Dynamic pick-up and delivery vehicle routing method and apparatus

Abstract

The application provides a dynamic goods taking and delivering vehicle path scheduling method and device, relates to the technical field of vehicle path scheduling, and particularly relates to a dynamic goods taking and delivering vehicle path scheduling method and device. The application divides a scheduling period into multiple rolling time slices, generates an initial distribution path in the first time slice, and re-plans in subsequent time slices based on newly added orders and an updated running state formed by a vehicle executing a previous distribution path. The vehicle binding node and the re-assignable node are determined based on the task execution state, the current target node is taken as the starting point of the re-planning path of the corresponding vehicle, and the delivery node paired with the current target taking node and the delivery node of the taken but not delivered order are kept in the corresponding vehicle path; the remaining unbound orders are re-assigned across vehicles in pairs of taking and delivering nodes. Therefore, the application can maintain the continuity of vehicle tasks, reduce path disturbance, and improve the adaptability of dynamic scheduling.

Description

Technical Field

[0001] This invention relates to the field of vehicle route scheduling technology, specifically to a method and apparatus for dynamic pickup and delivery vehicle route scheduling. Background Technology

[0002] In dynamic pickup and delivery scenarios such as wholesale logistics, retail logistics, and on-demand delivery, pickup and delivery orders typically include corresponding pickup and delivery tasks. The dispatch platform needs to determine vehicle task allocation and delivery routes based on order information, vehicle status, and road network information. Since orders may continuously arrive during vehicle operation, vehicle delivery routes need to be dynamically updated according to order changes.

[0003] In related technologies, scheduling platforms typically generate initial delivery routes for vehicles based on initially known orders. During vehicle operation, new orders are merged with incomplete orders for route replanning according to rolling time slices. However, once a vehicle begins executing a delivery task, it has already established an actual execution status, including its current location, current load, current destination or service target, and tasks en route. Different tasks in the original delivery route are also at different stages, such as completed, in progress, or yet to be executed.

[0004] If the execution status of different tasks is not distinguished during rolling replanning, and all incomplete tasks are uniformly included in vehicle reallocation and access order adjustment, it may alter the current target being executed by the vehicle or the task arrangement in transit. This would cause the regenerated delivery route to be discontinuously connected with the actual execution status of the vehicle, resulting in frequent changes to the recently issued routes, causing significant route disturbances and affecting the stability of the estimated arrival time. Conversely, if all incomplete tasks in the original delivery route are fixed to avoid route disturbances, it is difficult to readjust tasks that have not yet started execution based on new orders. This can easily compress the replanning space for subsequent tasks and reduce the adaptability of vehicle scheduling schemes in the case of dynamic order arrivals. Therefore, the relevant technology struggles to simultaneously ensure the stability of tasks currently being executed and the adjustability of tasks that have not yet been executed.

[0005] Therefore, a dynamic pickup and delivery vehicle route scheduling scheme is needed to ensure that the subsequent delivery routes obtained by rolling replanning are continuously connected with the vehicle's current location, current load, current target, and on-the-way task status. This would protect vehicles that are currently performing tasks while preserving replanning space for tasks that have not yet been performed, thereby reducing the disturbance of new orders to existing execution routes. Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a dynamic pickup and delivery vehicle route scheduling method and apparatus, which solves the problem of how to maintain the continuous execution of the vehicle's current task while preserving the adjustment space for unexecuted tasks when dynamic orders are continuously arriving, thereby reducing the disturbance of rolling replanning to existing delivery routes.

[0007] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a dynamic route scheduling method for pickup and delivery vehicles, comprising: The system acquires dynamically arriving pickup and delivery orders, vehicle operation status, and road network data, and divides the scheduling cycle into multiple rolling time slices. Pickup and delivery orders include pairs of pickup nodes and delivery nodes. Vehicle operation status includes the current target node and task execution status. Task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered. In the first rolling time slice, an initial planning model is constructed based on the initial order, the initial vehicle operating status, and road network data, and the first delivery route is obtained by solving the problem. In subsequent rolling time slices, the updated vehicle operation status is obtained based on the new orders and the delivery routes executed by the vehicles in the previous rolling time slice. Based on the task execution status in the updated vehicle operation status, the vehicle binding nodes are determined from the nodes that have not completed service. The remaining nodes that have not formed a vehicle binding relationship are reassignable nodes. The vehicle binding nodes include the current target node that the vehicle is heading to or serving, the delivery node paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the order but has not yet delivered it. A replanning model is constructed based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is taken as the starting point of the vehicle's replanning path. When the current target node is a pickup node, the delivery node paired with it is kept in the vehicle's replanning path. Reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis. Solve the replanning model to obtain the delivery route for the current rolling time slice, and then re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning process until the scheduling cycle ends.

[0008] Secondly, the present invention provides a dynamic pickup and delivery vehicle route scheduling device, comprising: The data acquisition and time slice division module is used to acquire dynamically arriving pickup and delivery orders, vehicle operation status and road network data, and divide the scheduling cycle into multiple rolling time slices; pickup and delivery orders include pairs of pickup nodes and delivery nodes, vehicle operation status includes the current target node and task execution status, and task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered. The initial planning module is used to build an initial planning model based on the initial orders, the initial vehicle operating status, and road network data in the first rolling time slice, and solve for the first delivery route. The node status determination module is used to obtain the updated vehicle operation status formed by new orders and the delivery routes executed by vehicles in the previous rolling time slice in subsequent rolling time slices. Based on the task execution status included in the updated vehicle operation status, it determines the vehicle-bound nodes from the nodes that have not completed service. The remaining nodes that have not formed a vehicle-bound relationship are reassignable nodes. Vehicle-bound nodes include the current target node that the vehicle is heading to or serving, the delivery node paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the order but has not yet delivered it. The replanning model construction module is used to build a replanning model based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is used as the starting point of the vehicle's replanning path. Delivery nodes of orders that have been picked up but not yet delivered are constrained in the replanning path of the vehicle that performed the corresponding pickup operation. Reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis. The path solving and rolling update module is used to solve the replanning model, obtain the delivery route for the current rolling time slice, and re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning until the scheduling cycle ends.

[0009] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, the pickup and delivery order also includes node demand, node service time and service time window, and the vehicle operation status also includes vehicle location, vehicle load and vehicle availability time; in subsequent rolling time slices, based on the updated vehicle operation status, the node demand of the current target node, the node service time and service time window and the road network data, the expected departure time and expected load of each vehicle when it completes the service of the current target node and leaves are determined, and the expected departure time and expected load are respectively used as the initial time state and initial load state of the corresponding vehicle in the replanning model.

[0010] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, vehicle capacity is also obtained; the optimization objectives and constraints of the initial planning model and the replanning model include: The initial planning model aims to maximize the expected net profit of the initial order; the replanning model aims to maximize the sum of the determined profit and the expected profit increment of the current order to be planned; wherein, the determined profit is the profit obtained by deducting the incurred transportation costs from the revenue of completed orders and orders in transit, the current order to be planned is the order belonging to the reassignable node, and the expected profit increment is the profit obtained by deducting the estimated transportation costs from the expected revenue of the current order to be planned. Both the initial planning model and the replanning model include order selection constraints, pairing pickup and delivery nodes serving in the same vehicle constraints, pickup preceding delivery time constraints, vehicle path flow balancing constraints, service time window constraints, and vehicle capacity constraints based on the vehicle capacity. The vehicle capacity constraints are used to limit the load when a vehicle leaves any access node to not exceed the corresponding vehicle capacity. The replanning model also includes path start constraints for the current target node, initial time state constraints corresponding to the expected departure time, initial load state constraints corresponding to the expected load, and constraints that the vehicle-bound node remains served by the corresponding vehicle.

[0011] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, solving both the initial planning model and the replanning model includes constructing an initial population. The process of constructing the initial population includes: Candidate schedules are decoded into a combination of access node sequence for each vehicle, unallocated order pool, and node status. Node status is used to identify whether a vehicle is bound or reassignable. Specifically, in the first rolling time slice, the pickup and delivery nodes corresponding to the initial order are marked as reassignable. In subsequent rolling time slices, node status is set based on the determination results of vehicle bound nodes and reassignable nodes. In the first rolling time slice, a basic access node sequence is established starting from the initial position of each vehicle. In subsequent rolling time slices, the vehicle access node sequence starting from the current target node and including the corresponding vehicle binding node is used as the basic access node sequence. The unassigned orders are sorted according to their expected delivery time. The sorted unassigned orders are then inserted into the basic access node sequence in pairs of pickup and delivery nodes, satisfying the constraints of pickup before delivery, same-vehicle service, service time window, and vehicle capacity, to generate a seed solution. The seed solution is copied, and only for orders corresponding to reassignable nodes, the pickup and delivery nodes of the same order are removed and re-inserted in pairs until the initial population reaches the population size determined based on the number of orders to be planned based on the current rolling time slice.

[0012] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, solving the initial planning model and the replanning model based on the initial population includes: Using the initial population as the current population, candidate scheduling solutions are selected from the current population as the main parent generation. Destruction operators and repair operators are adaptively selected from the destruction operator pool and repair operator pool, respectively. Orders in the main parent generation that are in a reassignable state are removed, inserted, or reorganized to generate offspring. After feasibility verification and selection, the next generation population is formed. Based on whether the offspring updates the current record of the globally optimal candidate scheduling solution, whether it is better than the parent generation that generated the offspring, and whether it is selected into the next generation population, the scores of the called destruction and repair operators are accumulated, and the selection weights of the corresponding operators are updated based on the accumulated scores and the number of times they are used. The destruction operator pool includes random removal operator, worst-case removal operator, correlation removal operator and path removal operator, and the repair operator pool includes greedy insertion operator, regret value insertion operator and correlation anchor insertion operator.

[0013] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, the insertion process of the correlation anchor insertion operator includes: The in-transit delivery nodes in the vehicle binding nodes are determined as the relevant anchor points of the corresponding vehicles; the in-transit delivery nodes include delivery nodes paired with the pickup node when the current target node is the pickup node, and delivery nodes that have picked up the goods but have not yet delivered the order. For each unassigned order, vehicles with non-empty relevance anchor point sets and insertion positions that satisfy the constraints of pickup before delivery, same-vehicle service, service time window, and vehicle capacity are identified as candidate vehicles. When the candidate vehicle exists, the spatial correlation term and the temporal correlation term between the unassigned order and each correlation anchor point of each candidate vehicle are calculated respectively. The weighted sum of the spatial correlation term and the temporal correlation term is used as the correlation value between the unassigned order and the corresponding correlation anchor point. The minimum value among the correlation values ​​corresponding to the same candidate vehicle is determined as the correlation measure between the unassigned order and the candidate vehicle. Among the candidate vehicles, the unassigned order is preferentially assigned to a candidate vehicle with a lower relevance metric, and the target insertion position is determined from the insertion positions of that candidate vehicle that satisfy the constraints.

[0014] As a further improvement to the above-mentioned dynamic pickup and delivery vehicle route scheduling method or device, the iterative solution based on the initial population includes: Offspring are generated through the elite track and the exploration track respectively; wherein, the main parent generation includes the main parent generation of the elite track and the main parent generation of the exploration track; The elite track determines the best candidate scheduling solution in the current population as the elite track's main parent generation, and performs destruction and repair on the elite track's main parent generation according to the first destruction ratio; The exploration track selects candidate scheduling solutions from the current population through a tournament to serve as the main parent of the exploration track, and performs destruction and repair on the main parent of the exploration track according to a second destruction ratio greater than the first destruction ratio; The first destruction ratio and the second destruction ratio are determined based on the number of orders corresponding to the reassignable nodes in the corresponding parent generation; When the number of consecutive algebras in which the currently recorded global optimal candidate scheduling solution has not been updated reaches the stagnation threshold, the first destruction ratio and the second destruction ratio are increased.

[0015] Thirdly, the present invention provides a storage medium storing a computer program, wherein the computer program causes a computer to execute the dynamic pickup and delivery vehicle route scheduling method described above.

[0016] Fourthly, the present invention provides an electronic device, the electronic device comprising: A processor and a memory; the memory stores program instructions; the processor is configured to run the program instructions to execute the dynamic pickup and delivery vehicle route scheduling method described above.

[0017] (III) Beneficial Effects This invention provides a method and apparatus for dynamic route scheduling of pickup and delivery vehicles. Compared with the prior art, it has the following advantages: This invention divides the scheduling cycle into continuous rolling time slices. An initial delivery route is formed in the first time slice, and subsequent time slices are used to replan based on the updated operating status of vehicles executing the previous delivery route. This allows the planning results to be continuously updated as new orders and actual tasks progress. By distinguishing between vehicle-bound nodes and reassignable nodes based on task execution status, the current target of a vehicle is used as the starting point for the corresponding vehicle's replanning route. When the current target is a pickup node, its paired delivery node is kept within the vehicle's route, preventing interruptions to ongoing tasks or changes to established pickup and delivery relationships. Simultaneously, unbound pickup and delivery nodes are reassigned across vehicles on an order-by-order basis, allowing unexecuted tasks to be adjusted based on new orders and road network status. Therefore, the continuity of recent vehicle task execution can be maintained, frequent changes to delivery routes can be reduced, and the adaptability of dynamic scheduling to order changes can be improved. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a dynamic pickup and delivery vehicle route scheduling method provided in an embodiment of the present invention; Figure 2 A schematic diagram of a rolling time-domain framework provided in an embodiment of the present invention; Figure 3 A flowchart of a hybrid meta-heuristic solution for HGA-ALNS provided in this embodiment of the invention; Figure 4 This is a schematic diagram of a direct path encoding for a candidate scheduling solution provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a random removal operator operation provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a worst-case removal operator operation provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a correlation removal operator operation provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of a path removal operator operation provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of a greedy insertion operator operation provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of a regret value insertion operator operation provided in an embodiment of the present invention; Figure 11 This is a schematic diagram of a correlation anchor insertion operator operation provided in an embodiment of the present invention; Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0020] Component labeling explanation: 100 - electronic device, 101 - memory, 102 - processor, 103 - display. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To address the aforementioned technical problems, this invention provides a dynamic pickup and delivery vehicle route scheduling method and apparatus for dynamically arriving orders. It primarily solves the problem of protecting tasks with established vehicle-to-vehicle relationships and preserving adjustment space for unexecuted orders during continuous delivery operations, ensuring continuous connection between replanned routes and the current vehicle execution state, and reducing the disturbance to existing delivery routes caused by rolling replanning. The main objectives and preferred embodiments of this invention further include: First, it balances dynamic response with path execution stability. The scheduling cycle is divided into multiple rolling time slices, and vehicle-bound nodes and reassignable nodes are determined based on the updated task execution status. The current target node is used as the starting point for the corresponding vehicle's replanning path, ensuring that delivery nodes paired with the current target node (which is a pickup node) and delivery nodes of orders that have been picked up but not yet delivered remain in the corresponding vehicle's path. At the same time, orders that have not formed vehicle-bound relationships are allowed to be redistributed among vehicles in pairs of pickup and delivery nodes.

[0023] Second, achieve continuous inheritance of vehicle operating status and two-stage rolling optimization. Based on the updated vehicle operating status formed by the vehicle executing the delivery path of the previous rolling time slice, determine the expected departure time and expected load when the vehicle completes the service of the current target node and leaves, and use them as the initial time state and initial load state of the replanning path; construct the initial planning model and the replanning model respectively, so that the replanning process can integrate the determined profit with the expected profit increment of the current order to be planned.

[0024] Third, improve the optimization capability and solution efficiency under strong constraints. A hybrid metaheuristic algorithm that integrates genetic algorithm and adaptive large neighborhood search is adopted. Through adaptive destruction-repair, elite track and exploration track, relevant anchor point insertion, and same-vehicle sliding optimization of vehicle-bound delivery nodes, the feasibility of candidate solutions, search diversity and rolling replanning efficiency are improved while maintaining vehicle binding relationship and pick-up and delivery pairing relationship.

[0025] The scheduling scenario described above can be abstracted as a Dynamic Pickup and Delivery Problem with Time Windows (DPDPTW) oriented towards dynamic order arrival. In this problem, each order includes paired pickup and delivery nodes. The pickup node must be visited before its corresponding delivery node, and the paired pickup and delivery nodes must be served by the same vehicle. After pickup, goods cannot be transferred to other vehicles for delivery. In the first rolling time slice, the vehicle departs from the starting station to execute the initial order. In subsequent rolling time slices, newly arrived dynamically ordered items and tasks not yet completed in the previous rolling time slice are included in the replanning process, and the replanned path is continuously expanded from the vehicle's current target node and corresponding operating status. After completing the accepted delivery task, the vehicle returns to the termination station. In an implementation that further considers service time windows and vehicle capacity constraints, by determining the order selection, vehicle allocation, and node access order, the sum of the determined profit and the expected profit increment of the orders to be planned can be maximized.

[0026] In one implementation method for establishing a benchmark model, the following assumptions can be made: 1. The vehicles are of the same type, and the load capacity of each vehicle is known and equal; 2. The travel time, travel distance, and transportation cost between stations and order nodes are fixed values, and the travel distance between nodes satisfies the triangle inequality; 3. The service time of each order node is known and can be set to be equal in the benchmark example; 4. Pick-up and delivery tasks for the same order must be completed by the same vehicle; order splitting and transshipment are prohibited. 5. Vehicles are allowed to arrive at the node and wait before the service window begins; 6. For orders where the pickup service has not yet started and the pickup node has not been determined as the current target node for any vehicle, the corresponding paired pickup and delivery nodes can be reallocated to vehicles on an order-by-order basis.

[0027] The method provided by this invention can be executed by a scheduling server, an edge scheduling device, or other scheduling computing devices with processors and memory. The scheduling computing devices communicate with the order management system, the road network service, and the vehicle terminals, respectively. The order management system is used to send initial orders and dynamically arriving new orders; the vehicle terminals are used to report vehicle identification, vehicle location, vehicle load, current target node, vehicle availability time, and events such as node arrival, service start, service completion, and load changes; the road network service is used to provide travel time, distance, and transportation costs between nodes and between the vehicle's current location and nodes.

[0028] The above baseline assumptions do not constitute a limitation on the scope of application of this invention. In other embodiments, heterogeneous vehicles with different capacities, unit mileage costs, operating speeds, and available time can be used, and vehicle subscripts can be set for the corresponding parameters; travel time and transportation costs can also be updated by the road network service at the beginning of each rolling time slice based on real-time or predicted traffic conditions, and used as planning parameters within the current rolling time slice.

[0029] like Figure 1 As shown, the method described in this embodiment is represented by steps S101 to S105: S101. Obtain dynamically arriving pickup and delivery orders, vehicle operation status, and road network data, and divide the scheduling cycle into multiple rolling time slices; pickup and delivery orders include pairs of pickup nodes and delivery nodes, vehicle operation status includes the current target node and task execution status, and task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered.

[0030] Each pickup and delivery order Including pickup nodes and delivery nodes And may include order revenue. Node demand Node service time Service Hour Window and expected delivery time, among which These represent the upper and lower limits of the service start time, respectively. The demand for pickup nodes is taken as a positive value, while the demand for delivery nodes is taken as a negative value equal to the absolute value of the pickup quantity. Pickup nodes must be accessed before delivery nodes, and both must be served by the same vehicle. Goods cannot be transferred to other vehicles from pickup completion to delivery completion.

[0031] Road network data can be represented by a directed graph G=(V,A), where V includes warehouse nodes, virtual nodes corresponding to the vehicle's current location, and order pickup / delivery nodes, and A is the set of arcs that allow direct vehicle travel. For any arc... Record driving time Driving distance and transportation costs ,in All are node indices. Travel time and transportation cost can be deterministic values, or values ​​predicted by the road network service at the start of the current time slice and remaining constant throughout the time slice.

[0032] like Figure 2 As shown, the scheduling period Divided into P consecutive rolling time slices, with boundary times as follows: , ... The first time slice Execute the initial plan; from Each subsequent time slice begins with a replanning process based on the delivery path execution results of the previous time slice. Time slices can be of equal length or triggered by order events or vehicle status events; the following example uses equal-length time slices.

[0033] S102. In the first rolling time slice, an initial planning model is constructed based on the initial order, the initial vehicle operating status, and road network data to solve for the first delivery route.

[0034] In the first rolling time slice, the scheduling computing device constructs an initial planning model using the initial order, the initial vehicle operating status, and road network data as input. It then calls a preset solver to solve the initial planning model, obtaining the first delivery route, and registers this first delivery route as the delivery route currently being executed by the vehicle. The objective, constraint expression, and solution method of the initial planning model are described in detail in the following implementation method.

[0035] S103. In subsequent rolling time slices, obtain the updated vehicle running status formed by the new orders and the delivery routes executed by the vehicles in the previous rolling time slice, and determine the vehicle binding nodes based on the task execution status in the updated vehicle running status from the nodes that have not completed the service. The remaining nodes that have not formed a vehicle binding relationship are reassignable nodes. The vehicle binding nodes include the current target node that the vehicle is heading to or serving, the delivery node that is paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the goods but has not delivered the order.

[0036] Once the first delivery route or the delivery route of the previous rolling time slice is formed, the vehicle executes tasks according to that route and continuously reports node arrival, service start, service completion, and load change events. Upon entering any subsequent rolling time slice, the scheduling computing device obtains the updated vehicle operating status formed by new orders and the vehicle's execution of the delivery route from the previous rolling time slice, and converts these execution events into updated task execution statuses.

[0037] The task execution status should at least indicate: the node has not yet started service, the node has been identified as the vehicle's current target and the vehicle is heading to the node, the node is receiving service, the node has completed service, the order has been picked up but not yet delivered, and the order has been delivered. The current target node is the only recent node that the vehicle is heading to or serving, and serves as the path connection point for the vehicle in the current replanning.

[0038] At the start of a subsequent time slice, historical nodes that have completed their service and are no longer involved in subsequent paths are first deleted. Then, vehicle-bound nodes and reassignable nodes are determined based on the updated task execution status. For any order, the status transition and processing rules are shown in the table below: In this context, reassignable nodes refer to nodes that have not completed their service and have not formed vehicle binding relationships. This includes paired pickup and delivery nodes corresponding to orders where pickup service has not yet started and the pickup node has not been determined as the current target node for any vehicle. The reassignable status is recorded at the order level, with pickup and delivery nodes for the same order entering the reassignable set together. When removing, inserting, or reorganizing across vehicles, the paired pickup and delivery nodes corresponding to the order are used as the unit to avoid splitting pickup and delivery tasks.

[0039] S104. Construct a replanning model based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is taken as the starting point of the vehicle's replanning path. When the current target node is a pickup node, the delivery node paired with it is kept in the vehicle's replanning path. Reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis.

[0040] The scheduling computing device constructs a replanning model based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In this replanning model, the current target node that the vehicle is heading to or serving is used as the starting point of the vehicle's replanning path; when the current target node is a pickup node, its paired delivery node is kept in the vehicle's replanning path; delivery nodes that have picked up goods but not yet delivered orders are kept in the replanning path of the vehicle performing the corresponding pickup operation; and reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis.

[0041] In this implementation, the "Locked" state indicates that vehicle binding relationships are protected, but this does not mean that the access positions of all vehicle-bound nodes in the same vehicle are absolutely frozen. The current target node is simultaneously subject to vehicle ownership locking and path start position locking; delivery nodes paired with the current target pickup node and delivery nodes that have picked up but not yet delivered orders are subject to vehicle ownership locking. Fine-tuning of the same vehicle position is allowed in the subsequent path of that vehicle, provided that the corresponding vehicle is not changed, the current target node is not crossed, and the pickup / delivery sequence, time window, and capacity constraints are met. The destruction operator, repair operator, and cross-parent mutation must not allow any vehicle-bound node to transfer across vehicles. Therefore, the vehicle binding mechanism does not conflict with the "in-transit delivery node sliding in the same vehicle" local search described later. S105. Solve the replanning model to obtain the delivery route for the current rolling time slice, and then re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning process until the scheduling cycle ends.

[0042] The scheduling computing device invokes a mathematical programming solver, a constraint programming solver, or a metaheuristic solver to solve the replanning model. It determines the target candidate solution from among candidate solutions that satisfy the requirements of the current target path origin, the vehicle service corresponding to the vehicle-bound node, and the pairwise processing of reassignable orders. The target candidate solution is then converted into the delivery path for the current rolling time slice. This implementation is not limited to a specific solution algorithm. Figure 3 The solution process of a hybrid meta-heuristic algorithm (HGA-ALNS) that integrates genetic algorithm and adaptive large neighborhood search is presented.

[0043] The scheduling computing device sends the delivery route of the current rolling time slice to the corresponding vehicle terminal and registers it as the delivery route currently being executed by the vehicle. If the scheduling cycle has not yet ended, the updated vehicle running status formed by the new orders and the vehicle executing the current delivery route is retrieved again in the next rolling time slice, and steps S103~S105 are repeated; if the scheduling cycle has ended, new replanning is stopped, and the vehicle completes the remaining tasks that have been accepted.

[0044] The above steps S101 to S105 constitute a complete implementation process of the method.

[0045] The following sections will describe the specific implementation methods for the state inheritance of the vehicle's expected departure time and expected load, the two-stage planning model, the initial population construction, the adaptive destruction-repair, the insertion of relevant anchor points, and the dual-track evolution mechanism based on this main process. Each implementation method can be combined and applied without technical conflicts.

[0046] To ensure that the replanning route unfolds continuously based on the actual operating status of the vehicles, further data can be obtained, including node demand, node service time, service time window, vehicle location, vehicle load, and vehicle availability time, to determine the expected departure time and expected load. Vehicle availability time represents the earliest point in time when a vehicle can continue performing its delivery task.

[0047] When determining the estimated departure time, calculations are performed based on the vehicle's location, available time, remaining travel time to the current target node, service time and service time window of the current target node, and task execution status. Specifically, when the vehicle is en route to the current target node, the service start time is determined based on the estimated arrival time, with any early arrival counted as waiting time; when the vehicle has already arrived at or is serving the current target node, the estimated departure time is determined based on the waiting status or remaining service time.

[0048] When determining the expected load, based on the vehicle's current load, the node demand of the current target node, and the task execution status, it is determined whether the load change corresponding to the current target node has been included in the vehicle's current load; if not, the vehicle load is updated based on the pickup or delivery operation; if it has been included, the current load reported by the vehicle is used as the expected load.

[0049] The expected departure time and expected load are used as the initial time state and initial load state of the corresponding vehicle in the replanning model, respectively, and the service time and vehicle load of subsequent nodes in the replanning path are recursively deduced.

[0050] For example, at the beginning of the p-th rolling time slice For vehicles Record its current position Current load Vehicle availability time Current target node And the current service phase of the target node, determine the estimated departure time according to the above method. and expected load Among them, the estimated departure time Indicates vehicle Complete the current target node The expected time when the service will leave the node, and the expected payload. Indicates vehicle Complete this node The load when leaving the node after the load increases or decreases.

[0051] Furthermore, the initial planning model for the first rolling time slice and the replanning model for subsequent rolling time slices can be constructed according to the following objective function and constraints.

[0052] First, we introduce the initial planning model for the first rolling time slice.

[0053] The following is a mixed-integer programming expression that enables the implementation of this invention. The formula is used to fully explain the relationships between model variables, and in actual solution, it can be implemented using equivalent linearization, constrained programming, or heuristic feasibility verification.

[0054] make For vehicle assembly, This is the set of pickup nodes for the first time slice. For the corresponding delivery node set, the set of all nodes. 0 and n+1 are the starting and ending warehouse nodes, respectively. Indicates pickup node The corresponding delivery node.

[0055] make Vehicles are 0-1 variables. By node Drive directly to the node Take 1 at the time; For 0-1 variables, the pickup order By vehicle Take 1 when providing service; Indicates vehicle At the node Service start time; Indicates vehicle Complete Node Service and leave node Load at time; For the revenue of pickup order o; and Arc Transportation costs and travel time; For nodes Service time windows, among which Representing nodes respectively Service start time The upper and lower limits; For nodes Service hours; For nodes The node demand is M, where M is a sufficiently large positive number.

[0056] The initial planning model aims to maximize the expected net profit of the initial order. Order revenue can be calculated based on the revenue per unit of demand. With the quantity of goods to be picked up determined, that is Alternatively, it can be directly provided from order data. The objective function is: (1) Each order can be served by a maximum of one vehicle: (2) Pickup and delivery nodes for the same order must be accessed by the same vehicle. Associating the outgoing arcs of both the pickup and corresponding delivery nodes with the order selection variable can be written as: (3) For each visited node, balance the inflow and outflow of vehicles: (4) Each activated vehicle departs from the starting warehouse and returns to the ending warehouse; when some vehicles are not activated, a value of less than or equal to 1 is allowed: (5) The node service start time is transmitted according to the vehicle access arc: (6) The node service start time should be within the time window: (7) Pick-up service must strictly precede the corresponding delivery service. To clarify the activation condition of the priority constraint, this implementation method directly activates the constraint using the order selection variable: (8) The load on a vehicle leaving any access node must not exceed the vehicle's capacity. (9) When the vehicle travels along the arc During driving, nodes The load after service equals the node Post-service load plus node The node demand. This relationship can be expressed by a pair of big-M inequalities: (10) The vehicle departs from the warehouse empty and returns empty after completing the selected orders. (11) Equations (2) to (11) together constitute order selection, same-vehicle service, flow balancing, first-pick-up-then-delivery, service time window, and vehicle capacity constraints. For orders that must be served, the inequality sign in equation (2) can be changed to an equality sign; for orders that are allowed to be rejected or postponed, the unselected orders enter the unallocated order pool and can continue to participate in planning in subsequent time slices.

[0057] Next, we will introduce the replanning model for subsequent rolling time slices.

[0058] At the beginning of subsequent rolling time slices, such as the p-th rolling time slice. First, based on the updated vehicle operating status formed by the vehicle executing the delivery route of the previous time slice, a current dataset to be planned is established. Let... Bind a set of delivery nodes to the vehicles corresponding to orders in transit; orders in transit include orders where the pickup node has completed service but the delivery node has not, as well as orders where the pickup node has become the current destination and the vehicle is heading to or serving that pickup node. For the set of pickup nodes for reassignable orders, and The corresponding set of delivery nodes Bind a set of delivery nodes to vehicles other than the current target node that serves as the starting point of the path.

[0059] For vehicles ,make For the current target node, For vehicles The vehicle is bound to a set of delivery nodes. For vehicles The set of accessible nodes in the current time slice. For the reason The nodes in the network constitute a set of feasible arcs that satisfy the road network connectivity conditions. The current target node of other vehicles and the delivery node bound to the vehicle are not considered part of the vehicle's network. The set of accessible nodes.

[0060] make and These are the estimated departure time and estimated load determined based on the state, and are used as the vehicle's... The initial time state and initial load state of the replanned path.

[0061] make This is the profit already determined before the start of the current time slice, which equals the revenue from completed orders and orders in transit minus the profit as of [date missing]. Transportation costs already incurred. Revenue from orders in transit is recorded as a definite item for which a carrier binding relationship has been established. The current orders to be planned belong to reassignable nodes, and their expected revenue minus the estimated transportation costs based on the current vehicle status constitutes the expected profit increment. Because... In the current replanning, it is a constant, which does not change the comparison of the merits between candidate paths, but can maintain the continuity of the total profit statistics for each time slice.

[0062] The objective function of the reprogramming model is: (12) For reassignable orders, at most one vehicle can be selected per order: (13) For a vehicle bound to a delivery node, both the inflow and outflow arcs of that node are restricted to the feasible arc set of the vehicle; since the accessible node set of other vehicles does not contain this node, other vehicles cannot access it. The flow constraint for a vehicle bound to a delivery node is: (14) For reassignable orders, the paired pickup and delivery nodes must be served by the same vehicle: (15) Set flow balancing for all visited nodes except the vehicle's current destination and destination warehouses: (16) The vehicle departs from its current target node and returns to the final warehouse: (17) The current target node will no longer have an entry arc set, and the terminated repository will no longer have a departure arc set. (18) For non-originating access arcs, service time is passed in the order of travel: (19) The service start time of the accessed node is located within the corresponding service time window: (20) For orders currently awaiting planning, the principle of picking up goods before delivery will continue: (twenty one) The service completion time of the vehicle's current target node coincides with its estimated departure time. Current target Being driven by vehicles During service, The actual service start time can be taken; when the vehicle is en route, Take the estimated service start time after considering the waiting time window: (twenty two) The vehicle can only reach the next node in the replanned path after it is expected to leave the current target node: (twenty three) The departure load at any node on the replanning path must not exceed the vehicle capacity. (twenty four) Load is transferred along the access arc: (25) The departure load of the current target node is equal to the predicted load calculated from the vehicle's operating status: (26) When the vehicle completes its accepted task and returns to the termination warehouse in this time slice, its load will be 0. (27) Equations (13) to (27) add corresponding vehicle service constraints for vehicle-bound nodes, path start constraints for the current target node, and initial state constraints for expected departure time and expected load to the basic constraints of the initial model. If the current target node is a pickup node, then equations (17), (18), (22), and (26) fix the pickup node as the path start, and its paired delivery node is constrained to be in the same vehicle path by equation (14); if the current target node is a delivery node, then the delivery node itself is the path start, and its original vehicle affiliation is fixed accordingly. This covers all types of vehicle-bound nodes listed above.

[0063] Equation (14) locks the vehicle affiliation. For vehicles that are not the starting point of the current target and are bound to delivery nodes, the solver can determine the position that satisfies the constraints in the subsequent access sequence of the same vehicle; the local search described later can also slide forward or backward without changing the vehicle affiliation. The current target node is fixed as the first node by Equations (17) and (18) and does not participate in this sliding.

[0064] As a solution preparation method for the initial planning model and the replanning model, direct path encoding and greedy insertion can be used to construct the initial population. When constructing the initial population, a preset perturbation method that satisfies the requirements of pairwise removal and re-insertion can be used, and its specific implementation process is as follows: like Figure 4 As shown, direct path encoding is used to represent candidate scheduling solutions. ,in, For vehicles The sequence of accessed nodes, For the unassigned order pool, This is the node state vector. The node state includes at least the vehicle-bound state (Locked) and the reassignable state (Activated). All orders are in the Activated state. For vehicle-bound nodes, the code also records their corresponding vehicle identifier; for the current target node, it also records its corresponding... The starting point. Pickup nodes in the code can be marked with "+", and delivery nodes can be marked with "-". Figure 4 This shows an instance of three vehicle path segments and an unassigned pool.

[0065] In the first rolling time slice, the initial positions of each vehicle or warehouse node constitute an empty path base, and the paired pick-up and delivery nodes of all initial orders are marked as reassignable. In subsequent rolling time slices, each... Starting from the current target node, and retaining the nodes belonging to the vehicle... Other vehicles are linked to delivery nodes; historically completed nodes are removed from the coding, and old orders and new orders that have not yet started execution are entered into the new nodes. The initial basis for the seed solution can first maintain the relative order of the vehicle-bound delivery nodes in the previous delivery path, and then allow feasible sliding within the same vehicle.

[0066] Will Orders are sorted in ascending order of expected delivery time. The expected delivery time can be the promised delivery time of the order; if no promised time is given separately, the center time or upper bound of the delivery node's time window can be used, and the same definition is applied uniformly within the same time slice. The valid insertion positions of the order's pickup and delivery nodes in each vehicle path are enumerated sequentially, validating pickup before delivery, same-vehicle service, service time windows, and capacity constraints. The insertion position is selected based on the nearest feasible vehicle priority, minimum additional driving cost, or maximum marginal profit. Orders without valid positions are retained in [the following time slice]. .

[0067] The first seed solution is obtained by the greedy insertion above. and used it as the initial population One of the individuals. Copy. A preset perturbation method is used for replicating individuals, and pairwise removal and re-insertion are performed only on orders in a redistributable state, repeating until the initial population reaches a certain size. .

[0068] One method for determining the number of disturbances is to select the disruption ratio from the range [0.1, 0.3]. and order ,in To round up, The number of orders that can be reassigned and planned at present. Set a removal limit for each order. This ensures that the damage scale varies with the number of orders while avoiding exceeding the preset removal limit. Alternatively, other uniform rounding rules and ratio ranges can be used based on the real-time budget calculation.

[0069] The above calculation of the number of disturbances is only applicable to Execute at that time. At that time, there are no realizable orders that can be removed by the destruction operator, so the initial population perturbation and subsequent ALNS iterations are skipped, and the feasibility of each vehicle path composed of vehicle-bound nodes is directly checked and output; if the same vehicle has multiple non-starting vehicle-bound delivery nodes, the same-vehicle sliding optimization that does not change the vehicle affiliation can be performed only.

[0070] Furthermore, the preset perturbation method can be implemented by selecting a destruction operator and a repair operator. That is, based on the initial population mentioned above, offspring can be generated by adaptively selecting the destruction operator and the repair operator, and the next generation population can be formed by combining feasibility verification, local search, sorting and selection.

[0071] Each time a child node is generated, an operator is selected from the destruction operator pool and the repair operator pool by roulette wheel based on the current selection weight. All destruction and repair operations first filter the node status, operate only on reassignable orders, and group pickup and delivery nodes of the same order together; vehicle-bound nodes cannot be stripped or reassembled across vehicles.

[0072] The pool of destruction operators includes random removal operators, worst-case removal operators, correlation removal operators, and path removal operators, as detailed below: Random removal operator such as Figure 5 As shown, orders are drawn with equal probability from the set of reassignable orders already allocated in the current solution. For each order, remove its paired pickup and delivery nodes from the corresponding path and place them in... This operator does not depend on the current cost ranking and is used to maintain search diversity.

[0073] Worst removal operator such as Figure 6 As shown. For each reassignable order Calculate the path cost for retaining the order. and the path cost after deleting the paired nodes of the order. , the difference As a marginal cost, priority should be given to removing items with higher marginal costs. One order.

[0074] Correlation removal operators such as Figure 7 As shown. For each pair of reassignable orders, the spatiotemporal correlation is pre-calculated and a lookup table is built. Let the orders... and orders The pickup nodes are respectively and The correlation between the two orders It can be calculated according to formula (28): (28) In equation (28), the first term is the sum of the distance between the pickup node and the delivery node of the two orders; the second term is the smaller of the two connecting distances between completing one order and then picking up the other order; and the third term is the difference between the start times of the pickup and delivery time windows. The smaller the value, the higher the spatiotemporal correlation. An example is... During the operation, a seed order is randomly selected, and the removal set is expanded from high to low relevance until the order is removed. One order.

[0075] Path removal operators such as Figure 8As shown, a path is randomly selected from the vehicle paths containing reassignable orders, and consecutive segments to be visited are selected; if the segment contains only one node of a certain order, the other paired node of that order is removed to ensure that the final destination is reached. This is a complete order. The vehicle binding node is skipped, and the fragment removal process is not initiated.

[0076] The repair operator pool includes greedy insertion operators, regret value insertion operators, and correlation anchor insertion operators, as detailed below: Greedy insertion operator such as Figure 9 As shown, enumeration For each order, calculate the marginal profit increment at each valid insertion position in each vehicle path, and select the "order-vehicle-pickup location-delivery location" combination with the largest marginal profit increment to perform the insertion until the predetermined number of insertions is reached or no valid position exists.

[0077] Regret value insertion operator such as Figure 10 As shown. Let Indicates order Insert the first The incremental marginal profit when in the optimal legal position This indicates the optimal position. Under profit maximization evaluation, orders... Regret value is defined as the non-negative profit loss caused by abandoning the optimal position: (29) Equation (29) uses the optimal marginal profit minus the first The optimal marginal profit allows the regret value to represent the non-negative profit loss incurred by abandoning the optimal position. If the order... have There are valid insertion positions and Then it can be made Where min represents the minimum value operator; when there is only one valid insertion position, the order degenerates into a greedy insertion, and the missing first value is not calculated. Optimal location. Algorithm selection. For the largest order, insert it into the optimal legal position of that order, and repeat until the predetermined number of insertions is completed.

[0078] Relevance anchor insertion operator such as Figure 11 As shown, this feature is only enabled when at least one vehicle has an order in transit linked to a delivery node. For vehicles The set of relevance anchor points, whose members include delivery nodes paired with the current target pickup node and vehicles. Delivery points for orders that have been picked up but not yet delivered. For unassigned orders. , only Vehicles that are not empty and have paired insertion positions that satisfy the constraints of pickup before delivery, same-vehicle service, time window, and capacity. Listed as candidate vehicles.

[0079] Assuming an order The pickup node is , As a relevance anchor point, then the order With vehicles The correlation measure is: (30) Equation (30) for orders With vehicles For each delivery anchor point in transit, spatial and temporal dependencies are calculated separately, and the minimum value among the weighted sums is taken. These are the spatial and temporal weighting coefficients, respectively. The smaller the value, the more relevant the order is to the vehicle's current on-the-go tasks. An example is... After selecting candidate vehicles with low relevance, it is still necessary to enumerate the legal pairwise insertion positions within each vehicle and select the position with the best marginal profit; all If all values ​​are empty, use greedy insertion or regret value insertion instead.

[0080] After constructing the initial population, the initial programming model and the reprogramming model are solved based on the initial population. Adaptive updates of operator selection weights, local search, feasibility verification, and next-generation population selection are performed sequentially, including: During algorithm initialization, the selection weights of each destruction and repair operator can be set to 1.0. The evolution process is divided into multiple statistical segments, each segment having a length of... Based on the number of offspring in each generation Determine, for example Within each segment, record the operators. Number of times of use and cumulative score .

[0081] One scoring rule is: if the offspring refreshes the globally optimal solution of the current record, record... If the global optimum is not refreshed but the result is better than the parent generation that generated the child generation, then record... If a generation is not superior to its parent generation but enters the next generation after merging and sorting, it is recorded as... No points are awarded for other cases. The destruction and repair operators corresponding to the same offspring receive points for the same event type.

[0082] After the statistical segment is completed, update the operator selection weights according to the following formula: (31) In equation (31), These are the selection weights of operator m in the t-th and t+1-th statistical segments, respectively; For reaction rate parameters, .like Then keep To avoid division by zero. After the update is complete, reset the current statistical segment. and The probability of choosing the next segment of the roulette wheel is determined based on the updated weights.

[0083] After each new path is generated but before fitness evaluation, vehicle-bound delivery nodes that are not the current target starting point can undergo same-vehicle sliding optimization. These vehicle-bound delivery nodes include paired delivery nodes of the current target pickup node and delivery nodes of orders that have been picked up but not yet delivered. Specifically, these vehicle-bound delivery nodes are extracted from each vehicle path, keeping their vehicle affiliation unchanged, and attempts are made to re-insert them forward or backward along the current sequence. If the new position does not exceed the current target starting point, satisfies the temporal relationship, time window constraints, and capacity constraints of the current target pickup node or the pickup node that has completed service preceding the corresponding delivery node, and shortens the travel distance or reduces travel and waiting costs, then the new position is adopted. This operation is used to fix the problem of vehicle-bound delivery nodes being squeezed into non-optimal positions when inserting new orders.

[0084] When the newly generated offspring is better than the currently recorded global optimal solution, further fine-tuning of adjacent nodes in the same vehicle can be performed. Adjacent nodes along the same path are swapped sequentially, excluding cases where pickup and delivery nodes belonging to the same order are directly swapped. It is verified that after the swap, pickup still precedes delivery, the distance is shortened, and both capacity and time windows are feasible. Alternatively, the above adjacent swapping can be performed on all new individuals, but a maximum number of checks should be set to control real-time solution time.

[0085] Feasibility verification includes at least the following: each accepted order's pickup and delivery nodes are accessed by exactly the same vehicle; pickup service precedes delivery service; the vehicle-bound node belongs to the corresponding vehicle in the record; the current target node is the starting point of the corresponding path; the service time of each node meets the time window; the load recursively along the path is between 0 and the vehicle capacity; and the path arc is valid in the road network. Offspring that fail the verification are repaired or discarded.

[0086] The current parent population and offspring pool are merged to form a candidate pool, which is then sorted in multiple levels according to the objective function value, the number of unassigned orders, and the total cost of vehicle travel and waiting. To eliminate logically duplicate individuals generated by the Adaptive Large Neighborhood Search (ALNS), the objective function value is rounded to four decimal places and combined with the number of unassigned orders to form a two-dimensional feature vector. Candidate scheduling solutions with the same feature vector are retained and ranked higher.

[0087] From the deduplicated candidate pool, take approximately 0.9 The top-ranked individuals are selected, and approximately 0.1 individuals are randomly drawn without replacement from the remaining individuals. Individuals were selected to balance the preservation of elites with diversity. The insufficient number of individuals was due to deduplication. When copying, a deep copy of the currently recorded globally best individual is used to complete the data; during copying, its temporary operator scoring and cache are reset to avoid different individuals sharing variable data. Population size.

[0088] Furthermore, based on the aforementioned iterative solution, a offspring generation strategy based on dual-track evolution is introduced. This involves selecting the primary parent generation through both the elite track and the exploration track, and generating offspring with different destruction ratios to balance local development and global exploration. Secondary parent generation and path replacement mutation are further optional implementation methods.

[0089] exist Figure 3 The generation stage of the overall solution process shown below, each generation generates... The population consists of several generations, with the top 40% allocated to the elite track and the remaining 60% to the exploration track. The ratio can be adjusted as needed. The elite track uses the individual ranked first in the current population by multi-level indicators as the primary parent, and selects a secondary parent through a five-person tournament. The exploration track selects both the primary and secondary parents through five-person tournaments.

[0090] ALNS corruption—repair is applied only to the primary parent of each track. During normal search, elite tracks are corrupted according to the first corruption ratio. Reconfigurable orders in the parent generation on a smaller scale. It can be determined within the range of 10% to 35%; the exploration trajectory is based on greater than Second proportion of damage Perform large-scale reconstruction. It can be determined within the range of 20% to 45%. For The current number of orders to be planned, and the actual number of orders to be removed for the Elite and Exploration tracks, can be determined by the following formulas: , (32) Each time offspring are generated, select the one that satisfies The specific value depends on the actual number of orders removed. Limitations: When the order size is small or the same removal limit is reached, and They can be the same, but the setting relationship that the second damage ratio is greater than the first damage ratio remains unchanged. Skip the damage-repair process as described above.

[0091] The secondary parent is used for optional path substitution mutations, rather than directly participating in unconstrained node crossings. It operates with a preset probability. When mutation is triggered, a complete feasible vehicle path or path segment consisting only of reassignable orders is selected from the initial offspring generated by ALNS. Using the secondary parent as the base, reassignable nodes that conflict with the selected path or segment are deleted. The selected path or segment is then placed into the corresponding vehicle segment, and the squeezed-out reassignable orders are repositioned using greedy insertion. During this process, the current target origin and vehicle-bound nodes of the secondary parent are preserved, and the vehicle affiliation of the vehicle-bound nodes is not changed. An example setup is provided. .

[0092] When the number of consecutive algebras in which the currently recorded globally optimal candidate scheduling solution has not been updated reaches the stagnation threshold. At that time, a large-scale anti-stagnation reconstruction is triggered, increasing the first destruction ratio of the elite track to 40% to 60% and the second destruction ratio of the exploration track to 50% to 70%, while maintaining a second destruction ratio greater than the first destruction ratio each time. Stagnation threshold. Can be calculated based on the maximum stagnation tolerance generation. Sure: (33) Increasing the destruction ratio only expands the removal scope of reassignable orders; vehicle-bound nodes remain protected by state filtering. Once the global optimum is updated, the destruction ratio can be restored to normal, or it can be gradually restored after the current statistical segment ends.

[0093] To illustrate the complete iterative relationship of the HGA-ALNS algorithm under the different combinations of technical features listed above, the main loop of HGA-ALNS is given below. This main loop is used to show the execution order after the corresponding implementation combinations: Step 1, Input the initial population The maximum number of generations in the current time slice Maximum tolerance algebra Population size Offspring size Weight update interval .

[0094] Step 2, Initialize the evolutionary algebra Patience value Statistical Fragment Counter And set the weight of each operator to 1.0.

[0095] Step 3, Calculate the large-scale damage threshold .

[0096] Step 4, when and Perform the following steps: Step 5, for The optimal candidate scheduling solutions for this generation are recorded by sorting the solutions at multiple levels according to the objective function value, the number of unassigned orders, and the travel and waiting costs. ;when At that time, with Initialize the historical global optimal solution .

[0097] Step 6: Invoke the above-mentioned offspring generation strategy based on dual-track evolution to generate offspring. Each child generation performs state protection, feasibility verification, and local search, and records the destruction operators, repair operators, and main parent generation called when generating each child generation.

[0098] Step 7, merge Combined with the offspring pool, multi-level sorting, two-dimensional feature deduplication, and selection of 90% elites plus 10% randomness are performed to obtain... After the next generation selection result is determined, the selection is based on whether the offspring is superior to the historical global optimal solution before generation. Whether it is superior to the corresponding parent generation and whether it enters The mutual exclusion priority is used to accumulate the score and usage count of the corresponding operators.

[0099] Step 8, let ;when achieve At that time, the weights of the destruction and repair operators are updated according to formula (31) based on the scores and usage times of each operator, and the weights of each operator are reset. and and order .

[0100] Step 9, record The optimal candidate scheduling solution ;like Better than the historical global optimal solution recorded before the generation of the current generation and its offspring. Then renew And order Otherwise, .

[0101] Step 10, let Return to step 4.

[0102] Step 11: Output the optimal candidate scheduling solution for the current time slice. .

[0103] The following describes the continuous execution process of this method using an 8-hour retail logistics delivery cycle. The scheduling cycle runs from 8:00 to 16:00, with a rolling time slice length of... =30 minutes, a total of 5 identical vehicles are deployed, each vehicle has a capacity of =100. Each order has one pickup node and one delivery node, and records the demand, service time, time window, revenue, and node coordinates; the road network data provides the definite travel time and transportation cost between nodes.

[0104] Steps S101~S102: 8:00 First rolling time slice.

[0105] At 8:00, the system received 30 initial orders, including pickup nodes P1 to P30 and corresponding delivery nodes D1 to D30. All 5 vehicles were located in the warehouse and were empty, having not yet performed any tasks. Therefore, no vehicle binding relationship was formed among the 60 order nodes, and all 30 orders were processed as reassignable orders.

[0106] The scheduling computing device loads the initial planning model shown in equations (1) to (11) to calculate the expected net profit of the initial orders. The goal is to maximize efficiency. Each vehicle starts from warehouse node 0. The model verifies that orders can be selected at most once, pickup and delivery nodes are in the same vehicle, pickup is first and delivery is later, service time windows, flow balancing, and vehicle capacity constraints are considered. Orders that cannot be scheduled in the current time slice are allowed to remain in the unallocated pool.

[0107] The total number of orders not yet completed =30, the current number of reassignable orders to be planned. =30. The population size is obtained by setting the parameters as described above and rounding them up. =20, Offspring Size =12, Maximum number of generations =167, Maximum Stasis Tolerance Algebra =42, ALNS order removal limit =8. Anti-stagnation threshold =34 and operator weight update interval =4.

[0108] In the initial population building phase, 30 orders are first sorted according to their expected delivery time, and a seed solution is formed using the nearest feasible vehicle priority rule. This seed solution is then replicated and subjected to ALNS perturbations involving only reallocatable order pairs, resulting in 20 feasible individuals. After generational evolution begins, the first 40% of offspring are generated via the elite track: the individual currently ranked first in the population serves as the primary parent, subject to a 10% to 35% small-scale ALNS reconstruction; the remaining 60% of offspring are generated via the exploration track: the primary parent is generated through a five-person tournament, subject to a 20% to 45% larger-scale ALNS reconstruction. Secondary parents for each track are generated through a tournament and are only used for path replacement mutations triggered with a 5% probability.

[0109] For each offspring generation, a feasibility check and a sliding motion of the same vehicle to delivery nodes in transit are performed. Since there are no vehicles bound to nodes in transit at this time slice, the sliding motion does not change the path. When the offspring generation refreshes the global optimum, fine-tuning of adjacent nodes in the same vehicle is then performed. For each generation, 12 offspring generations are merged with 20 parent generations, sorted by objective function value, number of unassigned orders, and travel and waiting costs. Duplicates are removed using four decimal places of the objective function value and the number of unassigned orders. Then, 90% elite retention and 10% random sampling are used to form the next generation of 20 individuals.

[0110] The algorithm terminates after 167 generations or 42 consecutive generations without updating the optimal solution for the current time slice. In one example runtime environment, the first time slice calculation takes approximately 3 seconds, outputting the first delivery route for 5 vehicles and sending it to the vehicle terminal. Based on this, the vehicle terminal displays or executes the access node sequence and continuously reports vehicle location, load, current target, and task service events.

[0111] Steps S103~S105: 8:30 Subsequent rolling time slices.

[0112] At 8:30, the system retrieves 8 new orders (orders 31 to 38) added between 8:00 and 8:30, and obtains the updated running status of the vehicles as they execute their first delivery routes. For example, vehicle 1 is heading to pickup node P15, a task already determined before 8:30; vehicle 2 has already loaded the goods for orders 5 and 8 and still needs to access delivery nodes D5 and D8; the five vehicles collectively completed services at 12 historical nodes between 8:00 and 8:30.

[0113] The system removes the aforementioned 12 completed service nodes from the set of nodes to be planned. For vehicle 1, the current target pickup node P15 is used as the starting point for the replanning path of vehicle 1, and its paired delivery node D15 is kept in the replanning path of vehicle 1; for vehicle 2, D5 and D8 are kept in the replanning path of vehicle 2. Thus, P15, D15, D5, and D8 are 4 vehicle-bound nodes, corresponding to 3 orders in transit. P15 is also locked by the path starting position, and D15, D5, and D8 are locked by vehicle ownership.

[0114] There were 15 old orders that hadn't started pickup in the previous time slot. Together with 8 new orders, this creates a total of 23 reassignable orders, or 46 reassignable pickup / delivery nodes. The "23" here refers to the number of orders, not the number of nodes; including the 3 orders already in transit with vehicle bindings, this represents the total number of incomplete orders. =26, the current number of reassignable orders to be planned. =23.

[0115] The scheduling calculation equipment calculates the estimated departure time when vehicle 1 completes pickup at P15 and leaves, based on the remaining travel time from vehicle 1's current location to P15, the time window of P15, and the service time. The current load is then added to the P15 requirement to obtain the projected load. Vehicle 2 and other vehicles perform the same calculations based on their current target, current load, and available time. The expected departure time and expected load are used as the initial states for equations (22), (23), and (26), respectively.

[0116] The system loads the reprogramming models shown in equations (12) to (27) with the determined profits. The objective is to maximize the sum of the expected profit increments from the 23 currently planned orders. Revenue from completed orders and orders in transit, including P15, Order 5, and Order 8, is included, while transportation costs incurred between 8:00 and 8:30 are deducted. The replanning model no longer starts from the warehouse but instead arranges subsequent routes starting from the expected departure state of each vehicle's current target node.

[0117] Press at this time =26、 Calculating with =23, we still get... =20、 =12、 =167、 =42、 =8、 =34 and =4. The seed solution is based on the inheritance path formed by the current target of each vehicle and its bound nodes. The 46 nodes of the 23 reassignable orders are inserted in pairs according to the expected delivery time. When the seed solution is copied to generate the remaining individuals, only the 23 reassignable orders are allowed to be destroyed and repaired. P15 must not be changed from its starting position, nor can D15, D5 or D8 be transferred to other vehicles.

[0118] During the iteration process, the relevance anchor insertion operator uses D15, D5 and D8 as the spatiotemporal anchors of the corresponding vehicles. For example, the relevance measure of the order to be inserted with vehicle 2 is calculated according to equation (30) between D5 and D8. If the order is more convenient to the task on the way of vehicle 2 and there is a valid insertion position, it is first placed in vehicle 2. Then, the specific location of the order's pickup node and delivery node is determined by the greedy logic.

[0119] After ALNS reconstruction is complete, same-vehicle sliding is performed on D15, D5, and D8. If reinsertion forward or backward can shorten the path while still satisfying capacity and time window requirements, the new position is adopted; this step does not change the corresponding vehicle and therefore does not violate vehicle binding rules. Mutations only swap path segments consisting of reassignable orders. Parent-child merging, deduplication, multi-level sorting, and next-generation selection are the same as in the first time slice.

[0120] After the algorithm reaches the termination condition, it generates the delivery route for that time slice in approximately 4 seconds under a sample running environment. The resulting route is continuously expanded from the actual current destination and expected departure status of each vehicle. While preserving the vehicle binding relationships of P15, D15, D5, and D8, the allocation and access locations of 23 vehicles that have not yet started executing orders are re-optimized.

[0121] Step S105: Full scheduling cycle loop After the 8:30 plan is issued and executed, the system repeats the above subsequent time slice processing every 30 minutes. At the replanning times of 9:00, 9:30, and 10:00, the system sequentially obtains the updated running status of new orders and vehicles following the paths of the previous time slice, deletes recently completed nodes, determines the new current target node, vehicle-bound delivery nodes, and reassignable orders, calculates the estimated departure time and estimated load of each vehicle, and constructs and solves the replanning model.

[0122] For each rolling time slice, newly captured orders and old orders that have not yet started pickup enter the reassignable order pool together; pickup nodes that have become the current target and their paired delivery nodes, as well as delivery nodes that have picked up goods but have not yet delivered them, continue to be protected by vehicle binding. The new path obtained by the solution is registered as the current execution path and sent to the corresponding vehicle terminal. The subsequent execution results of the vehicle terminal then form the input for the next time slice.

[0123] This process continues until the start of the last time slice at 15:30. At 15:30, the remaining unpicked orders, the orders added during the last period, and the remaining vehicle binding tasks for each vehicle are combined and planned to form the final path required to complete the accepted tasks from 15:30 to 16:00. This guides the vehicles to complete the follow-up services and return to the termination warehouse, thus ending the scheduling cycle.

[0124] Based on the same inventive concept, the present invention also provides a dynamic pickup and delivery vehicle route scheduling device, comprising: The data acquisition and time slice division module is used to acquire dynamically arriving pickup and delivery orders, vehicle operation status and road network data, and divide the scheduling cycle into multiple rolling time slices; pickup and delivery orders include pairs of pickup nodes and delivery nodes, vehicle operation status includes the current target node and task execution status, and task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered. The initial planning module is used to build an initial planning model based on the initial orders, the initial vehicle operating status, and road network data in the first rolling time slice, and solve for the first delivery route. The node status determination module is used to obtain the updated vehicle operation status formed by new orders and the delivery routes executed by vehicles in the previous rolling time slice in subsequent rolling time slices. Based on the task execution status included in the updated vehicle operation status, it determines the vehicle-bound nodes from the nodes that have not completed service. The remaining nodes that have not formed a vehicle-bound relationship are reassignable nodes. Vehicle-bound nodes include the current target node that the vehicle is heading to or serving, the delivery node paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the order but has not yet delivered it. The replanning model construction module is used to build a replanning model based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is used as the starting point of the vehicle's replanning path. Delivery nodes of orders that have been picked up but not yet delivered are constrained in the replanning path of the vehicle that performed the corresponding pickup operation. Reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis. The path solving and rolling update module is used to solve the replanning model, obtain the delivery route for the current rolling time slice, and re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning until the scheduling cycle ends.

[0125] The modules described above are interconnected according to the data flow of "data and time slice - first slice path or update status - node classification - replanning constraints - path solving and feedback", which correspond one-to-one with the steps listed in the dynamic pickup and delivery vehicle route scheduling method provided in any embodiment of the present invention. The module division is only a functional division; multiple modules can be implemented by the same program process, and a single module can also be implemented collaboratively by multiple processors or service processes.

[0126] Based on the same inventive concept, embodiments of the present invention also provide a storage medium storing a computer program, wherein the computer program causes a computer to execute the dynamic pickup and delivery vehicle route scheduling method provided in any embodiment of the present invention.

[0127] In embodiments of the present invention, any combination of one or more storage media may be used. The storage medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0128] Based on the same inventive concept, embodiments of the present invention also provide an electronic device. Figure 12 The diagram shown illustrates the structure of the electronic device 100 provided in an embodiment of the present invention. In some embodiments, the electronic device may be a mobile phone, tablet computer, wearable device, in-vehicle device, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), or other terminal device. Furthermore, the dynamic pickup and delivery vehicle route scheduling method provided in this embodiment can also be applied to databases, servers, and service response systems based on terminal artificial intelligence. This embodiment does not impose any limitations on the specific application scenarios of the dynamic pickup and delivery vehicle route scheduling method.

[0129] like Figure 12 As shown, the electronic device 100 provided in this embodiment of the invention includes a memory 101 and a processor 102.

[0130] The memory 101 is used to store computer programs; preferably, the memory 101 includes various media that can store program code, such as ROM, RAM, magnetic disk, USB flash drive, memory card or optical disk.

[0131] Specifically, memory 101 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. Electronic device 100 may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 101 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0132] The processor 102 is connected to the memory 101 and is used to execute the computer program stored in the memory 101 so that the electronic device 100 executes the dynamic pickup and delivery vehicle route scheduling method provided in any embodiment of the present invention.

[0133] In an optional implementation, the processor 102 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, evaluation index gate or transistor logic devices, or evaluation index hardware components.

[0134] In an optional embodiment of the present invention, the electronic device 100 may further include a display 103. The display 103 is communicatively connected to the memory 101 and the processor 102, and is used to display the relevant GUI interactive interface of the dynamic pickup and delivery vehicle route scheduling method.

[0135] It is understood that the dynamic pickup and delivery vehicle route scheduling device, storage medium and electronic device provided in the embodiments of the present invention correspond to the dynamic pickup and delivery vehicle route scheduling method provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the method, and will not be repeated here.

[0136] In summary, compared with existing technologies, it has the following beneficial effects: 1. A rolling replanning mechanism combining vehicle binding and reassignable states: In subsequent rolling time slices, vehicle-bound nodes and reassignable nodes are distinguished based on the updated task execution status. The current target node that the vehicle is heading to or serving is used as the starting point for the corresponding vehicle's replanning path. Delivery nodes paired with the current target as a pickup node, as well as delivery nodes for orders that have been picked up but not yet delivered, are kept in the corresponding vehicle's path. For orders that have not formed a vehicle-bound relationship, reassignment is allowed on a pair basis, with pickup and delivery nodes as units. This allows the replanning path to be continuously connected with the vehicle's current task, avoiding cross-vehicle transfers of goods, while retaining adjustment space for unexecuted orders and reducing the disturbance of rolling replanning to existing delivery routes.

[0137] 2. A two-stage planning model that connects initial planning with dynamic replanning: An initial planning model and a replanning model are constructed for the first rolling time slice and subsequent rolling time slices, respectively. The initial planning model uses the expected net profit of the initial order as the optimization objective; the replanning model uses the sum of the determined profit and the expected profit increment of the current order to be planned as the optimization objective, and sets the path start constraint of the current target node, the initial state constraint corresponding to the expected departure time and expected load, and the original vehicle service constraint of the vehicle-bound node. This can maintain the continuity of profit statistics across different periods while allowing subsequent paths to be developed based on the actual operating state of the vehicles, thus improving the continuity of vehicle paths throughout the entire scheduling cycle.

[0138] 3. Dynamic adjustment mechanism for control parameters based on the size of the orders to be planned: Based on the total number of incomplete orders and the number of orders available for reallocation in the current time slice, the population size, offspring size, maximum number of generations, maximum stagnation tolerance generations, order removal upper limit, and operator weight update interval are dynamically determined, and the range of parameter changes is controlled by upper and lower bound constraints. This allows the algorithm's search scale and computational budget to vary with order size and order density, reducing search inadequacy or computational redundancy caused by fixed parameters in different dynamic scenarios, and improving the adaptability of rolling replanning.

[0139] 4. Relevance anchor insertion operator that integrates spatiotemporal correlation: By binding delivery nodes to vehicles in the vehicle path as relevance anchors, spatial and temporal relevance terms are calculated between unassigned orders and each candidate vehicle's relevance anchor, and the minimum weighted sum is used as the relevance metric between the order and the candidate vehicles. When a vehicle has a valid insertion position that satisfies pickup and delivery sequence, same-vehicle service, time window, and capacity constraints, vehicles with lower relevance metrics are prioritized. This allows for guiding order insertion using the spatiotemporal distribution of existing vehicle tasks, reducing driving detours and timing conflicts caused by blind allocation, and increasing the probability of generating feasible candidate paths during the repair process.

[0140] 5. A hybrid meta-heuristic algorithm that combines genetic algorithms with adaptive large neighborhood search: Candidate schedules are decoded into a combination of vehicle access node sequences, an unallocated order pool, and node states. An initial population is constructed using greedy insertion and paired perturbation. During population iteration, disruptive operators such as random removal, worst-case removal, relevance removal, and path removal, as well as repair operators such as greedy insertion, regret value insertion, and relevance anchor insertion, are used to remove and re-insert reallocatable orders in pairs. Simultaneously, the operator selection weights are updated based on the results of the offspring updating the globally optimal solution, being better than the parent, or entering the next generation of the population. This allows for adjusting the invocation probability of different operators based on their historical search performance, balancing neighborhood development capabilities and population search diversity.

[0141] 6. Dual-track evolution and stagnation-triggered large-scale reconstruction mechanism: Offspring are generated using an elite track and an exploration track. The elite track uses the currently preferred candidate scheduling solution as the primary parent and employs a smaller initial disruption ratio. The exploration track selects its parent through a tournament and employs a second disruption ratio greater than the initial disruption ratio. When the number of generations without a continuously updated globally optimal candidate scheduling solution reaches a stagnation threshold, the disruption ratios of both tracks are increased, while maintaining the second disruption ratio greater than the first. This allows for the expansion of the search range during stagnation phases while preserving the local exploitation capabilities of high-quality solutions, reducing the likelihood of the population remaining in locally optimal regions for extended periods.

[0142] 7. A two-level local search mechanism that does not change vehicle ownership: For delivery nodes bound to vehicles that are not the starting point of the replanning route, while keeping their vehicle affiliation unchanged, attempts are made to slide and re-insert them along the same vehicle path forward or backward. For candidate paths that meet preset conditions, adjacent nodes within the same vehicle path can also be swapped. The adjustment result is adopted only if it still meets the constraints of pickup and delivery timing, same-vehicle service, service time window, and vehicle capacity, and can shorten the travel distance or reduce travel and waiting costs. This can repair local path detours caused after inserting new orders and improve path quality without disrupting vehicle binding relationships and task execution continuity.

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0144] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic route scheduling method for pickup and delivery vehicles, characterized in that, include: The system acquires dynamically arriving pickup and delivery orders, vehicle operation status, and road network data, and divides the scheduling cycle into multiple rolling time slices. Pickup and delivery orders include pairs of pickup nodes and delivery nodes. Vehicle operation status includes the current target node and task execution status. Task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered. In the first rolling time slice, an initial planning model is constructed based on the initial order, the initial vehicle operating status, and road network data, and the first delivery route is obtained by solving the problem. In subsequent rolling time slices, the updated vehicle running status is obtained by acquiring new orders and the delivery routes executed by vehicles in the previous rolling time slice. Based on the task execution status in the updated vehicle running status, the vehicle binding nodes are determined from the nodes that have not completed the service. The remaining nodes that have not formed a vehicle binding relationship are reassignable nodes. Vehicle-bound nodes include the current target node that the vehicle is heading to or serving, the delivery node paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the order but has not yet delivered it. A replanning model is constructed based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is taken as the starting point of the vehicle's replanning path. When the current target node is a pickup node, the delivery node paired with it is kept in the vehicle's replanning path, and reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis. Solve the replanning model to obtain the delivery route for the current rolling time slice, and then re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning process until the scheduling cycle ends.

2. The dynamic pickup and delivery vehicle route scheduling method as described in claim 1, characterized in that, The pickup and delivery order also includes node demand, node service time, and service time window. The vehicle operation status also includes vehicle location, vehicle load, and vehicle availability time. In subsequent rolling time slices, based on the updated vehicle operation status, the node demand of the current target node, the node service time and service time window, and the road network data, the estimated departure time and estimated load of each vehicle when it completes the service of the current target node and leaves are determined. The estimated departure time and estimated load are then used as the initial time state and initial load state of the corresponding vehicle in the replanning model, respectively.

3. The dynamic pickup and delivery vehicle route scheduling method as described in claim 2, characterized in that, The vehicle capacity is also obtained; the optimization objectives and constraints of the initial planning model and the replanning model include: The initial planning model aims to maximize the expected net profit of the initial order; the replanning model aims to maximize the sum of the determined profit and the expected profit increment of the current order to be planned; wherein, the determined profit is the profit obtained by deducting the incurred transportation costs from the revenue of completed orders and orders in transit, the current order to be planned is the order belonging to the reassignable node, and the expected profit increment is the profit obtained by deducting the estimated transportation costs from the expected revenue of the current order to be planned. Both the initial planning model and the replanning model include order selection constraints, pairing pickup and delivery nodes serving in the same vehicle constraints, pickup preceding delivery time constraints, vehicle path flow balancing constraints, service time window constraints, and vehicle capacity constraints based on the vehicle capacity. The vehicle capacity constraints are used to limit the load when a vehicle leaves any access node to not exceed the corresponding vehicle capacity. The replanning model also includes path start constraints for the current target node, initial time state constraints corresponding to the expected departure time, initial load state constraints corresponding to the expected load, and constraints that the vehicle-bound node remains served by the corresponding vehicle.

4. The dynamic pickup and delivery vehicle route scheduling method as described in claim 3, characterized in that, Solving both the initial planning model and the replanning model involves constructing an initial population. The process of constructing the initial population includes: Candidate schedules are decoded into a combination of access node sequence for each vehicle, unallocated order pool, and node status. Node status is used to identify whether a vehicle is bound or reassignable. Specifically, in the first rolling time slice, the pickup and delivery nodes corresponding to the initial order are marked as reassignable. In subsequent rolling time slices, node status is set based on the determination results of vehicle bound nodes and reassignable nodes. In the first rolling time slice, a basic access node sequence is established starting from the initial position of each vehicle. In subsequent rolling time slices, the vehicle access node sequence starting from the current target node and including the corresponding vehicle binding node is used as the basic access node sequence. The unassigned orders are sorted according to their expected delivery time. The sorted unassigned orders are then inserted into the basic access node sequence in pairs of pickup and delivery nodes, satisfying the constraints of pickup before delivery, same-vehicle service, service time window, and vehicle capacity, to generate a seed solution. The seed solution is copied, and only for orders corresponding to reassignable nodes, the pickup and delivery nodes of the same order are removed and re-inserted in pairs until the initial population reaches the population size determined based on the number of orders to be planned based on the current rolling time slice.

5. The dynamic pickup and delivery vehicle route scheduling method as described in claim 4, characterized in that, Solving the initial programming model and the reprogramming model based on the initial population includes: Using the initial population as the current population, candidate scheduling solutions are selected from the current population as the main parent generation. Destruction operators and repair operators are adaptively selected from the destruction operator pool and repair operator pool, respectively. Orders in the main parent generation that are in a reassignable state are removed, inserted, or reorganized to generate offspring. After feasibility verification and selection, the next generation population is formed. Based on whether the offspring updates the current record of the globally optimal candidate scheduling solution, whether it is better than the parent generation that generated the offspring, and whether it is selected into the next generation population, the scores of the called destruction and repair operators are accumulated, and the selection weights of the corresponding operators are updated based on the accumulated scores and the number of times they are used. The destruction operator pool includes random removal operator, worst-case removal operator, correlation removal operator and path removal operator, and the repair operator pool includes greedy insertion operator, regret value insertion operator and correlation anchor insertion operator.

6. The dynamic pickup and delivery vehicle route scheduling method as described in claim 5, characterized in that, The insertion process of the correlation anchor insertion operator includes: The in-transit delivery nodes in the vehicle binding nodes are determined as the relevant anchor points of the corresponding vehicles; the in-transit delivery nodes include delivery nodes paired with the pickup node when the current target node is the pickup node, and delivery nodes that have picked up the goods but have not yet delivered the order. For each unassigned order, vehicles with non-empty relevance anchor point sets and insertion positions that satisfy the constraints of pickup before delivery, same-vehicle service, service time window, and vehicle capacity are identified as candidate vehicles. When the candidate vehicle exists, the spatial correlation term and the temporal correlation term between the unassigned order and each correlation anchor point of each candidate vehicle are calculated respectively. The weighted sum of the spatial correlation term and the temporal correlation term is used as the correlation value between the unassigned order and the corresponding correlation anchor point. The minimum value among the correlation values ​​corresponding to the same candidate vehicle is determined as the correlation measure between the unassigned order and the candidate vehicle. Among the candidate vehicles, the unassigned order is preferentially assigned to a candidate vehicle with a lower relevance metric, and the target insertion position is determined from the insertion positions of that candidate vehicle that satisfy the constraints.

7. The dynamic pickup and delivery vehicle route scheduling method as described in claim 5, characterized in that, Iterative solutions based on the initial population include: Offspring are generated through the elite track and the exploration track respectively; wherein, the main parent generation includes the main parent generation of the elite track and the main parent generation of the exploration track; The elite track determines the best candidate scheduling solution in the current population as the elite track's main parent generation, and performs destruction and repair on the elite track's main parent generation according to the first destruction ratio; The exploration track selects candidate scheduling solutions from the current population through a tournament to serve as the main parent of the exploration track, and performs destruction and repair on the main parent of the exploration track according to a second destruction ratio greater than the first destruction ratio; The first destruction ratio and the second destruction ratio are determined based on the number of orders corresponding to the reassignable nodes in the corresponding parent generation; When the number of consecutive algebras in which the currently recorded global optimal candidate scheduling solution has not been updated reaches the stagnation threshold, the first destruction ratio and the second destruction ratio are increased.

8. A dynamic pickup and delivery vehicle route scheduling device, characterized in that, include: The data acquisition and time slice division module is used to acquire dynamically arriving pickup and delivery orders, vehicle operation status and road network data, and divide the scheduling cycle into multiple rolling time slices; pickup and delivery orders include pairs of pickup nodes and delivery nodes, vehicle operation status includes the current target node and task execution status, and task execution status is used to identify whether the node has completed the service and whether the order is in the state of being picked up but not delivered. The initial planning module is used to build an initial planning model based on the initial orders, the initial vehicle operating status, and road network data in the first rolling time slice, and solve for the first delivery route. The node status determination module is used to obtain the updated vehicle operation status formed by new orders and the delivery routes executed by vehicles in the previous rolling time slice in subsequent rolling time slices. Based on the task execution status included in the updated vehicle operation status, it determines the vehicle-bound nodes from the nodes that have not completed service. The remaining nodes that have not formed a vehicle-bound relationship are reassignable nodes. Vehicle-bound nodes include the current target node that the vehicle is heading to or serving, the delivery node paired with the pickup node when the current target node is a pickup node, and the delivery node that has picked up the order but has not yet delivered it. The replanning model building module is used to build a replanning model based on new orders, vehicle-bound nodes, reassignable nodes, and road network data. In the replanning model, the current target node that the vehicle is heading to or serving is taken as the starting point of the vehicle's replanning path. Delivery nodes that have picked up but not yet delivered orders are constrained in the replanning path of the vehicle that performed the corresponding pickup operation, and reassignable nodes are allowed to be redistributed among vehicles on an order-by-order basis. The path solving and rolling update module is used to solve the replanning model, obtain the delivery route for the current rolling time slice, and re-acquire new orders and updated vehicle operating status in the next rolling time slice to repeat the replanning until the scheduling cycle ends.

9. A storage medium, characterized in that, It stores a computer program, wherein the computer program causes the computer to perform the dynamic pickup and delivery vehicle route scheduling method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: A processor and a memory; the memory stores program instructions; the processor is configured to run the program instructions to execute the dynamic pickup and delivery vehicle route scheduling method as described in any one of claims 1 to 7.

Images