A high-speed rail express end distribution scheduling method based on dynamic unloading decision

Abstract

This invention discloses a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making, belonging to the field of route scheduling technology. This method incorporates unloading station allocation into the delivery scheduling optimization process, enabling coordinated adjustment of unloading stations and vehicle routes. This allows unloading stations to adaptively adjust their unloading locations based on the spatial distribution of delivery demand points and changes in transportation costs, thereby reducing the overall transportation distance. Simultaneously, through iterative optimization and strategy feedback adjustment based on comprehensive delivery costs, and by introducing local structural adjustments and constraint verification mechanisms during the optimization process, constraints such as vehicle load, time windows, and unloading batches are always met while reducing path length, thus improving the executability of the scheduling scheme. This invention uses each station as a regional delivery starting point in the scheduling decision-making process, achieving zonal coverage of delivery demand points without the need for additional transfer nodes.

Description

Technical Field

[0001] This invention relates to the field of freight route scheduling technology, specifically a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decisions. Background Technology

[0002] Currently, high-speed rail express trains in cities with multiple high-speed rail stations operate under two modes: single-station stops and multi-station stops. The multi-station stop mode offers flexible unloading options at multiple stations for freight distribution. High-speed rail stations are mostly distributed across different areas of the city, forming a regional radiation pattern and are well-connected to urban transportation. This gives it a natural advantage in the "decentralized distribution center" model, allowing it to directly serve as a point of demand for cargo distribution and thus, to some extent, replacing traditional secondary distribution centers and reducing transshipment costs.

[0003] In the organization and delivery methods for multi-station stopping patterns, a static allocation method based on distance cost rules is usually adopted. The allocation results of unloading high-speed rail stations are not adjusted in subsequent route planning. This fails to treat high-speed rail stations as dynamic decision variables that can participate in route optimization, resulting in the separation between unloading station selection and last-mile delivery route planning. It is difficult to make coordinated adjustments based on the spatial distribution of delivery demand, time constraints, and changes in transportation costs, thereby limiting the potential advantages of multi-station stopping patterns in reducing overall delivery costs and improving delivery efficiency. Summary of the Invention

[0004] (a) Technical problems to be solved This invention provides a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision, which realizes the coordinated adjustment of unloading stations and high-speed rail vehicle delivery routes.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision, comprising the following steps: Obtain and calculate the transportation distance between each delivery demand point and the high-speed rail station; Spatial clustering is performed on all delivery demand points, and delivery demand points that are close to each other are grouped into the same cluster; for each cluster, the high-speed rail station with the shortest transportation distance is assigned as the initial unloading station of the cluster based on the midpoint of all delivery demand points in the cluster. For each set of delivery demand points at the initial unloading station, starting from the corresponding high-speed rail station, the nearest neighbor heuristic algorithm is used to form an initial scheduling scheme covering all delivery demand points under the condition of satisfying the set load constraints. Each initial scheduling scheme is iteratively optimized; in each iteration, delivery demand points that do not meet the constraints are removed based on the set delivery time constraints, and then a minimum increment greedy insertion strategy is adopted to traverse different high-speed rail stations and select the potential insertion position with the minimum delivery cost increment to construct the delivery path. After each iteration, the overall delivery cost is calculated for the updated scheduling scheme. The overall delivery cost includes transportation distance and time constraint penalty. The changes in overall delivery cost caused by different delivery demand point removal and insertion strategies in each iteration are recorded to increase the selection probability of strategies that reduce the overall delivery cost. The selection probability is normalized and updated based on the cumulative cost reduction of each strategy in historical iterations. When the decrease in overall delivery cost falls below the set range in multiple consecutive iterations, the optimization process terminates, and the final scheduling scheme, including the allocation results of unloading stations for delivery demand points and vehicle delivery routes for each high-speed rail station, is output.

[0006] In some feasible embodiments, the set of high-speed rail stations includes the spatial location of each high-speed rail station, and the set of delivery demand points includes the geographical coordinates of all delivery demand points in the selected city, and obtains the number of dispatchable delivery vehicles and the capacity of vehicles that can enter the station for each high-speed rail station, as well as the demand and delivery time window demand for each of the delivery demand points.

[0007] In some feasible implementations, when performing spatial clustering based on the K-means algorithm on all delivery demand points, the following steps are performed: Will Each delivery demand point is divided according to its geographical coordinates. For each cluster, the centroid position of the cluster is updated iteratively to minimize the sum of squared Euclidean distances between delivery demand points within the same cluster. The clustering function is: ; in, Indicates the first Clusters, Represents the centroid coordinates of the cluster. The coordinate vector of the delivery demand point; Subsequently, based on the centroid coordinates of each cluster, the high-speed rail station with the closest spatial distance to the delivery demand point of each cluster is taken as the unloading station, thus obtaining the initial allocation relationship between the delivery demand point and the unloading station.

[0008] In some feasible embodiments, after the initial allocation relationship between delivery demand points and unloading stations is completed, for each cluster, the access order is gradually generated using a nearest neighbor heuristic algorithm, starting from the corresponding high-speed rail station, including the following steps: Initialize the delivery route starting from the high-speed rail station. ; In the cluster of delivery demand points that have not yet been visited, select the one corresponding to the current high-speed rail station. Select the delivery point closest to your current delivery request point. Join path: ; in, This indicates the optimal delivery demand point to be selected. The parameter represents the minimum value, that is, finding the value that minimizes the expression. value, Indicates the first The set of delivery demand points in a cluster, This represents the currently constructed sequence of route delivery demand points. This represents the set of delivery demand points in the cluster that have not yet been visited. Indicates the current delivery demand point and the delivery demand point The Euclidean distance between them; Update the current delivery demand points and repeat the above process until all delivery demand points have been visited. Finally, it returns to the high-speed rail station to form a complete delivery loop.

[0009] In some feasible embodiments, after obtaining the initial delivery plan, an adaptive large neighborhood search algorithm is introduced to iteratively optimize the scheduling plan; the following operations are performed sequentially in each iteration: An adaptive destruction operator is selected to remove some delivery demand points in the current initial delivery plan. The destruction operator includes a random removal operator and a correlation removal operator. The correlation removal operator removes delivery demand points that are geographically close and have similar time constraints in batches, based on the spatial distance between delivery demand points and the difference in time window deadlines. After removing delivery demand points, a repair operation is performed on the unassigned delivery demand points. The repair phase employs a minimum incremental greedy insertion strategy, traversing all feasible delivery paths and their possible insertion positions for each delivery demand point to be inserted, and calculating the incremental delivery cost after insertion. ; in, This is the current route delivery demand point sequence. For the delivery demand points to be inserted; Insertion is performed at the location with the smallest increase in delivery cost, thereby reconstructing the delivery route.

[0010] In some feasible embodiments, after each iteration completes the insertion of new delivery demand points, a partial update optimization is performed on the current delivery plan, including the following optimization processes: Using a 2-opt swapping strategy, the access order of delivery demand points in each vehicle's delivery route is partially adjusted to reduce route detours by swapping some access orders in the route; if the delivery distance is reduced after the adjustment, the corresponding delivery route is updated. A consistency check is performed on the allocation relationship between delivery demand points and high-speed rail stations in the current delivery plan. For delivery demand points that increase transportation costs due to re-insertion, their available high-speed rail stations are reassessed, and stations with lower transportation costs are selected for reallocation.

[0011] In some feasible embodiments, constraint verification is performed on the delivery plan after the partial update, checking the load limit of the high-speed rail vehicles, the delivery time window constraint, and the unloading batch constraint; if there are any constraints that are not met, the corresponding delivery route is adjusted or the current adjustment result is revoked.

[0012] In some feasible embodiments, the initial scheduling scheme is iteratively processed using an adaptive large neighborhood search method. In each iteration, based on the selection probabilities of various removal and insertion strategies currently set, the corresponding strategy is selected to remove some delivery demand points in the current initial scheduling scheme, and the delivery path is reconstructed through the insertion method of the minimum potential increment to obtain a new scheduling scheme after each iteration. For each new scheduling scheme generated in each iteration, its comprehensive delivery cost is calculated and compared with the scheduling scheme before the iteration; if the delivery cost of the new scheduling scheme decreases, it is updated, and the change in comprehensive delivery cost caused by the removal and insertion strategies used in this iteration is recorded. Based on the cumulative change in the overall delivery cost generated by various removal and insertion strategies in multiple iterations, the selection probability is normalized and updated, and the removal and insertion strategies that reduce the overall delivery cost are assigned a high selection probability value. If the decrease in overall delivery cost falls below a set range in multiple consecutive iterations, the search process is considered to have converged and the iteration is terminated.

[0013] In some feasible embodiments, during the iteration process, the new scheduling scheme generated in each round is taken as a candidate solution, and the newly generated candidate set is compared one by one with the scheduling schemes corresponding to each existing solution in the current Pareto solution set. If there exists a scheduling scheme corresponding to any of the existing solutions, and its overall delivery cost is not higher than the overall delivery cost of the candidate solution scheduling scheme, then the candidate solution is removed. If the overall delivery cost of the candidate solution scheduling scheme is lower than the existing solution scheduling schemes in the current Pareto solution set, then the corresponding existing solution is removed from the current Pareto solution set, and the candidate solution is added to the Pareto solution set after this round of update. If the overall delivery cost of the candidate solution scheduling scheme is lower than the existing solution scheduling schemes in the current Pareto solution set, then the candidate solution will be directly added to the Pareto solution set after this round of updates.

[0014] (III) Beneficial Effects: Compared with the prior art, this invention has the following beneficial effects: This invention combines the layout of high-speed rail express network stations with the characteristics of cargo flow. It uses two core stages: dynamic unloading decision-making and last-mile vehicle delivery. The dynamic unloading decision-making introduces a dynamic adjustment mechanism for unloading stations at delivery demand points during the scheduling process. This transforms the unloading stations from pre-fixed ones into decision variables that can participate in optimization. The unloading location can be adaptively adjusted according to the spatial distribution of delivery demand points and changes in transportation costs, thereby reducing the overall transportation distance.

[0015] Last-mile delivery takes into account the coordinated optimization of unloading site allocation and vehicle delivery route construction. During the re-insertion of demand points, site selection and route structure are considered simultaneously, so that the two are no longer independent of each other, reducing the route detour problem caused by unreasonable site allocation and improving the overall rationality of last-mile delivery routes. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the algorithm executed by a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision provided in an embodiment of the present invention. Figure 2 The planned vehicle delivery route for the first batch provided in this embodiment of the invention; Figure 3 The planned delivery route for the second batch of vehicles provided in this embodiment of the invention; Figure 4 The planned third batch of vehicle delivery routes provided in this embodiment of the invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0019] Combination Figures 1 to 4This paper presents a high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision. The method first generates an initial scheme of "unloading station + delivery route" through spatial clustering. Then, under the ALNS framework, it continuously adjusts the station to which the delivery demand point belongs and the vehicle delivery route through destruction, repair and local optimization mechanisms. Under the condition of satisfying capacity and time window constraints, it searches for the Pareto optimal delivery scheme between cost and satisfaction.

[0020] At the beginning of the scheduling phase, basic information about the delivery tasks is first obtained, including the set of high-speed rail stations, the set of delivery demand points, vehicle resource information, and constraints such as vehicle capacity and delivery time windows. Based on this, a distance matrix is ​​constructed between high-speed rail stations and delivery demand points, as well as between delivery demand points themselves. The purpose of this step is to establish a complete delivery spatial network structure.

[0021] In urban logistics and distribution problems, transportation costs and route planning are highly dependent on the spatial relationships between delivery demand points. Therefore, by constructing a distance matrix, the actual urban road spatial relationships can be transformed into a computable network structure, providing basic data support for subsequent cluster analysis, route construction, and neighborhood search.

[0022] Meanwhile, in the embodiments of the present invention, by unifying and integrating delivery demand points, vehicle resources and delivery constraints, a delivery task environment with clear delivery demand point relationships and constraints can be formed, enabling subsequent algorithms to perform optimization searches within the same data framework.

[0023] In a certain year, the postal industry in a certain city, such as Wuhan, completed 2.264 billion postal and delivery services, of which 1.881 billion were express delivery services. In order to make reasonable use of the city's existing high-speed rail facilities, Station A (Hankou Station), Station B (Wuchang Station), and Station C (Wuhan Station) were selected as stations where high-speed rail freight trains could unload goods. 25 delivery demand points were selected as delivery destinations. The delivery demand points were selected from commercial stores of a certain size, and their geographical locations were evenly distributed.

[0024] To facilitate the observation of the calculation results, the destination delivery demand point number corresponding to each cargo is set to the same number as the cargo number when designing the calculation example. For example, the destination corresponding to cargo group 19 is delivery demand point 19, and the weight of the sub-cargoes constituting each cargo group meets the loading and unloading requirements of high-speed rail express operation. The coordinates of each delivery demand point are obtained through Gaode Map and the latest expected delivery time is set. The specific information of each delivery demand point is shown in Table 1 below.

[0025] Table 1. Detailed information on each delivery demand point Considering the scenario of multiple trains arriving for unloading, it is planned to use five high-speed express trains, D352, G1772, G586, G676, and G1724, to carry an average of 25 sets of goods. The specific information of each train is shown in Table 2.

[0026] Table 2 Specific information for each train Considering the loading time after trains arrive at the station and integrating train arrival times, trains D352 and G1772 are designated as the first batch of arriving trains, with delivery operations commencing at 12:15; trains G586 and G676 are designated as the second batch of arriving trains, with delivery operations commencing at 14:25; and train G1724 is designated as the third batch of arriving trains, with delivery operations commencing at 16:24. Each high-speed rail station is equipped with 5 last-mile delivery trains available for use. Based on research into the operating costs and load capacity of ZTO Express trucks, the vehicle parameters are as follows: vehicle speed 45 km / h, maximum load 15 t, cost per kilometer of operation 2 yuan, and fixed operating cost per truck 100 yuan.

[0027] After obtaining the basic network structure, spatial clustering based on the K-means algorithm is performed on the set of delivery demand points. K-means is a classic unsupervised clustering method. Its basic principle is to continuously update the cluster center position to minimize the sum of squared distances from samples within a cluster to the center point, thereby achieving a relatively compact spatial partitioning structure.

[0028] In the high-speed rail express last-mile delivery scenario, different delivery demand points usually exhibit certain spatial clustering characteristics in urban space. Therefore, the K-means algorithm can be used to divide delivery demand points with similar geographical locations into the same cluster.

[0029] Specifically, it will Each delivery demand point is divided according to its geographical coordinates. For each cluster, the centroid position of the cluster is updated iteratively to minimize the sum of squared Euclidean distances between delivery demand points within the same cluster. The clustering function is: ； in, Indicates the first Clusters, Represents the centroid coordinates of the cluster. The coordinate vector of the delivery demand point; Subsequently, based on the spatial location of each cluster, the delivery demand points of each cluster are matched with the nearest high-speed rail station to obtain the initial allocation relationship between the delivery demand points and the unloading stations.

[0030] This spatial zoning strategy forms several local service areas in the distribution network, and selects high-speed rail stations that are close to or have matching service capabilities as unloading stations based on the spatial location of each cluster.

[0031] First, a data point is randomly selected as the first cluster center, and then subsequent center points are selected according to a probability distribution: ; in Point The shortest distance to the selected center.

[0032] The purpose of this approach is to leverage the distribution advantages of high-speed rail stations in cities to pre-distribute delivery demand based on spatial proximity, thereby avoiding station congestion caused by all goods being unloaded at the terminal. At the same time, it shortens the average service radius of subsequent vehicle deliveries, providing a better initial structure for optimizing last-mile delivery routes.

[0033] After establishing the mapping between delivery demand points and unloading stations, for each cluster of delivery demand points, the corresponding high-speed rail station is used as the delivery starting point. A nearest neighbor heuristic algorithm is then used to progressively generate the access order, including initializing the delivery path. Then, from the clusters of delivery demand points that have not yet been visited, select the one corresponding to the current high-speed rail station. Select the delivery point closest to your current delivery request point. Join path: ; in, This indicates the optimal delivery demand point to be selected. The parameter represents the minimum value, that is, finding the value that minimizes the expression. value, Indicates the first The set of delivery demand points in a cluster, This represents the currently constructed sequence of route delivery demand points. This represents the set of delivery demand points in the cluster that have not yet been visited. Indicates the current delivery demand point and the delivery demand point The Euclidean distance between them; Update the current delivery demand points and repeat the above process until all delivery demand points have been visited. Finally, return to the high-speed rail station to form a complete delivery loop. It is worth noting that in practical applications, the path length generated by the nearest neighbor algorithm is typically within 10%-15% of the optimal solution, fully meeting the quality requirements of the initial solution.

[0034] After constructing the initial route based on "batch-site" spatial clustering, some delivery demand points may still not be included in any vehicle route due to capacity limitations, cluster boundary effects, or spatial isolation. To ensure the feasibility and completeness of the initial solution, embodiments of the present invention introduce a systematic "capacity-feasible insertion" mechanism in the second stage to process the uncovered delivery demand points one by one.

[0035] Specifically, after obtaining the initial scheduling scheme, an adaptive large neighborhood search algorithm is introduced to iteratively optimize the scheme.

[0036] The Adaptive Large Neighborhood Search (ALNS) here is a metaheuristic search method. Its basic idea is to reconstruct part of the structure of the current solution by repeatedly performing "destruction-repair" operations, thereby finding a better solution in a larger solution space.

[0037] In each iteration, the algorithm first adaptively selects a disruptive operator based on the operator weights, and removes some delivery demand points in the current scheduling scheme.

[0038] Disruption operators are typically designed using two strategies: random removal and relevance removal. Random removal expands the search scope and prevents the algorithm from getting stuck in local optima too early. Relevance removal, on the other hand, removes delivery demand points that are geographically close or have similar delivery constraints in batches based on the spatial distance between delivery demand points and the similarity of their time window deadlines, thereby changing the delivery structure of an area at once. This disruption of local structures can effectively break the original path structure, creating new possible combinations for subsequent path reconstruction.

[0039] After removing delivery request points, perform a repair operation on the unassigned delivery request points.

[0040] During the repair phase, a minimum incremental greedy insertion strategy is adopted. For each delivery demand point to be inserted, all feasible delivery paths and possible insertion positions within those paths are traversed, and the delivery cost increment caused by inserting the delivery demand point is calculated. ; in, This is the current route delivery demand point sequence. For the delivery demand points to be inserted, select the location with the smallest cost increment to complete the insertion.

[0041] The principle behind this strategy is to minimize the additional costs by evaluating the impact of inserting delivery demand points on path length and delivery costs, thereby gradually building a more compact delivery path structure. Through repeated combinations of disruption and repair, the algorithm can continuously adjust the service stations and vehicle access order of delivery demand points, gradually optimizing the unloading station allocation and delivery path structure during the search process.

[0042] Based on the above example of urban freight transport, some Pareto solutions obtained by the hybrid heuristic algorithm based on adaptive large neighborhood search are shown in Table 3.

[0043] Table 3 Partial Pareto Solution Results Comparing the above delivery solutions, although Solution 1 has the lowest cost, its satisfaction rate at delivery demand points is only 75.66%, which is far lower than other solutions. This does not conform to the core positioning of high-speed rail express as high-timeliness and is likely to lead to the loss of delivery demand points. Although Solution 5 has the highest satisfaction rate, its cost of 1239.53 yuan reaches the highest value in the Pareto solution set. Excessive operating costs will weaken market competitiveness.

[0044] Option 4 has a delivery cost that is only about 100 yuan higher than Options 2 and 3, while improving satisfaction by about 2%, achieving a high satisfaction rate of 87.72% for delivery demand points. It only requires one more vehicle and a slight increase in mileage, achieving a service quality that better meets market demands at a relatively small cost, thus achieving the best balance between cost and satisfaction. Therefore, taking Option 4 as an example, the unloading decision scheme and vehicle delivery scheme are shown in Tables 4 and 5, and the visualization of vehicle routes for each batch is shown below. Figures 2 to 4 As shown.

[0045] Table 4 Unloading Plan and Delivery Schedule Table 5 Specific Delivery Plans for Each Batch of Vehicles After path reconstruction, a local search optimization is performed on the current scheduling scheme to improve the internal structure of the path and ensure the feasibility of the scheme.

[0046] In the local search phase, a 2-opt swap strategy is used to optimize the access order of delivery demand points in the vehicle path. The principle is to disconnect the two edges connecting any two non-adjacent delivery demand points in a single vehicle path, reverse the order of the sub-paths between the two delivery demand points, and then reconnect them to form a new path. By iterating this swap operation repeatedly, the path scheme with the shorter total distance is selected, ultimately eliminating any potential redundancy in the original path.

[0047] Considering the involvement of multi-station unloading rules, the local search must ensure that all delivery demand points are assigned to vehicle routes matching their unloading stations. Therefore, through station consistency repair and batch feasibility maintenance, necessary adjustments are made to the allocation relationship between delivery demand points and high-speed rail stations, as well as the vehicle route structure, to ensure that the generated scheduling scheme meets vehicle capacity constraints, delivery time window constraints, and station service capacity constraints.

[0048] This process can further reduce delivery distance and improve delivery efficiency while maintaining route feasibility.

[0049] After each iteration, the current scheduling scheme is evaluated, and the Pareto solution set is updated based on two objectives: total delivery cost and average satisfaction at delivery demand points.

[0050] Since there is usually a trade-off between delivery costs and the service level of delivery demand points, maintaining multiple sets of non-dominant optimal solutions using Pareto fronts can provide alternative results with different emphases for actual scheduling decisions. Simultaneously, based on the contribution of each destruction operator and repair operator to generating high-quality solutions during the search process, their weights are adaptively updated, ensuring that better-performing operators are selected more frequently in subsequent iterations, thereby gradually improving the algorithm's search efficiency.

[0051] When the preset number of iterations is reached or the search process stabilizes, the adaptive large neighborhood search process terminates, and the final delivery scheduling scheme is output. This scheme includes the unloading and allocation results of goods at each high-speed rail station and the corresponding vehicle delivery routes for each high-speed rail station. By realizing the reasonable diversion of goods between different stations in advance, and combined with the optimized vehicle delivery route structure, the last-mile delivery distance in cities can be effectively reduced, the pressure of concentrated unloading of goods at the terminal station can be alleviated, and the overall delivery efficiency and service level of delivery demand points can be improved while controlling delivery costs.

[0052] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention should also be included within the scope of protection of the present invention.

Claims

1. A high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision, characterized in that, This includes performing the following steps: Obtain and calculate the transportation distance between each delivery demand point and the high-speed rail station; Spatial clustering is performed on all delivery demand points, and delivery demand points that are close to each other are grouped into the same cluster; for each cluster, the high-speed rail station with the shortest transportation distance is assigned as the initial unloading station of the cluster based on the midpoint of all delivery demand points in the cluster. For each set of delivery demand points at the initial unloading station, starting from the corresponding high-speed rail station, the nearest neighbor heuristic algorithm is used to form an initial scheduling scheme covering all delivery demand points under the condition of satisfying the set load constraints. Each initial scheduling scheme is iteratively optimized; in each iteration, delivery demand points that do not meet the constraints are removed based on the set delivery time constraints, and then a minimum increment greedy insertion strategy is adopted to traverse different high-speed rail stations and select the potential insertion position with the minimum delivery cost increment to construct the delivery path. After each iteration, the overall delivery cost is calculated for the updated scheduling scheme. The overall delivery cost includes transportation distance and time constraint penalty. The changes in overall delivery cost caused by different delivery demand point removal and insertion strategies in each iteration are recorded to increase the selection probability of strategies that reduce the overall delivery cost. The selection probability is normalized and updated based on the cumulative cost reduction of each strategy in historical iterations. When the decrease in overall delivery cost falls below the set range in multiple consecutive iterations, the optimization process terminates, and the final scheduling scheme, including the allocation results of unloading stations for delivery demand points and vehicle delivery routes for each high-speed rail station, is output.

2. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision as described in claim 1, characterized in that, The set of high-speed rail stations includes the spatial location of each high-speed rail station, and the set of delivery demand points includes the geographical coordinates of all delivery demand points within the selected city, and obtains the number of dispatchable delivery vehicles and the capacity of vehicles that can enter the station for each high-speed rail station, as well as the demand and delivery time window demand for each of the delivery demand points.

3. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making as described in claim 1, characterized in that, When performing spatial clustering based on the K-means algorithm on all delivery demand points, the following steps are performed: Will Each delivery demand point is divided according to its geographical coordinates. For each cluster, the centroid position of the cluster is updated iteratively to minimize the sum of squared Euclidean distances between delivery demand points within the same cluster. The clustering function is: ； in, Indicates the first Clusters, Represents the centroid coordinates of the cluster. The coordinate vector of the delivery demand point; Subsequently, based on the centroid coordinates of each cluster, the high-speed rail station with the closest spatial distance to the delivery demand point of each cluster is taken as the unloading station, thus obtaining the initial allocation relationship between the delivery demand point and the unloading station.

4. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making according to claim 3, characterized in that, After completing the initial allocation relationship between delivery demand points and unloading stations, for each cluster, the nearest neighbor heuristic algorithm is used to gradually generate the access order, starting from the corresponding high-speed rail station, including the following steps: Initialize the delivery route starting from the high-speed rail station. ; In the cluster of delivery demand points that have not yet been visited, select the one corresponding to the current high-speed rail station. Select the delivery point closest to your current delivery request point. Join path: ； in, This indicates the optimal delivery demand point to be selected. The parameter represents the minimum value, that is, finding the value that minimizes the expression. value, Indicates the first The set of delivery demand points in a cluster, This represents the currently constructed sequence of route delivery demand points. This represents the set of delivery demand points in the cluster that have not yet been visited. Indicates the current delivery demand point and the delivery demand point The Euclidean distance between them; Update the current delivery demand points and repeat the above process until all delivery demand points have been visited. Finally, it returns to the high-speed rail station to form a complete delivery loop.

5. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making according to claim 4, characterized in that, After obtaining the initial delivery plan, an adaptive large neighborhood search algorithm is introduced to iteratively optimize the scheduling plan; the following operations are performed in each iteration: An adaptive destruction operator is selected to remove some delivery demand points in the current initial delivery plan. The destruction operator includes a random removal operator and a correlation removal operator. The correlation removal operator removes delivery demand points that are geographically close and have similar time constraints in batches, based on the spatial distance between delivery demand points and the difference in time window deadlines. After removing delivery demand points, a repair operation is performed on the unassigned delivery demand points. The repair phase employs a minimum incremental greedy insertion strategy, traversing all feasible delivery paths and their possible insertion positions for each delivery demand point to be inserted, and calculating the incremental delivery cost after insertion. ； in, This is the current route delivery demand point sequence. For the delivery demand points to be inserted; Insertion is performed at the location with the smallest increase in delivery cost, thereby reconstructing the delivery route.

6. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision as described in claim 1, characterized in that, After each iteration inserts a new delivery demand point, a partial update and optimization is performed on the current delivery plan, including the following optimizations: Using a 2-opt swapping strategy, the access order of delivery demand points in each vehicle's delivery route is partially adjusted to reduce route detours by swapping some access orders in the route; if the delivery distance is reduced after the adjustment, the corresponding delivery route is updated. A consistency check is performed on the allocation relationship between delivery demand points and high-speed rail stations in the current delivery plan. For delivery demand points that increase transportation costs due to re-insertion, their available high-speed rail stations are reassessed, and stations with lower transportation costs are selected for reallocation.

7. A high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making as described in claim 6, characterized in that, The delivery plan after the partial update is subjected to constraint verification, which checks the load limit of the high-speed rail vehicles, the delivery time window constraint, and the unloading batch constraint. If any constraint is not met, the corresponding delivery route is adjusted or the current adjustment result is revoked.

8. The high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision as described in claim 1, characterized in that, The initial scheduling scheme is iteratively processed using an adaptive large neighborhood search method. In each iteration, based on the selection probabilities of various removal and insertion strategies, the corresponding strategy is selected to remove some delivery demand points in the current initial scheduling scheme, and the delivery path is reconstructed by inserting the minimum potential increment to obtain a new scheduling scheme after each iteration. For each new scheduling scheme generated in each iteration, its comprehensive delivery cost is calculated and compared with the scheduling scheme before the iteration; if the delivery cost of the new scheduling scheme decreases, it is updated, and the change in comprehensive delivery cost caused by the removal and insertion strategies used in this iteration is recorded. Based on the cumulative change in the overall delivery cost generated by various removal and insertion strategies in multiple iterations, the selection probability is normalized and updated, and the removal and insertion strategies that reduce the overall delivery cost are assigned a high selection probability value. If the decrease in overall delivery cost falls below a set range in multiple consecutive iterations, the search process is considered to have converged and the iteration is terminated.

9. A high-speed rail express last-mile delivery scheduling method based on dynamic unloading decision-making as described in claim 8, characterized in that, During the iteration process, the new scheduling scheme generated in each round is taken as a candidate solution. The newly generated candidate set is compared one by one with the scheduling schemes corresponding to the existing solutions in the current Pareto solution set. If there exists a scheduling scheme corresponding to any of the existing solutions, and its overall delivery cost is not higher than the overall delivery cost of the candidate solution scheduling scheme, then the candidate solution is removed. If the overall delivery cost of the candidate solution scheduling scheme is lower than the existing solution scheduling schemes in the current Pareto solution set, then the corresponding existing solution is removed from the current Pareto solution set, and the candidate solution is added to the Pareto solution set after this round of update. If the overall delivery cost of the candidate solution scheduling scheme is lower than the existing solution scheduling schemes in the current Pareto solution set, then the candidate solution will be directly added to the Pareto solution set after this round of updates.

Images