A Method and System for Intercity Customized Passenger Transport Scheduling Integrating Spatiotemporal Coupled Clustering and State Compression Dynamic Programming

By integrating spatiotemporal coupled clustering with state compression dynamic programming, the intercity customized passenger transport scheduling method solves the problems of spatiotemporal correlation and computational efficiency in customized passenger transport scheduling, realizes efficient order aggregation and route optimization, and improves service utility and order acceptance rate.

CN123070541APending Publication Date: 2026-11-10GUANGDONG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611157308.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-31
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing customized passenger transport scheduling solutions do not fully consider the spatiotemporal correlation of order pick-up and drop-off points, the directional characteristics of intercity travel, and the computational efficiency issues in large-scale order scenarios. This results in severe vehicle detours, excessive search space, and decreased operating revenue. Furthermore, it fails to effectively coordinate and optimize the operating profits of passenger transport companies and the quality of passenger service.

Method used

A method combining spatiotemporal coupled clustering and state compression dynamic programming is adopted. Orders are clustered by constructing a spatiotemporal integrated distance metric to generate logical stations. Virtual stations are matched with the distance of the map road network to perform directional consistency classification and filtering of candidate departure tasks. An optimization model is constructed to maximize the overall service utility, and state compression dynamic programming is used for scheduling.

Benefits of technology

It effectively reduced the scale of the path optimization problem, improved order aggregation efficiency, increased the overall service utility value and order acceptance rate, while reducing computational complexity and detour costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070541A_ABST
    Figure CN123070541A_ABST
Patent Text Reader

Abstract

This application relates to an intercity customized passenger transport scheduling method and system that integrates spatiotemporal coupled clustering and state compression dynamic programming. The method includes: acquiring customized passenger transport passenger order information; constructing a spatiotemporal comprehensive distance metric integrating spatial and temporal distances to cluster the orders and form logical stations; using the centroids of the logical stations as virtual stations, retrieving candidate bus stations, and combining map road network distance calculations to complete the matching mapping between virtual and real stations; constructing a rolling time window based on the customized passenger transport passenger order information to form candidate departure tasks; performing pre-filtering on the candidate departure tasks to obtain shortlisted departure tasks; constructing an optimization model; and performing passenger transport scheduling. This invention constructs a spatiotemporal comprehensive distance metric integrating spatial and temporal distances to cluster customized passenger transport orders and form logical stations; simultaneously, it makes decisions to retain or remove isolated orders, achieving adaptive aggregation and demand integration of large-scale orders; and further, through accurate mapping from virtual stations to real bus stations, it effectively reduces the problem scale of subsequent path optimization.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to an intercity customized passenger transport scheduling method and system that integrates spatiotemporal coupled clustering and state compression dynamic programming. Background Technology

[0002] With the integrated development of urban clusters and the increasing demand for cross-city commuting, short- and medium-distance intercity passenger transport has become an important component of the comprehensive transportation system. Traditional intercity passenger transport operates on a fixed-station, fixed-schedule model, which suffers from insufficient flexibility, low load factors, and low operational efficiency. Meanwhile, rail transport and ride-hailing services face issues such as inconvenient connections or high travel costs. Therefore, customized passenger transport, which can dynamically organize capacity based on passenger demand, is gradually becoming an important supplement to intercity travel.

[0003] However, most existing customized passenger transport scheduling solutions do not fully consider the spatiotemporal correlation of order pick-up and drop-off points, the directional characteristics of intercity travel, and the computational efficiency issues in large-scale order scenarios, which can easily lead to severe vehicle detours, excessive search space, and decreased operating revenue. At the same time, existing studies mostly focus on the shortest distance or lowest cost as optimization objectives, and do not adequately consider the synergistic optimization of passenger transport companies' operating profits and passenger service quality. Summary of the Invention

[0004] Based on this, it is necessary to provide an intercity customized passenger transport scheduling method and system that integrates spatiotemporal coupled clustering and state compression dynamic programming to address the above-mentioned technical problems and improve the level of intelligent scheduling.

[0005] This invention provides an intercity customized passenger transport scheduling method that integrates spatiotemporal coupled clustering and state compression dynamic programming, including: S1: Obtain customized passenger transport order information; S2: Aggregate order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; construct a spatiotemporal comprehensive distance metric that integrates spatial distance and temporal distance to cluster orders and form logical sites; S3: Using the centroid of the logical station as the virtual station, retrieve candidate bus stops and combine the distance calculation of the map road network to complete the matching and mapping between the virtual station and the real station; S4: Construct a rolling time window based on customized passenger transport order information, aggregate orders with similar times and compatible locations to form candidate departure tasks; S5: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the candidate departure tasks are divided into directions based on consistency; and based on the spatial detour boundary and time window boundary, obviously infeasible candidate tasks are pre-filtered to obtain the candidate departure tasks. S6: With the goal of maximizing the overall utility value of customized passenger transport services, construct an optimization model that includes total revenue, total operating costs, and total user-side time penalties; S7: Conduct passenger dispatching based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.

[0006] Further, in step S2, specifically: S2.1: Construct a spatiotemporal integrated distance metric that integrates spatial and temporal distances, and use historical data to perform closed-loop fine-tuning of scheduling parameters. The spatiotemporal integrated distance metric is defined as follows:

[0007] in, It is an order With orders Geographical distance between boarding points The absolute difference between the two appointment times. , As the normalization factor, , The weighting coefficients are fine-tuned using historical data in a closed-loop manner. S2.2: Canopy coarse clustering: Utilizes T1 and T2 thresholds to quickly define the demand pool, reducing computation and avoiding infinite loops. It is defined as:

[0008] in, .

[0009] S2.3: After the demand pool is constructed, orders that fail to enter any demand pool are judged as candidate isolated orders. An isolated point acceptance coefficient is introduced to quantitatively weigh the additional detour cost of connecting to isolated points and the number of orders. And calculate the isolated point acceptance coefficient, defined as:

[0010] in, It refers to the number of orders. This is the expected additional detour distance to accommodate this isolated point. , It is the adjustment coefficient; when If the isolated point is selected, it will be retained as an independent demand unit for subsequent scheduling; otherwise, it will be discarded.

[0011] S2.4: Find the geometric centroid μ within each Canopy and generate virtual site coordinates, defined as:

[0012] in, It is the geometric centroid of the cluster, serving as the coordinates of the virtual site.

[0013] Further, in step S4, specifically: S4.1 Time-series sorting: Sort all virtual sites in ascending order of average departure time; S4.2 Dynamic Window Storage: One-dimensional scrolling clustering is performed using time_window, storing each... All station requests within the pool are merged into the same shift pool.

[0014] S4.3 Constraint Filtering: Remove assigned orders and verify them. This ensures that the shifts are profitable.

[0015] Furthermore, in step S5, based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the candidate departure tasks are divided into categories based on directional consistency, specifically as follows: S5.1 Four-Dimensional Feature Coupling: Extracting order latitude and longitude coordinates and preprocessing them. Direction vector; S5.2 Feature Matrix Construction Definition:

[0016] in, , These are the latitude and longitude coordinates of the order's pick-up point. It is the order travel direction vector. , These are weighting coefficients used to balance the importance of spatial location and orientation features; S5.3K-Means++ Task Splitting: Calculate the number of vehicles to be dispatched, K = [total number of people / capacity]; perform K-classification in the four-dimensional feature space to assign orders that are going the same way to the same vehicle.

[0017] Furthermore, in step S5, based on the spatial detour boundary and the time window boundary, the obviously infeasible candidate tasks are pre-filtered to obtain the candidate departure tasks, specifically as follows: S5.1 Spatial Detour Prevention and Removal: Based on the Haversine formula and driving direction angle, the spatial topological relationship between stations is evaluated in advance; if the connection of the pick-up and drop-off point sequence deviates significantly from the intercity trunk line, it is directly pruned. S5.2 Time Window Boundary Blocking: Verify the estimated travel time between any two nodes; if the node If the shortest arrival time exceeds the passenger's tolerable extreme waiting time window, then the weight of that edge in the state space matrix is ​​directly set to infinity to block the generation of invalid state branches from the source.

[0018] Further, in step S6, specifically: S6.1 Phased Incremental Utility Actuarial Calculation: The algorithm will transfer the entire path. It can be decomposed into two ordered subsequences. Let Starting point city logical station set The set of logical stations for the destination city.

[0019] Pick-up phase: Starting from point o, in The optimal access order is found within the middle, ending at the high-speed entrance. The state transition is denoted as:

[0020] Drop-off phase: From the highway exit Departure, at The optimal traversal order is achieved by searching through the middle section and ending at the destination. The state transition is denoted as:

[0021] (3) Intercity connection point: The final state of the receiving segment is used as the initial cost of the sending segment, that is:

[0022] in The cost of traveling on intercity expressways (including road and bridge tolls and fuel consumption).

[0023] S6.2 Core Recursive Mechanism: To reduce computational complexity, a single objective function U is directly embedded, employing utility increments. U is accumulated: (1) Single objective function U: maximizing the overall utility of the service (total revenue) - Total operating costs (fixed costs) + Variable Costs Time window penalty cost The target is defined as:

[0024] Expanded to:

[0025] in: It is the unit price of order r. It refers to the number of orders. It is the fixed operating cost per vehicle. It refers to the number of vehicles dispatched. It is the variable cost per kilometer. To maintain costs, For road and bridge tolls, It is a website Arrive at the station driving distance, Is vehicle k from Drive to Decision variables, It is the time window penalty weight coefficient. The mixed time window penalty function is defined as follows:

[0026] in Penalty coefficients for arriving early and late.

[0027] (2) Single-step utility increment, for each additional station It is only necessary to calculate the new revenue of the station minus the variable costs and time window penalty costs of the current road segment, defined as:

[0028] in, It is the total ticket revenue of passengers boarding at station i. It is the variable cost per unit distance. It is a website Arrive at the station driving distance, It is the time window penalty coefficient. This is the time penalty value caused by the arrival time deviation.

[0029] (3) Global state transition, using bitmasks to record the set of stations already served. The overall service utility U is maximized through incremental accumulation, defined as:

[0030] in, It is the last site visited. It was the last site visited.

[0031] S6.3 Constraint-Driven Algorithm Pruning: During the state transition process, constraints such as passenger waiting time window, detour coefficient, vehicle passenger capacity, order uniqueness service, and vehicle entry and exit balance are embedded into the search process; among them, paths exceeding the limit tolerance time window are directly pruned, while waiting deviations within the allowable range are optimized through the time window penalty cost in the objective function.

[0032] Furthermore, the S6.3 constraint-driven algorithm pruning is specifically as follows: (1) Passenger waiting time window constraint: the time when the vehicle arrives at station j Passengers' patience boundaries must be met. Defined as:

[0033] in, This is the maximum tolerable waiting time; (2) Detour coefficient constraint: Define passenger The ratio of the actual driving distance to the optimal distance ensures that the detour is within an acceptable range; defined as:

[0034] in, It is the maximum detour coefficient; (3) Combined constraints of passenger capacity and minimum departure number: The upper limit of single vehicle capacity and the bottom line of profitability are defined as:

[0035] in, This is the minimum number of passengers required for departure. This is the vehicle's maximum passenger capacity; (4) Order Uniqueness Service Constraint: Ensure that each accepted order is served only once and that vehicles are not dispatched repeatedly, defined as:

[0036] (5) Constraints based on a departure strategy that combines fixed and flexible elements; (6) Flow direction deviates from the upper limit:

[0037] in, It is an order The travel direction vector, It is the direction vector of the current road segment traveled by vehicle k.

[0038] Furthermore, the specific constraints of the departure strategy based on the complementarity of fixed and flexible departures are as follows: Triggering integration into a fixed shift: If a fixed shift exists Its preset departure time is And satisfy:

[0039] in, If it is a time window threshold, then the order will be merged into a fixed shift.

[0040] Trigger the creation of flexible shifts: for all This situation opens up a new flexible schedule set for this demand. Its preset departure time To become an optimal decision variable, the order itself must satisfy a flexible time window:

[0041] in:

[0042] The earliest booking time in the current task pool is used as the time base for flexible shifts.

[0043] Further, in step S7, specifically: S7.1 Capacity Splitting: If the number of people in the task pool exceeds the passenger capacity Q, according to... The principle is to automatically select the optimal subset of orders and prioritize vehicle dispatch, thereby achieving task splitting; S7.2 Departure command triggered: When the overall service utility value corresponding to the optimal path is... This indicates that the overall utility of the plan is positive, meaning that ticket revenue is sufficient to cover operating costs and service quality penalties, thus justifying its operation. If the departure is cancelled, the departure will be cancelled to avoid operating losses or substandard service quality.

[0044] Furthermore, this invention provides an intercity customized passenger transport scheduling system that integrates spatiotemporal coupled clustering and state compression dynamic programming, including: Acquisition Module: Retrieves customized passenger transport order information; Logical site construction module: Aggregates order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; constructs a spatiotemporal comprehensive distance metric that integrates spatial distance and temporal distance to cluster orders and form logical sites; Mapping module: Using the centroid of the logical station as the virtual station, it retrieves candidate bus stops and, combined with the distance calculation of the map road network, completes the matching and mapping between the virtual station and the real station; Candidate departure task module: Based on customized passenger transport order information, a rolling time window is constructed to aggregate orders with similar times and compatible locations to form candidate departure tasks; Pre-filtering module: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the module classifies candidate departure tasks according to directional consistency; and based on spatial detour boundaries and time window boundaries, it pre-filters obviously infeasible candidate tasks to obtain the candidate departure tasks. Optimization module: With the goal of maximizing the overall utility value of customized passenger transport services, an optimization model is constructed that includes total revenue, total operating costs, and total user-side time penalties; Scheduling module: Performs passenger transport scheduling based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.

[0045] This invention constructs a spatiotemporal integrated distance metric that integrates spatial and temporal distances, clusters customized passenger transport orders to form logical stations, and makes decisions to retain or remove isolated orders, thereby achieving adaptive aggregation and demand integration of large-scale orders. Furthermore, through the accurate mapping of virtual stations to real bus stations, it effectively reduces the problem scale of subsequent route optimization. Attached Figure Description

[0046] Figure 1 This is a flowchart of an intercity customized passenger transport scheduling method that integrates spatiotemporal coupled clustering and state compression dynamic programming in one embodiment. Figure 2 Customized passenger transport route map for dividing major cities into regions in one embodiment Figure 3 Visualization of spatiotemporal coupled clustering in one embodiment Figure 4 This is a virtual station → real bus stop matching graph in one embodiment. Figure 5 Temporal rolling concatenation clustering visualization in one embodiment Figure 6 This is a decoupled end-to-end optimization path diagram for intercity customized bus services in one embodiment; Figure 7 This is a system block diagram of an intercity customized passenger transport scheduling method that integrates spatiotemporal coupled clustering and state compression dynamic programming in one embodiment. Detailed Implementation To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] Example 1: This embodiment provides, as follows: Figure 1 The intercity customized passenger transport scheduling method shown integrates spatiotemporal coupled clustering and state compression dynamic programming, including: S1: Obtain customized passenger transport order information; S2: Aggregate order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; construct a spatiotemporal comprehensive distance metric that integrates spatial distance and temporal distance to cluster orders and form logical sites; S3: Using the centroid of the logical station as the virtual station, retrieve candidate bus stops and combine the distance calculation of the map road network to complete the matching and mapping between the virtual station and the real station; S4: Construct a rolling time window based on customized passenger transport order information, aggregate orders with similar times and compatible locations to form candidate departure tasks; S5: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the candidate departure tasks are divided into directions based on consistency; and based on the spatial detour boundary and time window boundary, obviously infeasible candidate tasks are pre-filtered to obtain the candidate departure tasks. S6: With the goal of maximizing the overall utility value of customized passenger transport services, construct an optimization model that includes total revenue, total operating costs, and total user-side time penalties; S7: Conduct passenger dispatching based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.

[0048] This embodiment constructs a spatiotemporal integrated distance metric that combines spatial and temporal distances, clusters customized passenger transport orders to form logical stations, and makes decisions to retain or remove isolated orders, thereby achieving adaptive aggregation and demand integration of large-scale orders. Furthermore, through precise mapping from virtual stations to real bus stations, it effectively reduces the problem scale of subsequent route optimization.

[0049] Example 2: This embodiment further discloses information based on Embodiment 1: Demand aggregation and dispatch task generation Obtain customized passenger order information and perform macro-level partitioning and preprocessing.

[0050] The system first retrieves all passenger order information for the day from the booking platform. For large cities, it divides the city into administrative districts, creating a 'district-to-district' intercity shuttle connection, such as... Figure 2 As shown.

[0051] Each order must include at least: pick-up point coordinates, drop-off point coordinates, scheduled departure time, and number of passengers. The preprocessing stage cleans the order data, removing invalid orders with abnormal coordinates or missing times, and merging duplicate orders. Simultaneously, a direction feature vector (dir_x, dir_y) is generated for each order for subsequent direction-aware task decomposition. Orders that are geographically close and occur at similar times are grouped into "logical sites," such as... Figure 3 As shown, this is to reduce the scale of subsequent scheduling issues.

[0052] To avoid excessively large time spans due to relying solely on spatial distance, a metric that integrates spatial and temporal distance is designed, defined as:

[0053] in, It is the geographical distance between the pick-up points of order i and order j. The absolute difference between the two appointment times. , As the normalization factor, , These are weighting coefficients, which can be fine-tuned through closed-loop adjustments using historical data.

[0054] Use the Canopy algorithm to quickly partition the demand pool. Set two distance thresholds. If the order With a certain cluster center The distance is less than If the value is less than T2, it is added to the cluster's demand pool; otherwise, it is directly used as a core member of the cluster. This mechanism can significantly reduce the computational cost of subsequent K-Means iterations and avoid infinite loops.

[0055] For orders that fail to enter any demand pool, the system marks them as candidate orphan orders. An "orphan acceptance coefficient" Ωi is introduced to weigh the additional detour costs and benefits of accepting this orphan order:

[0056] in, It refers to the number of orders. This is the expected additional detour distance to accommodate this isolated point. , It is the adjustment coefficient.

[0057] when If the isolated point is selected, it will be retained as an independent demand unit for subsequent scheduling; otherwise, it will be discarded to ensure operational economy.

[0058] Within each Canopy demand pool, the K-Means++ algorithm is used for refined clustering to generate the final logical sites. The optimization objective is to minimize the intra-cluster squared error.

[0059] in, It is the geometric centroid of the cluster, serving as the coordinates of the virtual site.

[0060] The real-site mapping uses the generated logical site centroid as the virtual site and leverages a KD-tree to quickly retrieve nearby real bus stops. After initial screening, the map service interface is further called to calculate the actual road network distance, and the real site with the smallest road network distance is selected as the mapping result, such as... Figure 4 As shown. This mapping result will be used for subsequent route planning. The final selected optimal mapping real-world station. Definition:

[0061] in, It is a set of candidate real bus stops retrieved using a KD-tree. From virtual site To the real site .

[0062] The rolling time-domain shift aggregation step aggregates logical stations with similar times and spatial compatibility into candidate departure tasks.

[0063] Sort all virtual sites in ascending order of average departure time.

[0064] Set a time window threshold (e.g., 1800 seconds, or 30 minutes) to merge station demands with time differences within the threshold into the same shift pool. Figure 5 As shown, temporal rolling clustering links stations with similar times together as candidate departure tasks.

[0065] Remove stations that have already been assigned orders and verify whether the total number of passengers for the trip meets the minimum departure requirement (e.g., = 5), to ensure that the shifts have basic operational profitability.

[0066] To avoid detours caused by mixing orders from different travel directions, the direction-aware task is split into candidate tasks based on the origin, destination, and travel direction of the order.

[0067] Extract the latitude and longitude coordinates and direction vector for each order to construct a four-dimensional feature vector. The direction vector is obtained by normalizing the difference between the origin and destination coordinates of the order, denoted as . .

[0068] The feature vector of each order is defined as:

[0069] in, , These are the latitude and longitude coordinates of the order's pick-up point. This is the order travel direction vector (normalized). , These are weighting coefficients used to balance the importance of spatial location and orientation features.

[0070] Estimate the required number of vehicles based on the total number of people: Calculate the number of vehicles needed, K = [total number of people / capacity]. Perform K-classification in the four-dimensional feature space to assign "on-the-way" orders to the same vehicle.

[0071] For each clustering result, calculate the average direction of the order direction vector within the cluster, and verify the cosine value of the angle between each order and the average direction of the cluster:

[0072] in, It is the travel direction vector of order i. It is the direction vector of the current road segment traveled by vehicle k.

[0073] If the threshold is not met, the order is removed from the current cluster and enters the next round of clustering until the directional consistency of all orders meets the requirements.

[0074] Route search and departure decision To effectively address the "combinatorial explosion" problem that may occur in real large-scale road networks, heuristic rules are introduced to perform pre-pruning in the spatiotemporal dimensions before constructing the state transition diagram.

[0075] The geographical distance and driving direction angle between stations are calculated based on the Haversine formula. If the connection between pick-up and drop-off points deviates significantly from the main intercity route (such as the Guangzhou to Shenzhen direction), the path is directly pruned. The calculation formula and threshold are based on the cosine formula in step 5, S4.

[0076] Verify the estimated travel time between any two nodes. Let the estimated arrival time of node j be... Passengers' expected boarding time window is ,in The maximum tolerable waiting time is set to 10 minutes. If the shortest arrival time of node j exceeds the passenger's tolerable extreme waiting time window, the weight of that edge is directly set to infinity in the state space matrix to block the generation of invalid state branches from the source.

[0077] For the shift task pool, this stage aims to maximize the overall service utility value U, using state-compressed dynamic programming (SC-DP) to find the globally optimal physical path. Since intercity transportation has a distinct characteristic of "pick-up and drop-off at both ends, and high-speed transport in the middle," the algorithm is deconstructed into a pick-up phase and a drop-off phase for segmented optimization. Specifically, step 7, the service utility optimization based on state-compressed dynamic programming, includes the following steps: Phased incremental utility actuarial calculation: The algorithm will transfer the entire path. It can be decomposed into two ordered subsequences. Let Starting point city logical station set The set of logical stations for the destination city.

[0078] (1) Pick-up phase: Starting from point o, in The optimal access order is found within the middle, ending at the high-speed entrance. The state transition is denoted as:

[0079] (2) Drop-off stage: from the highway exit Departure, at The optimal traversal order is achieved by searching through the middle section and ending at the destination. The state transition is denoted as .

[0080]

[0081] (3) Intercity connection point: The final state of the receiving segment is used as the initial cost of the sending segment, that is:

[0082] in The cost of traveling on intercity expressways (including road and bridge tolls and fuel consumption).

[0083] Core recursive mechanism: To reduce computational complexity, a single objective function U is directly embedded, employing utility increments. U is accumulated: (1) Single objective function U: maximizing the overall utility of the service (total revenue) - Total operating costs (fixed costs) + Variable costs Time window penalty cost The target is defined as:

[0084] Expanded to:

[0085] in: It is the unit price of order r. It refers to the number of orders. It is the fixed operating cost per vehicle. It refers to the number of vehicles dispatched. It is the variable cost per kilometer. To maintain costs, For road and bridge tolls, It is the travel distance from station i to station j. Is vehicle k from Drive to Decision variables, It is the time window penalty weight coefficient. The mixed time window penalty function is defined as follows:

[0086] in Penalty coefficients for arriving early and late.

[0087] (2) Single-step utility increment, for each additional station It only requires calculating the new revenue of the station minus the variable costs and time window penalty costs of the current road segment.

[0088] in, It is the total ticket revenue of passengers boarding at station i. It is the variable cost per unit distance. It is the travel distance from station j to station i. It is the time window penalty coefficient. This is the time penalty value caused by the arrival time deviation.

[0089] (3) Global state transition: using a bitmask to record the set of sites served (mask), and maximizing the overall global service utility U through incremental accumulation.

[0090] in, It is the last site visited. It was the last site visited.

[0091] Constraint-Driven Pruning: After pre-pruning, all paths entering this stage satisfy the basic time window boundaries for passengers. The following constraints are used to further optimize waiting deviations within the allowable range. During state transitions, constraints such as passenger waiting time windows, detour coefficients, vehicle passenger capacity, order uniqueness service, and vehicle entry / exit balance are embedded into the search process. Paths exceeding the limit tolerance time window are directly pruned, while waiting deviations within the allowable range are optimized through the time window penalty cost in the objective function. These constraints include the following: (1) Passenger waiting time window constraint: The time tj when the vehicle arrives at station j must satisfy the passenger patience boundary. Defined as:

[0092] in, This is the maximum tolerable waiting time (e.g., 10 minutes).

[0093] (2) Detour coefficient constraint: Define the ratio of passenger i's actual travel distance to the optimal distance to ensure that the detour is within an acceptable range. Defined as:

[0094] in, It is the maximum detour coefficient (usually taken as 1.2 to 1.5).

[0095] (3) Combined constraints of passenger capacity and minimum departure number: upper limit of single vehicle capacity and bottom line of profitability:

[0096] in, This is the minimum number of passengers required for departure (e.g., 5 people). This refers to the vehicle's maximum passenger capacity (e.g., 10 people).

[0097] (4) Order Uniqueness Service Constraint: Ensure that each accepted order is served only once and that vehicles are not dispatched repeatedly.

[0098] (5) The departure strategy constraint based on the complementarity of fixed and flexible departures is specifically defined as follows: Triggering Merging into Fixed Shifts (Flexible to Fixed): If a fixed shift exists Its preset departure time is And satisfy:

[0099] in, If it is a time window threshold (such as 30 minutes), then the order will be merged into a fixed shift.

[0100] Trigger the creation of flexible shifts (when fixed shifts are mismatched): For all This situation opens up a new flexible schedule set for this demand. Its preset departure time To become an optimal decision variable, the order itself must satisfy a flexible time window:

[0101] in:

[0102] The earliest booking time in the current task pool is used as the time base for flexible shifts.

[0103] (6) Flow direction deviates from the upper limit:

[0104] in, It is the travel direction vector of order i. It is the direction vector of the current road segment traveled by vehicle k.

[0105] like Figure 6 As shown, this invention generates an optimized path from the starting city to the ending city through a two-layer SC-DP architecture.

[0106] The determination of optimal utility value and the output of the solution include the following steps: Capacity splitting: If the number of people in the task pool exceeds the passenger capacity Q (10 people), the algorithm will split according to... The principle is to automatically select the optimal subset of orders and prioritize vehicle dispatch, thereby achieving task splitting.

[0107] Departure command triggered: When the overall service utility value corresponding to the optimal path is... This indicates that the overall utility of the plan is positive, meaning that ticket revenue is sufficient to cover operating costs and service quality penalties, thus justifying its operation. If the departure is cancelled, the departure will be cancelled to avoid operating losses or substandard service quality.

[0108] To verify the effectiveness of this embodiment, experiments were conducted on a simulated intercity customized passenger transport dataset containing 100 orders, and compared with two baseline methods: K-Means+SC-DP and DBSCAN+Time Window+SC-DP. The experiment used service utility value, order acceptance rate, average detour coefficient, and computation time as evaluation indicators. Key parameters were set as follows: vehicle passenger capacity of 10 people, minimum departure passengers of 5 people, maximum tolerable waiting time of 10 minutes, and maximum detour coefficient of 1.3.

[0109] The experimental results are shown in Table 1, demonstrating that the method of this invention significantly outperforms the two existing baseline methods in improving the utility of customized passenger transport services. Specifically, in terms of the overall service utility value, the method of this invention reaches 3797.93 yuan, a 29.4% improvement compared to the K-Means+SC-DP method and a 19.7% improvement compared to the DBSCAN+Time Window+SC-DP method. Regarding the order acceptance rate, the method of this invention reaches 63.0%, which is 18% and 15% higher than the two baseline methods, respectively, proving that the spatiotemporal integrated distance metric and isolated point acceptance coefficient proposed in this invention can effectively aggregate more valuable orders. In terms of computational efficiency, the method of this invention requires only 0.0233 seconds, a reduction of 80.8% and 20.0% compared to the two baseline methods, respectively, verifying the effectiveness of Canopy coarse clustering and spatiotemporal boundary pre-pruning mechanism in reducing the state space size. The average detour coefficients of the three methods are all within a reasonable range of 1.04 to 1.07, indicating that this invention does not introduce unacceptable detours while improving utility and order acceptance rate.

[0110] Table 1 Performance Comparison of Different Customized Passenger Dispatch Methods

[0111] Example 3: This embodiment provides, as follows: Figure 7 The intercity customized passenger transport scheduling system shown integrates spatiotemporal coupled clustering and state compression dynamic programming, including: Acquisition Module: Retrieves customized passenger transport order information; Logical site construction module: Aggregates order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; constructs a spatiotemporal comprehensive distance metric that integrates spatial distance and temporal distance to cluster orders and form logical sites; Mapping module: Using the centroid of the logical station as the virtual station, it retrieves candidate bus stops and, combined with the distance calculation of the map road network, completes the matching and mapping between the virtual station and the real station; Candidate departure task module: Based on customized passenger transport order information, a rolling time window is constructed to aggregate orders with similar times and compatible locations to form candidate departure tasks; Pre-filtering module: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the module classifies candidate departure tasks according to directional consistency; and based on spatial detour boundaries and time window boundaries, it pre-filters obviously infeasible candidate tasks to obtain the candidate departure tasks. Optimization module: With the goal of maximizing the overall utility value of customized passenger transport services, an optimization model is constructed that includes total revenue, total operating costs, and total user-side time penalties; Scheduling module: Performs passenger transport scheduling based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.

[0112] This embodiment constructs a spatiotemporal integrated distance metric that combines spatial and temporal distances, clusters customized passenger transport orders to form logical stations, and makes decisions to retain or remove isolated orders, thereby achieving adaptive aggregation and demand integration of large-scale orders. Furthermore, through precise mapping from virtual stations to real bus stations, it effectively reduces the problem scale of subsequent route optimization.

Claims

1. A method for intercity customized passenger transport scheduling that integrates spatiotemporal coupled clustering and state compression dynamic programming, characterized in that, include: S1: Obtain customized passenger transport order information; S2: Aggregate order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; A spatiotemporal integrated distance metric that integrates spatial and temporal distances is constructed to cluster orders and form logical sites; S3: Using the centroid of the logical station as the virtual station, retrieve candidate bus stops and combine the distance calculation of the map road network to complete the matching and mapping between the virtual station and the real station; S4: Based on the customized passenger transport order information, a rolling time window is constructed to aggregate orders with similar times and compatible locations to form candidate departure tasks; S5: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the candidate departure tasks are divided into directions based on consistency; and based on the spatial detour boundary and time window boundary, obviously infeasible candidate tasks are pre-filtered to obtain the candidate departure tasks. S6: With the goal of maximizing the overall utility value of customized passenger transport services, construct an optimization model that includes total revenue, total operating costs, and total user-side time penalties; S7: Conduct passenger transport scheduling based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.

2. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 1, characterized in that, In step S2, specifically: S2.1: Construct a spatiotemporal integrated distance metric that integrates spatial and temporal distances, and use historical data to perform closed-loop fine-tuning of scheduling parameters. The spatiotemporal integrated distance metric is defined as follows: in, It is an order With orders Geographical distance between boarding points The absolute difference between the two appointment times. , As the normalization factor, , The weighting coefficients are fine-tuned using historical data in a closed-loop manner. S2.2: Canopy coarse clustering: Utilizes T1 and T2 thresholds to quickly define the demand pool, reducing computation and avoiding infinite loops. It is defined as: in, ; S2.3: After the demand pool is constructed, orders that fail to enter any demand pool are judged as candidate isolated orders. An isolated point acceptance coefficient is introduced to quantitatively weigh the additional detour cost of connecting to isolated points and the number of orders. And calculate the isolated point acceptance coefficient, defined as: in, It refers to the number of orders. This is the expected additional detour distance to accommodate this isolated point. , It is the adjustment coefficient; when If the isolated point is selected, it will be retained as an independent demand unit for subsequent scheduling; otherwise, it will be discarded. S2.4: Find the geometric centroid μ within each Canopy and generate virtual site coordinates, defined as: in, It is the geometric centroid of the cluster, serving as the coordinates of the virtual site.

3. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 1, characterized in that, In step S4, specifically: S4.1 Time-series sorting: Sort all virtual sites in ascending order of average departure time; S4.2 Dynamic Window Consolidation: One-dimensional scrolling clustering is performed using time_window to merge the demand of all stations within a certain time range into the same shift pool; S4.3 Constraint Filtering: Remove assigned orders and verify them. This ensures that the shifts are profitable.

4. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 1, characterized in that, In step S5, based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the candidate departure tasks are divided into categories according to directional consistency, specifically as follows: S5.1 Four-Dimensional Feature Coupling: Extracting order latitude and longitude coordinates and preprocessing them. Direction vector; S5.2 Feature Matrix Construction Definition: in, , These are the latitude and longitude coordinates of the order's pick-up point. It is the order travel direction vector. , These are weighting coefficients used to balance the importance of spatial location and orientation features; S5.3K-Means++ Task Splitting: Calculate the number of vehicles to be dispatched, K = [total number of people / capacity]; perform K-classification in the four-dimensional feature space to assign orders that are going the same way to the same vehicle.

5. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 1, characterized in that, In step S5, based on the spatial detour boundary and the time window boundary, obviously infeasible candidate tasks are pre-filtered to obtain the candidate departure tasks, specifically as follows: S5.1 Spatial Detour Prevention and Removal: Based on the Haversine formula and driving direction angle, the spatial topological relationship between stations is evaluated in advance; if the connection of the pick-up and drop-off point sequence deviates significantly from the intercity trunk line, it is directly pruned. S5.2 Time Window Boundary Blocking: Verify the estimated travel time between any two nodes; if the node If the shortest arrival time exceeds the passenger's tolerable extreme waiting time window, then the weight of that edge in the state space matrix is ​​directly set to infinity to block the generation of invalid state branches from the source.

6. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 1, characterized in that, In step S6, specifically: S6.1 Phased Incremental Utility Actuarial Calculation: The algorithm will transfer the entire path. Decompose into two ordered subsequences; let Starting point city logical station set For the destination city logical station set; Pick-up phase: Starting from point o, in The optimal access order is found within the middle, ending at the high-speed entrance. The state transition is denoted as: Drop-off phase: From the highway exit Departure, at The optimal traversal order is achieved by searching through the middle section and ending at the destination. The state transition is denoted as: Intercity connection point: The final state of the receiving segment serves as the initial cost of the sending segment, i.e.: in The cost of travel on intercity expressways; S6.2 Core Recursive Mechanism: To reduce computational complexity, a single objective function U is directly embedded, employing utility increments. U is accumulated: Single objective function U: Maximizing the overall utility of the service (total revenue) - Total operating costs (fixed costs) + Variable costs Time window penalty cost The target is defined as: Expanded to: in: It is the unit price of order r. It refers to the number of orders. It is the fixed operating cost per vehicle. It refers to the number of vehicles dispatched. It is the variable cost per kilometer. To maintain costs, For road and bridge tolls, It is a website to station driving distance, Is vehicle k from Drive to Decision variables, It is the time window penalty weight coefficient. The mixed time window penalty function is defined as follows: in Penalty coefficients for arriving early and late; Single-step utility increment, for each additional site It is only necessary to calculate the new revenue of the station minus the variable costs and time window penalty costs of the current road segment, defined as: in, It is the total ticket revenue of passengers boarding at station i. It is the variable cost per unit distance. It is a website to station driving distance, It is the time window penalty coefficient. This is the time penalty value caused by the arrival time deviation; Global state transition, using bitmasks to record the set of sites already served. The overall service utility U is maximized through incremental accumulation, defined as: in, It is the last site visited. It is the last site visited; S6.3 Constraint-Driven Algorithm Pruning: During the state transition process, constraints such as passenger waiting time window, detour coefficient, vehicle passenger capacity, order uniqueness service, and vehicle entry and exit balance are embedded into the search process; among them, paths exceeding the limit tolerance time window are directly pruned, while waiting deviations within the allowable range are optimized through the time window penalty cost in the objective function.

7. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 6, characterized in that, The constraint-driven pruning algorithm in S6.3 is as follows: Passenger waiting time window constraint: the time when the vehicle arrives at station j Passengers' patience boundaries must be met, defined as: in, This is the maximum tolerable waiting time; Detour coefficient constraint: Define passenger The ratio of the actual driving distance to the optimal distance ensures that the detour is within an acceptable range; defined as: in, It is the maximum detour coefficient; The combined constraint of passenger capacity and minimum departure number: the upper limit of single-vehicle capacity and the bottom line of profitability are defined as follows: in, This is the minimum number of passengers required for departure. This is the vehicle's maximum passenger capacity; Order uniqueness service constraint: Ensures that each accepted order is served only once, without duplicate vehicle dispatch, defined as: Constraints based on a departure strategy that combines fixed and flexible elements; Flow direction deviates from the upper limit: in, It is an order The travel direction vector, It is the direction vector of the current road segment traveled by vehicle k.

8. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming as described in claim 6, characterized in that, The specific constraints of the departure strategy based on the complementarity of fixed and flexible mechanisms are as follows: Triggering integration into a fixed shift: If a fixed shift exists Its preset departure time is And satisfy: in, If it is a time window threshold, then the order will be merged into a fixed shift; Trigger the creation of flexible shifts: for all This situation opens up a new flexible schedule set for this demand. Its preset departure time To become an optimal decision variable, the order itself must satisfy a flexible time window: in: The earliest booking time in the current task pool is used as the time base for flexible shifts.

9. The intercity customized passenger transport scheduling method integrating spatiotemporal coupled clustering and state compression dynamic programming according to claim 1, characterized in that, In step S7, specifically: S7.1 Capacity Splitting: If the number of people in the task pool exceeds the passenger capacity Q, according to... The principle is to automatically select the optimal subset of orders and prioritize vehicle dispatch, thereby achieving task splitting; S7.2 Departure command triggered: When the overall service utility value corresponding to the optimal path is... This indicates that the overall utility of the plan is positive, meaning that ticket revenue is sufficient to cover operating costs and service quality penalties, thus justifying its operation. If the departure is cancelled, the departure will be cancelled to avoid operating losses or substandard service quality.

10. An intercity customized passenger transport scheduling system integrating spatiotemporal coupled clustering and state compression dynamic programming, characterized in that, include: Acquisition Module: Retrieves customized passenger transport order information; Logical site construction module: Aggregates order demand based on spatiotemporal coupling clustering according to customized passenger transport order information; A spatiotemporal integrated distance metric that integrates spatial and temporal distances is constructed to cluster orders and form logical sites; Mapping module: Using the centroid of the logical station as the virtual station, it retrieves candidate bus stops and, combined with the distance calculation of the map road network, completes the matching and mapping between the virtual station and the real station; Candidate departure task module: Based on customized passenger transport order information, a rolling time window is constructed to aggregate orders with similar times and compatible locations to form candidate departure tasks; Pre-filtering module: Based on the origin and destination areas and mainstream travel directions in the customized passenger transport order information, the module classifies candidate departure tasks according to directional consistency; and based on spatial detour boundaries and time window boundaries, it pre-filters obviously infeasible candidate tasks to obtain the candidate departure tasks. Optimization module: With the goal of maximizing the overall utility value of customized passenger transport services, an optimization model is constructed that includes total revenue, total operating costs, and total user-side time penalties; Scheduling module: Performs passenger transport scheduling based on the candidate departure tasks and an optimization model that includes total revenue, total operating costs, and total user-side time penalties.