specifically include the following constraints 9 to 11: Constraint 9: Each drone must fly no more than once between any two nodes, as shown in the following formula (2-10): (2-10).

[0056] Constraint 10: Each drone's delivery path has only one takeoff node and one landing node, ensuring that each drone's journey has only one takeoff node and one landing node, as shown in the following formula (2-11): (2-11); In formula (2-11), Definition and Similarly, Representative The node is the drone in the The landing node in the flight.

[0057] Constraint 11: The takeoff and landing nodes of each drone's delivery path are different, ensuring that the node used to launch the drone and the node provided for the drone to land are different in a single trip of the same drone, as shown in the following formula (2-12): (2-12).

[0058] The special delivery constraints of the first drone include the following constraints 12 to 15: Constraint 12: Each first UAV serves at least one node, and the delivery path of each first UAV is different from the delivery path of the vehicle to prevent the first UAV from flying along the route of the vehicle, as shown in the following formulas (2-13) and (2-14): (2-13); In formula (2-13), The meaning and Similarly, Indicates the The first node is the first drone in the The take-off node in the flight; and The meaning and Similarly, Representing the first drone in Visit the first flight nodes, Indicates that the first drone is in Visit the first flight nodes; (2-14); In formula (2-14), The meaning and Similarly, Indicates the The first node is the first drone in the The landing node in the flight.

[0059] Constraint 13: Due to the capacity characteristics of the first drone's single storage, after performing the delivery service, the first drone can fly to another node that needs pickup service. This stipulates that the first drone must first serve the node with delivery needs, and then serve the node with pickup needs. In other words, the priority of each first drone relative to the delivery node is higher than the priority of each first drone relative to the shipping node, as shown in the following formulas (2-15) and (2-16): (2-15); In formula (2-15), Represents a collection of delivery nodes; (2-16).

[0060] Constraint 14: After completing its service for a shipping node and / or a delivery node, each first drone must fly to the corresponding landing node. That is, after completing its service for a node that provides shipping services or both shipping and delivery services, the first drone must fly directly to the landing point, as shown in the following formula (2-17): (2-17); In formula (2-17), The meaning and Similarly, Indicates that the vehicle visits nodes.

[0061] Constraint 15: Due to the limited battery life of the first UAV, the first UAV is subject to flight range and payload restrictions. This limits the delivery path of each first UAV to a maximum flight range. That is, the flight distance (delivery path) of the first UAV in a single flight must not exceed the flight range of the first UAV, as shown in the following formula (2-18): (2-18); In formula (2-18), Indicates the maximum endurance of the first UAV, which is a pre-calibrated value. Optionally, the maximum endurance of the first UAV is in the range of [5 km, 10 km], but is not limited thereto.

[0062] The special delivery constraints for the second drone include the following constraints 16 to 18: Constraint 16: Each second drone serves at least one delivery node, and the delivery path of each second drone is different from the delivery path of the vehicle to prevent the second drone from flying along the route of the vehicle, as shown in the following formulas (2-19) and (2-20): (2-19); In formula (2-19), The meaning and Similarly, Indicates the The second UAV is in the The take-off node in the flight; and The meaning and Similarly, Indicates that the second drone is in Visit the first flight nodes, Indicates that the second drone is in Visit the first flight nodes; (2-20); In formula (2-20), The meaning and Similarly, Indicates the The second UAV is in the The landing node in the flight.

[0063] Constraint 17: Due to the limited battery life of the second UAV, the second UAV is subject to flight range and payload restrictions. This limits the delivery path of each second UAV to a maximum flight range. That is, the flight distance (delivery path) of the second UAV in a single flight must not exceed the flight range of the second UAV, as shown in the following formula (2-21): (2-21); In formula (2-21), Indicates the maximum endurance of the second UAV, which is a pre-calibrated value. Optionally, the maximum endurance of the second UAV is in the range of [5 km, 10 km], but is not limited thereto.

[0064] Constraint 18: Since the second UAV can carry multiple cargoes during a single flight, it is required that the total weight of all cargoes carried by the second UAV during each flight does not exceed the maximum payload of the UAV. In other words, the payload of each second UAV does not exceed the corresponding maximum payload, as shown in the following formula (2-22): (2-22); In formula (2-22), Indicates the The cargo weight of each node, It indicates the maximum weight that the second UAV can bear, that is, the maximum load, which can be preset according to actual conditions.

[0065] As can be seen, this embodiment first imposes overall constraints on the delivery behavior of drones based on their takeoff and landing nodes, as well as the number of flights. This achieves preliminary constraints on the delivery behavior of drones in the path planning model while fully considering the common characteristics of the two types of drones. It then further imposes specific constraints on the delivery behavior of the first drone based on its service to the node, its delivery path, landing conditions, and endurance. Furthermore, it imposes specific constraints on the delivery behavior of the second drone based on its service to the node, its delivery path, payload, and endurance. This achieves precise constraints on the delivery behavior of the first and second drones in the path planning model while fully considering the delivery characteristics of the two types of drones, improving the adaptability of the constraints to the drone types. In this way, the delivery behavior of the two types of drones in vehicle-machine collaborative delivery is constrained based on the common characteristics of drones and the specific characteristics of the two types of drones, thereby accurately constraining the delivery behavior of the two types of drones and helping to improve the accuracy of the path planning model.

[0066] 2.5. Vehicle delivery constraints include the following constraints 19 to 22: Constraint 19: All takeoff nodes and landing nodes in the delivery path of each drone must be visited and served by vehicles. That is, all takeoff nodes and all landing nodes must be visited and served by vehicles, as shown in the following formulas (2-23) and (2-24): (2-23); (2-24); In formula (2-23), The meaning and Similarly, Indicates that the vehicle is Drive to the node nodes.

[0067] Constraint 20: In a vehicle's delivery path, the vehicle must travel from one node to another, and then from another node to another node to balance the vehicle flow at any node and ensure that all nodes in the vehicle's delivery path are visited once by the vehicle, as shown in the following formula (2-25): (2-25); In formula (2-25), The meaning and Similarly, Indicates that the vehicle is Drive to the node nodes.

[0068] Constraint 21: The vehicle departs from the warehouse node and returns to the warehouse node when the service is completed. That is, the vehicle can leave and return to the warehouse exactly once, as shown in the following formula (2-26): (2-26); In formula (2-26), The meaning and Similarly, Indicates the warehouse node where the vehicle starts Drive to nodes.

[0069] Constraint 22: Since this application considers the case where a vehicle carries multiple drones, the sum of the number of drones on board when the vehicle leaves the warehouse node and the number of drones that take off from the warehouse node must be equal to the total number of drones, as shown in the following formula (2-27): (2-27); In formula (2-27), Indicates that the vehicle is leaving the starting warehouse node The number of drones carried at the time; Indicates that the drone is The warehouse node from which the flight started Fly to nodes.

[0070] It can be seen that this embodiment imposes special constraints on the vehicle's delivery behavior from the aspects of the vehicle's access to nodes, the number of drones carried by the vehicle, and the vehicle's round-trip situation. In this way, the vehicle's delivery behavior in vehicle-machine collaborative delivery is constrained while fully considering the vehicle's delivery characteristics, thereby accurately constraining the vehicle's delivery behavior and helping to improve the accuracy of the path planning model.

[0071] In some embodiments, reference Figure 2 , the above-mentioned adaptive large neighborhood search algorithm is used to optimize the path planning model to obtain the best delivery solution, which may include: Based on the delivery scenario, an initial delivery plan is obtained as the current delivery plan; In each round of iteration, a target destruction operator is selected from multiple destruction operators, the target destruction operator is used to destroy the current distribution plan, and the repair operator is used to repair the damaged current distribution plan to obtain a feasible distribution plan. Based on the fitness value of the feasible distribution plan, it is determined whether the feasible distribution plan is not accepted by the Metropolis criterion. If so, the process returns to the step of selecting the target destruction operator from the multiple destruction operators. Otherwise, the feasible distribution plan is determined as the current distribution plan and the process jumps to the next round of iteration. The fitness value of the feasible distribution plan is obtained through the path planning model, and both the destruction operator and the repair operator are associated with the path planning model. The current delivery plan at the end of the last iteration is determined as the optimal delivery plan.

[0072] It should be noted that the distribution plan includes the distribution path of the vehicle and the distribution path of multiple drones. The distribution path includes at least one node. This definition applies to the initial distribution plan, current distribution plan and feasible distribution plan of the embodiment of this application.

[0073] In this implementation, an adaptive large neighborhood search algorithm is used as an effective tool for solving complex path planning models. The process is as follows: The first step is to generate an initial delivery plan (i.e., an initial solution) based on the delivery scenario provided in the embodiment of the present application, and determine the initial delivery plan as the current delivery plan (i.e., the current solution).

[0074] Optionally, the method for generating the initial delivery plan can be customized based on practical needs, and this embodiment does not impose any restrictions. For example, to accelerate the optimization convergence of the path planning model and reduce the number of iterations, the initial delivery plan can be generated using any existing heuristic algorithm, such as the greedy algorithm or the nearest neighbor algorithm. Of course, the initial delivery plan can also be generated randomly, but is not limited to this. Taking the nearest neighbor algorithm as an example, all nodes are first clustered using the K-means algorithm to obtain K regions. Next, a node is randomly selected from each region to form an initial set of nodes visited by the vehicle. Based on this initial set of nodes visited by the vehicle, the nearest neighbor algorithm is then used to plan an initial delivery route for the vehicle. The remaining nodes are then assigned one by one to a drone or truck for access. Specifically, a node is randomly selected from the remaining nodes and assigned to a drone of the corresponding type based on its service type. If a drone cannot meet the requirements for accessing the node, it is assigned to a vehicle; otherwise, it is assigned to a drone. This method accurately generates the initial delivery plan, ensuring both accuracy and efficiency.

[0075] The second step, in In a round of iteration, first, a destruction operator is selected from multiple destruction operators as the target destruction operator. The destruction operator is used to destroy the current solution using its configured destruction operation to obtain a partial solution. Then, the target destruction operator is used to destroy the current distribution plan to obtain the destroyed current distribution plan, and the repair operator is used to repair the destroyed current distribution plan to obtain a feasible distribution plan. The repair operator uses its configured repair operation to repair the partial solution to obtain a new feasible solution. In this embodiment, both the repair operator and the destruction operator are associated with the constraint set of the path planning model to constrain the delivery behavior of vehicles and drones. Afterwards, based on the Metropolis criterion of simulated annealing and the fitness value of the feasible distribution plan, it is decided whether to accept the feasible distribution plan; wherein, the fitness value of the feasible distribution plan is calculated by the fitness function of the path planning model; if accepted, the feasible distribution plan is determined as the current distribution plan, and it is judged Has the maximum number of iterations been reached? If so, jump to step 3, otherwise jump to step round of iteration; otherwise, keep the current distribution plan unchanged and return to the step of selecting the target destruction operator from multiple destruction operators.

[0076] Optionally, the destruction operator selection method can be set based on actual circumstances, and this embodiment does not impose any restrictions on this. For example, the destruction operator selection method can be roulette selection, probabilistic selection, random selection, etc., but is not limited to these. In addition, the maximum number of iterations can be set based on actual circumstances, and this embodiment does not impose any restrictions on this.

[0077] The third step is to determine the current delivery plan at the end of the last round of iteration as the optimal delivery plan (i.e., the optimal solution). The optimal delivery plan satisfies the requirements of minimizing delivery costs and satisfying the delivery behaviors constrained by vehicles and drones.

[0078] It can be seen that in this embodiment, the adaptive large neighborhood search algorithm is used as an effective tool for solving complex path planning models. It can solve the optimal delivery plan with the purpose of minimizing delivery costs while dealing with the constraints on the delivery behavior of vehicles and drones, helping vehicles and drones to plan reasonable delivery paths. At the same time, the acceptance criterion of the simulated annealing algorithm can reduce the possibility of falling into local optimality, thereby balancing the needs of delivery, mailing and simultaneous mailing, meeting diverse delivery needs, effectively improving the efficiency and accuracy of vehicle-machine collaborative delivery, and reducing the cost of vehicle-machine collaborative delivery.

[0079] The specific implementation of the adaptive large neighborhood search algorithm will be further explained below.

[0080] Existing adaptive large neighborhood search is usually applicable to the multi-flying partner traveling salesman problem with a single delivery demand, that is, its destruction operator and repair operator are only applicable to exploring solutions that meet a single delivery demand. However, the delivery scenario of the embodiment of the present application not only considers three different types of nodes, but also two different types of drones. If the adaptive large neighborhood search is continued to be used to explore the optimal solution, it may occur that the destruction operator cannot fully destroy the current solution, and some types of nodes cannot be allocated during repair, resulting in low accuracy of vehicle-machine collaborative delivery. For the delivery scenario with three different types of nodes and two different types of drones considered in the embodiment of the present application, the embodiment of the present application introduces a new repair operator and multiple new destruction operators into the adaptive large neighborhood search algorithm, and adjusts the adaptive weight mechanism of the destruction operator, thereby effectively improving the accuracy of vehicle-machine collaborative delivery, which is as follows: In some embodiments, selecting a target destruction operator from a plurality of destruction operators may include: According to the adaptive weights corresponding to the multiple destruction operators, the selection probability of each destruction operator is obtained; According to the selection probabilities corresponding to the multiple destruction operators, a destruction operator is selected from the multiple destruction operators as a target destruction operator.

[0081] In this embodiment, each destruction operator is assigned a specific weight, which is used to determine the probability of the destruction operator being selected. After each iteration of the destruction operator and the repair operator, the weight of the destruction operator used will change adaptively, so this weight is also called adaptive weight. First, obtain the weight of the destruction operator in the first The adaptive weights corresponding to multiple destruction operators in the iteration are The adaptive weight of the destruction operator is denoted as Then, for the In the round of iteration, the adaptive weights corresponding to multiple destruction operators are calculated and processed to obtain the weights of each destruction operator in the first The selection probability in the round iteration satisfies the following formula (3-1): (3-1); In formula (3-1), Indicates the The selection probability of a destruction operator in the current iteration, the sum of the selection probabilities of all destruction operators is one; Represents the sum of the adaptive weights of all destruction operators in the current iteration.

[0082] Next, a destruction operator is selected as the target destruction operator based on the selection probabilities corresponding to each of the multiple destruction operators. For example, if there are three destruction operators with weights of 1, 1, and 0.5, the selection probabilities of the first and second destruction operators are 1 / (1+1+0.5)=0.4, and the selection probability of the third destruction operator is 0.5 / (1+1+0.5)=0.2. Therefore, the probabilities of these three destruction operators being selected as the target destruction operator are 40%, 40%, and 20%, respectively. Based on these probabilities, one destruction operator is selected as the target destruction operator.

[0083] In this way, on the one hand, the probability of selecting a destruction operator is based on its historical performance (i.e., adaptive weight), rather than a fixed strategy. For example, if a destruction operator successfully generates an optimal solution over multiple iterations, its adaptive weight will increase accordingly, increasing its probability of selection. Conversely, the adaptive weight of a poorly performing destruction operator will decrease, reducing its chance of selection. This approach encourages high-performing operators to be selected frequently, accelerating the discovery of the global optimal solution and gradually eliminating inefficient destruction operators, reducing the waste of computational resources. Furthermore, even if a destruction operator has a high adaptive weight, other destruction operators still retain a certain probability of selection. This ensures that low-probability destruction operators still have a chance of selection, avoiding complete reliance on high-performing destruction operators. This ensures that the optimization process continuously explores different solution spaces, reducing the risk of prematurely falling into local optima, and thus balancing the chances of different destruction operators being selected.

[0084] In some embodiments, the plurality of destruction operators may include: The first destruction operator is used to assign any node in the delivery path of the vehicle that is not a take-off node or a landing node of the UAV to any UAV; The second destruction operator is used to assign any node in the delivery path of the target drone to a vehicle; wherein the target drone is a randomly selected drone; A third destruction operator is used to swap the delivery path of the target UAV with the path segment of the vehicle's delivery path that corresponds to the delivery path of the target UAV; The fourth destruction operator is used to swap nodes in the delivery paths of any two drones; The fifth destruction operator is used to assign any node in the delivery path of one of the drones to another drone; The sixth destruction operator is used to adjust the take-off node or landing node of the target UAV; The seventh destruction operator is used to randomly select a preset number of nodes in the current delivery plan, delete the selected nodes, and delete the delivery paths of the drones corresponding to the selected nodes that are take-off nodes or landing nodes.

[0085] In this embodiment, the destruction operator will obtain a partial solution by destroying the current solution in each iteration. According to the distribution scenario of the embodiment of the present application, this embodiment provides seven destruction operators to update the current distribution plan. The proposed destruction operators not only include the replanning of single or multiple nodes between vehicles and drones, and between drones, but also involve the replanning of the take-off and landing points of drones. These seven destruction operators cover various changes in the distribution scenario of the embodiment of the present application, providing a basis for comprehensive exploration of the solution space.

[0086] The following will be combined with the Figure 3 The destruction operator provided in the embodiments of the present application is described in detail.

[0087] The first destruction operator is used to assign any node in the delivery path of the vehicle that is not a take-off node or a landing node of the drone to any drone. Specifically, Figure 3As shown in (a), among the multiple nodes in the vehicle's delivery path that are not the take-off and landing nodes of the drone, a node is randomly selected as a destruction node, the destruction node is deleted from the vehicle's delivery path, and the destruction node is randomly assigned to any drone. Then, an algorithm such as the nearest neighbor algorithm is used to replan a new delivery path for the drone with the destruction node. If the new delivery path is infeasible, the destruction node is deleted from the current delivery plan. Otherwise, the destruction node is retained from the current delivery plan, thereby obtaining the current delivery plan after destruction. Among them, if the delivery path satisfies the constraint set of the path planning model, the delivery path is determined to be feasible; otherwise, the delivery path is determined to be infeasible. The first destruction operator provides flexible service opportunities for drones by releasing fixed nodes in the vehicle's delivery path. It can effectively utilize the advantage of drones' straight-line flight to shorten delivery time and improve delivery efficiency, thereby forming a new partial solution.

[0088] The second destruction operator is used to assign any node in the delivery path of the target drone to a vehicle; wherein the target drone is a randomly selected drone. Specifically, Figure 3 As shown in (b), a drone is randomly selected, denoted as the target drone, and a node is randomly selected from the target drone's delivery path as a destruction node. The destruction node is removed from the target drone's delivery path and assigned to the vehicle for access. Next, an algorithm such as the nearest neighbor algorithm is used to replan a new delivery path for the vehicle with the destruction node. If the new delivery path is not feasible, the destruction node is removed from the current delivery plan. Otherwise, the destruction node is retained from the current delivery plan, thus obtaining the current delivery plan after destruction. The second destruction operator assigns certain nodes in the drone's delivery path that are close to the vehicle to the vehicle. This allows these nodes to be visited without launching a single drone, effectively shortening delivery time and cost, improving delivery efficiency, and thus forming a new partial solution.

[0089] The third destruction operator is used to exchange the delivery path of the target UAV with the path segment corresponding to the delivery path of the vehicle. Specifically, Figure 3 As shown in (c), a drone is randomly selected as the target drone. The target drone's delivery path is determined as the destruction path. The path segment corresponding to the destruction path is then extracted from the vehicle's delivery path. This path segment, consisting of the takeoff and landing nodes of the destruction path, is then swapped with the destruction path to obtain the current delivery solution after the destruction. The third destruction operator provides a variety of options for the vehicle's delivery path. Compared to a single-point swap approach, this increases the exploration of the solution space and encourages the formation of more and better partial solutions.

[0090] The fourth destruction operator is used to exchange nodes in the delivery paths of any two drones. Specifically, Figure 3 As shown in (d), two drones are randomly selected, denoted as the first target drone and the second target drone. Next, a node is randomly selected from the first target drone's delivery path as the first destruction node, and a node is randomly selected from the second target drone's delivery path as the second destruction node. Finally, the first destruction node and the second destruction node are swapped to obtain the current delivery solution after destruction. Optionally, to improve the scalability of the destruction operator and expand the explorability of the solution space, the method of randomly selecting any two drones in the fourth destruction operator can be further limited to randomly selecting a first drone and a second drone, thereby swapping nodes in the delivery paths of two different types of drones. This facilitates global exploration of the solution space and generates more high-quality solutions.

[0091] The fifth destruction operator is used to assign any node in the delivery path of one of the drones to another drone. Specifically, Figure 3 As shown in (e), two drones are randomly selected, denoted as the first target drone and the second target drone. Next, a node is randomly selected from the first target drone's delivery path as a destruction node. This node is removed from the first target drone's delivery path and inserted into the second target drone's delivery path with the goal of minimizing flight distance, thereby obtaining the current delivery plan after destruction. Minimizing flight distance refers to minimizing the distance between the drones during that flight segment (takeoff node, node 1, ..., landing node). Optionally, to improve the scalability of the destruction operator and expand the explorability of the solution space, the random selection of two drones in the fifth destruction operator can be further limited to randomly selecting a first drone and a second drone, thereby assigning the delivery path of one type of drone to another type. This facilitates global exploration of the solution space and generates more high-quality solutions.

[0092] The fourth and fifth destruction operators aim at the node allocation among multiple drones and try to obtain shorter drone paths through local adjustments, effectively shortening the delivery time and cost, improving the delivery efficiency, and thus forming a new partial solution.

[0093] The sixth destruction operator is used to adjust the take-off node or landing node of the target UAV. Specifically, Figure 3As shown in (f), a random drone is selected, denoted as the target drone. The target drone's takeoff node is then randomly selected and adjusted to the node immediately before or after the takeoff node. Alternatively, the target drone's landing node is randomly selected and adjusted to the node immediately before or after the landing node, thereby obtaining the current delivery plan after the disruption. The sixth disruption operator focuses on optimizing the connection hub between drones and vehicles. This significantly impacts the feasibility of subsequent paths, encourages more diverse solutions, and ensures that the solution space is fully explored.

[0094] The seventh destruction operator is used to randomly select a preset number of nodes in the current delivery plan, delete the selected nodes, and delete the delivery paths of the drones corresponding to the selected nodes as take-off nodes or landing nodes. Specifically, Figure 3 As shown in (g), a preset number of nodes are randomly selected from all nodes and marked as destruction nodes. All destroyed nodes are removed from the current delivery plan. If the destroyed node is a takeoff or landing node, the delivery path of the drone it is located in is also deleted. The seventh destruction operator acts as a global perturbation. By forcibly removing nodes (including the drone's takeoff and landing nodes), the structure of the current delivery plan is disrupted, preventing it from falling into local optimality and ensuring the explorability of the solution space.

[0095] Optionally, the preset number can be set according to actual conditions, and this embodiment does not limit this. For example, the range of the preset number is , but not limited to this.

[0096] It can be seen that the above seven operators can fully destroy the current solution according to the delivery scenario of the embodiment of this application, achieve a balance between the breadth and depth of exploration of the solution space, provide a basis for comprehensive exploration of the solution space, ensure a full global search of the solution space, and at the same time increase the diversity and excellence of the solution, laying the foundation for finding the optimal solution, thereby effectively improving the accuracy of vehicle-machine collaborative delivery.

[0097] In some embodiments, the above-mentioned repair operator is used to repair the damaged current delivery plan to obtain a feasible delivery plan, which may include: Insert the current repair node into the delivery path of the current drone, and perform path planning on the current drone to obtain the first candidate path for the current drone; If the first candidate path of the current UAV meets the feasibility conditions, the current repair node is assigned to the current UAV and jumps to the next repair node. Otherwise, the current UAV serves the current repair node alone, and the path planning is performed on the current UAV to obtain the second candidate path of the current UAV. If the second candidate path of the current UAV meets the feasibility condition, the current repair node is assigned to the current UAV and jumps to the next repair node; otherwise, jumps to the next UAV until all UAVs are traversed; If the current repair node does not meet the feasibility conditions relative to the second candidate paths of all drones, the current repair node is assigned to the vehicle and jumps to the next repair node until all repair nodes are traversed and a feasible delivery plan is obtained; Among them, the repair node represents any one of the nodes that do not exist in the current distribution plan after being destroyed by the seventh destruction operator, or the nodes that do not satisfy the constraint condition set of the path planning model in the current distribution plan after being destroyed by any destruction operator other than the seventh destruction operator.

[0098] In this embodiment, a repair operator is used to repair the current delivery plan (i.e., an infeasible solution) after being corrupted by the destruction operator, which does not conform to the constraints of the routing model, to obtain a feasible delivery plan (i.e., a new feasible solution). Of the seven destruction operators mentioned above, the seventh destruction operator produces a partial solution because some nodes are not assigned, necessitating direct use of the destruction operator to repair them. Other destruction operators, other than the seventh, produce complete solutions, requiring the use of constraints to determine whether they are infeasible solutions before determining whether they require repair. To improve the quality of feasible solutions, this embodiment constructs a heuristic repair operator that prioritizes assigning nodes to drones with lower delivery costs. If a drone cannot access a node, it is assigned to a vehicle. Specifically, repair nodes are divided into two types: nodes that do not exist in the current delivery plan after being corrupted by the seventh destruction operator, and nodes that do not conform to the constraints of the routing model after being corrupted by any destruction operator other than the seventh destruction operator. This method allows for the generation of repair nodes for the current delivery plan after being corrupted by different destruction operators. For each repair node, the following repair operations are performed: The first step is to insert the current repair node into the current drone's delivery path and replan the drone's optimal path using an algorithm such as the nearest neighbor algorithm to obtain the current drone's first candidate path. The first candidate path for the current drone includes the current repair node and all nodes in the current drone's delivery path.

[0099] The second step is to determine whether the current drone's first candidate path meets the feasibility conditions. Feasibility conditions refer to the set of constraints that satisfy the path planning model. If so, the current drone's first candidate path is feasible. The current repair node is assigned to the current drone, and the route jumps to the next repair node to achieve traversal repair. If not, the current drone's first candidate path is infeasible. The current drone will then independently serve the current repair node, performing the delivery task at the current repair node independently. A new path, such as the nearest neighbor algorithm, is then planned to obtain the current drone's second candidate path. The current drone's second candidate path includes the current repair node, as well as the current drone's takeoff and landing nodes relative to the current repair node, allowing the drone to independently access the current repair node.

[0100] The third step is to determine whether the second candidate path for the current drone meets the feasibility criteria. If so, the second candidate path for the current drone is feasible. The current repair node is assigned to the current drone, and the process jumps to the next repair node to complete the traversal repair. If not, the second candidate path for the current drone is infeasible. The process jumps to the next drone to complete the traversal repair until all drones have been traversed.

[0101] In the fourth step, when all drones have been traversed, if the current repair node does not meet the feasibility conditions relative to the second candidate path of all drones, it means that all drones cannot access the current repair node. At this time, it is inserted into the delivery path of the vehicle with the minimum driving distance as the goal, and then jumps to the next repair node to achieve traversal repair until all repair nodes are traversed. At this time, all repair nodes are inserted into the delivery plan and a feasible delivery plan is obtained.

[0102] It can be seen that this embodiment takes into account that the flexibility of drones in vehicle-machine collaborative delivery is higher than the flexibility of vehicles in vehicle-machine collaborative delivery. Therefore, the repair operator first considers inserting the partially solved repair node into the drone's delivery path, and adopts a two-stage insertion mechanism to realize the insertion of the repair node, that is, first integrating it into the existing drone delivery path, and secondly selecting the drone delivery path that serves separately. This can accurately evaluate whether the drone can serve the repair node and allocate reasonable drones to the repair node, which helps to ensure the service efficiency of the repair node. Secondly, when all drones cannot access the node, consider inserting it into the vehicle's delivery path. This can ensure that the repair node has the corresponding path insertion, avoiding the situation where the repair node is isolated from the delivery plan, thereby improving the repair accuracy.

[0103] In some embodiments, after using the repair operator to repair the damaged current delivery plan to obtain a feasible delivery plan, the method further includes the following steps: According to the first judgment information, the second judgment information and the fitness value of the feasible delivery plan, combined with the fitness value of the current delivery plan, the adaptive weight of the target destruction operator is updated; wherein, the first judgment information is used to indicate whether the feasible delivery plan is the current optimal solution of the adaptive large neighborhood search algorithm, and the second judgment information is used to indicate whether the feasible delivery plan is accepted by the Metropolis criterion.

[0104] In this embodiment, every time a round of solution iteration is completed through the destruction operator and the repair operator, the weight value of the destruction operator used will be adaptively adjusted according to whether the new solution is accepted by the Metropolis criterion and whether the new solution is the current optimal solution. In this way, by increasing the adaptive weight of excellent destruction operators and reducing the adaptive weight of poor destruction operators, the probability of the destruction operator being selected can be adaptively changed according to the quality of the current solution, so that better destruction operators stand out and the probability of poor destruction operators being selected is reduced, thereby accelerating the convergence of the adaptive large neighborhood search algorithm and improving the efficiency of vehicle-machine collaborative delivery. Specifically, referring to Figure 4 , divided into the following four cases, the first value is the largest, the fourth value is the smallest, and the second value is greater than the third value: If the first judgment information is used to indicate that the feasible delivery plan is the current optimal solution of the adaptive large neighborhood search algorithm, then it means that the feasible delivery plan is the current optimal solution, which means that the target destruction operator performs well. High-frequency use of the target destruction operator helps to accelerate the discovery of the global optimal solution. At this time, the adaptive weight of the target destruction operator is replaced with the first value.

[0105] If the first judgment information is used to indicate that the feasible distribution plan is not the current optimal solution of the adaptive large neighborhood search algorithm, and the second judgment information is used to indicate that the feasible distribution plan is accepted by the Metropolis criterion, and the fitness value of the feasible distribution plan is greater than or equal to the fitness value of the current distribution plan, then it means that although the feasible distribution plan is not the optimal solution, the feasible distribution plan is accepted because its fitness value is better than the old solution, which means that the target destruction operator performs well, and the appropriate use of the target destruction operator helps to accelerate the discovery of the global optimal solution. At this time, the adaptive weight of the target destruction operator is replaced with the second value.

[0106] If the first judgment information is used to indicate that the feasible distribution plan is not the current optimal solution of the adaptive large neighborhood search algorithm, and the second judgment information is used to indicate that the feasible distribution plan is accepted by the Metropolis criterion, and the fitness value of the feasible distribution plan is less than the fitness value of the current distribution plan, then it means that although the feasible distribution plan is not the optimal solution, and the fitness value of the feasible distribution plan is worse than that of the old solution, it is still accepted, which means that the performance of the target destruction operator is average, and it can be called but not multiple times. At this time, the adaptive weight of the target destruction operator is replaced with the third value.

[0107] If the second judgment information is used to indicate that the feasible delivery plan is not accepted by the Metropolis criterion, it means that the feasible delivery plan is not a solution that satisfies the path planning model, which means that the target destruction operator performs poorly and should not be called multiple times. At this time, the adaptive weight of the target destruction operator is replaced with the fourth value.

[0108] It should be understood that the fitness value of the delivery plan is calculated by the fitness function of the path planning model.

[0109] Optionally, the first value, the second value, the third value, and the fourth value can be set according to actual conditions, and this embodiment does not impose any specific restrictions on this. For example, the first value is 1, the second value is 0.8, the third value is 0.6, and the fourth value is 0.4, but the present invention is not limited thereto.

[0110] In addition, refer to Figure 5 , the embodiment of the present application further provides a vehicle-heterogeneous drone collaborative delivery device, which may include: A first processing module 201 is configured to construct a path planning model based on a delivery scenario, wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, wherein the node includes any one of a delivery node, a consignment node, or a consignment-delivery node, and the drone includes any one of a first drone and a second drone, wherein the second drone is only used to access the delivery node. The path planning model aims to minimize delivery costs and constrains the delivery behaviors of the vehicles and drones. The second processing module 202 is used to optimize the path planning model using an adaptive large neighborhood search algorithm to obtain the best delivery solution; The third processing module 203 is used to control the vehicle and multiple drones to perform corresponding delivery actions according to the optimal delivery plan.

[0111] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0112] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and intent of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0113] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A vehicle-heterogeneous drone collaborative delivery method, characterized in that: The following steps are involved: Constructing a path planning model based on a delivery scenario; wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, wherein the nodes include any one of a delivery node, a consignment node, or a consignment-delivery node; the drone includes any one of a first drone and a second drone, wherein the second drone is used only to access the delivery node; and the path planning model aims to minimize delivery costs and constrains the delivery behaviors of the vehicle and the drone. The path planning model is optimized using an adaptive large neighborhood search algorithm to obtain the best delivery solution; According to the optimal delivery plan, the vehicle and the plurality of drones are controlled to perform corresponding delivery actions.

2. The method according to claim 1, characterized in that The path planning model is constructed based on the delivery scenario, including: Constructing a fitness function based on attribute information associated with the delivery behavior of the vehicle and the drone; wherein the fitness function is used to characterize the minimization of the delivery cost; Determining a set of constraints for the fitness function; wherein the set of constraints is used to constrain the delivery behavior of the vehicle and the drone; The path planning model is constructed according to the fitness function and the set of constraints.

3. The method according to claim 2, characterized in that The attribute information includes a node edge set, a flight set of the UAV, a unit distance cost of the vehicle, a unit distance cost of the UAV, driving attribute information of the vehicle, and driving attribute information of the UAV; The constructing of a fitness function based on attribute information associated with the delivery behavior of the vehicle and the drone includes: Constructing a first cost function based on the unit distance cost and driving attribute information of the vehicle and the node edge set; Constructing a second cost function based on the flight set, unit distance cost, and travel attribute information of the UAV and the node edge set; The fitness function is constructed according to the first cost function and the second cost function.

4. The method according to claim 2, characterized in that The constraint condition set includes: the delivery constraint conditions of the node, the overall delivery constraint conditions of the drone, the special delivery constraint conditions of the first drone, the special delivery constraint conditions of the second drone and the delivery constraint conditions of the vehicle.

5. The method according to claim 1, wherein The method of optimizing the path planning model using an adaptive large neighborhood search algorithm to obtain an optimal delivery solution includes: Based on the delivery scenario, an initial delivery plan is obtained as the current delivery plan; In each round of iteration, a target destruction operator is selected from the plurality of destruction operators, the target destruction operator is used to destroy the current distribution plan, and the repair operator is used to repair the destroyed current distribution plan to obtain a feasible distribution plan. Based on the fitness value of the feasible distribution plan, it is determined whether the feasible distribution plan is not accepted by the Metropolis criterion. If so, the process returns to the step of selecting the target destruction operator from the plurality of destruction operators. Otherwise, the feasible distribution plan is determined as the current distribution plan, and the process jumps to the next round of iteration; wherein the fitness value of the feasible distribution plan is obtained by the path planning model, and both the destruction operator and the repair operator are associated with the path planning model; Determine the current delivery plan at the end of the last round of iteration as the optimal delivery plan; The distribution plan includes the distribution path of the vehicle and the distribution paths of the plurality of drones, and the distribution path includes at least one node.

6. The method according to claim 5, characterized in that The selecting a target destruction operator from the plurality of destruction operators comprises: Obtaining a selection probability of each of the destruction operators according to the adaptive weights corresponding to the plurality of destruction operators; According to the selection probabilities respectively corresponding to the multiple destruction operators, one destruction operator is selected from the multiple destruction operators as the target destruction operator.

7. The method according to claim 5, characterized in that The plurality of said destruction operators include: a first destruction operator, configured to assign any node in the delivery path of the vehicle that is not a take-off node or a landing node of the drone to any one of the drones; a second destruction operator, configured to assign any one of the nodes in the delivery path of the target UAV to the vehicle; wherein the target UAV is a randomly selected UAV; a third destruction operator, configured to swap the delivery path of the target UAV with a path segment in the delivery path of the vehicle that corresponds to the delivery path of the target UAV; a fourth destruction operator, configured to exchange the nodes in the delivery paths of any two of the drones; a fifth destruction operator, configured to assign any node in the delivery path of one of the drones to another drone; a sixth destruction operator, configured to adjust a takeoff node or a landing node of the target UAV; The seventh destruction operator is used to randomly select a preset number of nodes in the current delivery plan, delete the selected nodes, and delete the delivery paths of the drones corresponding to the selected nodes that serve as the take-off nodes or the landing nodes.

8. The method according to claim 5, characterized in that The repairing operator is used to repair the damaged current delivery plan to obtain a feasible delivery plan, including: Inserting the current repair node into the delivery path of the current drone, and performing path planning for the current drone to obtain a first candidate path for the current drone; If the first candidate path of the current UAV meets the feasibility condition, the current repair node is assigned to the current UAV and the process jumps to the next repair node. Otherwise, the current UAV serves the current repair node alone, and a path is planned for the current UAV to obtain a second candidate path for the current UAV. If the second candidate path of the current UAV meets the feasibility condition, the current repair node is assigned to the current UAV, and the process jumps to the next repair node; otherwise, the process jumps to the next UAV until all the UAVs are traversed. If the current repair node does not meet the feasibility condition relative to the second candidate paths of all the drones, the current repair node is assigned to the vehicle and the process jumps to the next repair node until all the repair nodes are traversed to obtain the feasible delivery solution. Among them, the repair node represents any one of the nodes that does not exist in the current delivery plan after being destroyed by the seventh destruction operator, or the nodes that do not satisfy the constraint condition set of the path planning model in the current delivery plan after being destroyed by any destruction operator other than the seventh destruction operator.

9. The method according to claim 5, characterized in that After repairing the damaged current delivery plan by using the repair operator to obtain a feasible delivery plan, the method further includes the following steps: updating the adaptive weight of the target destruction operator according to the first determination information, the second determination information, and the fitness value of the feasible delivery plan, in combination with the fitness value of the current delivery plan; The first determination information is used to indicate whether the feasible delivery plan is the current optimal solution of the adaptive large neighborhood search algorithm, and the second determination information is used to indicate whether the feasible delivery plan is accepted by the Metropolis criterion.

10. A vehicle-heterogeneous drone collaborative delivery device, characterized in that: include: A first processing module is configured to construct a path planning model based on a delivery scenario; wherein the delivery scenario includes a vehicle, multiple drones, and multiple nodes, wherein the nodes include any one of a delivery node, a consignment node, or a consignment-delivery node; the drone includes any one of a first drone and a second drone, wherein the second drone is used only to access the delivery node; and the path planning model aims to minimize delivery costs and constrains the delivery behaviors of the vehicle and the drone. The second processing module is used to optimize the path planning model using an adaptive large neighborhood search algorithm to obtain an optimal delivery solution; The third processing module is used to control the vehicle and the plurality of drones to perform corresponding delivery actions according to the optimal delivery plan.

Images