Mixed scheduling method for airport shuttle vehicles considering driver behavior uncertainty

Abstract

This invention proposes a hybrid scheduling method for airport shuttle buses that considers driver behavior uncertainty, comprising: Step 1, collection and preprocessing of operational information, including flight dynamics, driver information, and shuttle bus information; Step 2, generating a candidate task-driver set for each service flight number based on flight dynamics and driver information; the candidate task-driver set includes task entries for each driver; Step 3, solving for driver uncertainty indices, including task acceptance probability and short-window response reliability index; wherein, the task acceptance probability is used for ranking in Step 4; the short-window response reliability index is used for unified automatic allocation in Step 4; Step 4, generating scheduling instructions based on the hybrid scheduling mechanism. This invention improves the on-time completion rate and service stability of airport shuttle buses.

Description

Technical Field

[0001] This invention belongs to the field of airport ground transportation and operation support scheduling technology, and in particular relates to a hybrid scheduling method for airport shuttle vehicles that takes into account the uncertainty of driver behavior. Background Technology

[0002] With the continuous growth of air traffic and the expansion of hub airports, the demand for passenger shuttle services between terminals and remote gates is high-frequency and highly volatile. Airport shuttle operations differ from general urban transportation; the airside has strict access boundaries, geofencing, and temporary controls, and vehicles and drivers must meet the corresponding area permissions and qualifications. Meanwhile, shuttle tasks are usually tied to critical moments such as flight takeoffs and landings, gate docking / removal, and remote gate deployment, with clear service time windows and high punctuality requirements; multiple vehicles often need to serve the same group of passengers concurrently. Operational disturbances such as flight delays, gate adjustments, and temporary lockdowns frequently alter task timeliness and route feasibility, further increasing the difficulty of real-time scheduling.

[0003] Existing solutions mainly fall into two categories: one is a pure order-grabbing model borrowed from ride-hailing services, relying on real-time confirmation from the driver. This is simple to implement and has a certain degree of on-site acceptance, but it lacks explicit airside permission verification and multi-vehicle capacity constraints, and lacks centralized decision-making capabilities for high-priority and urgent tasks, making it difficult to guarantee time window compliance and overall on-time performance. The other is a centralized dispatching model based on rules or static optimization, which can uniformly allocate resources, but usually ignores the uncertainty of driver behavior (such as confirmation delays and preference differences due to fatigue). It lacks the ability to replan when task information changes rapidly, resulting in an overly rigid system and friction with on-site execution. Overall, existing technologies struggle to simultaneously address airside compliance, time window achievement, driver behavior uncertainty, and continuous optimization under operational disturbances. There is an urgent need for a hybrid dispatching framework that integrates airport-specific constraints and driver response characteristics to achieve real-time, explainable, and executable assignment of shuttle tasks. Summary of the Invention

[0004] The purpose of this invention is to provide a hybrid scheduling method for airport shuttle vehicles that takes into account the uncertainty of driver behavior. This method can improve the on-time completion rate and service stability of airport shuttle vehicles, reduce the risk of empty runs and delays, and improve driver acceptance and scheduling fairness.

[0005] A hybrid scheduling method for airport shuttle vehicles that considers the uncertainty of driver behavior includes the following steps: Step 1: Information collection and preprocessing.

[0006] Operational information includes: flight status, driver information, shuttle bus information, and airport flight area road network data.

[0007] Flight status is the real-time updated flight status, including aircraft type and standard passenger capacity, estimated arrival / departure time, parking stand allocation information, terminal allocation information, flight number, inbound / departure flights, and international / domestic flights; The allocation information includes a number and a location, where the location is a geographic coordinate. Driver information includes: driver's continuous working hours per session, rest time, and driver's historical order response time.

[0008] Single continuous working time: The time a driver continues to perform a task from the end of their last rest period until the current moment.

[0009] One driver is assigned to each shuttle bus.

[0010] The shuttle bus information includes: the shuttle bus's rated passenger capacity, the shuttle bus's location, and the shuttle bus's status.

[0011] Shuttle bus location: The current location of the shuttle bus; Shuttle bus status includes: the average speed of the shuttle bus in the work area where the driver is located. The remaining time of the currently executing task (if any), the current destination of the currently executing task (if any), the idle status, and the passenger-carrying status; Among them, idle status: the vehicle is empty and not performing a task; passenger-carrying status: the vehicle is performing a predetermined task. Airport flight area vehicle road network data: including the topological structure of road nodes and edges and their latitude and longitude coordinates; Preprocessing: Perform routine cleaning on abnormal or missing data, including duplicate record removal, standardization of time field format / time zone and anomaly correction, completion of missing fields according to preset completion rules, and standardization of coordinates and units to form standardized input; the completion rules include: using one or more of the default values ​​of the same flight / same parking position or inferred values ​​based on business constraints; Step 2: Based on flight dynamics and driver information, generate a candidate task-driver set for service flights. .

[0012] Each task (ferry task) includes: the number of vehicles required, the task start point, the task end point, the service time window, the task priority, and the task status.

[0013] Specifically, the following steps are included: Step 2A: Generate a task pool.

[0014] The task pool contains several tasks waiting to be executed.

[0015] Based on real-time updated flight information, tasks to be executed are dynamically generated; Task generation is specifically as follows: Map the field information in flight dynamics to task attributes; The field information includes: aircraft type, estimated arrival / departure time, parking stand allocation information for the flight, and terminal allocation information for the flight; Task attributes include: number of vehicles required, task start point, task end point, service time window, task priority, and task status.

[0016] The required number of vehicles is determined by matching the corresponding passenger capacity demand with the aircraft type. Specifically, the required number of vehicles is obtained by dividing the standard passenger capacity corresponding to the aircraft type by the rated passenger capacity of the shuttle bus and rounding up. Determine the start and end points of the mission by using the location of the aircraft's parking stand and terminal (for departing flights, the mission start point is the terminal location, and the mission end point is the aircraft's parking stand; the opposite is true for arriving flights). Service time windows are generated using estimated arrival / departure times of flights; for inbound flights, the estimated arrival time... Set the preset preparation time and window width Its service time window is set to ; Estimated departure time for departing flights Set preset advance time and window width Its service time window is set to ; Task priorities are obtained by using the current time and service time window, based on a priority calculation model. The priority calculation model is as follows: Assuming the current time ,Task The service window is ; Preset first time threshold With the second time threshold ,and > >0; Task priority is determined using a piecewise function model: ; in, It refers to the timeliness and urgency of the task;

[0017] - Weights are set based on flight information set for the task service; flight information includes: aircraft type, inbound / departure flights, and international / domestic flights; and and These are weight parameters, and ; ; Task status, such as: pending assignment, candidate, assigned, executing, completed; Step 2B: Filter drivers to obtain a candidate task-driver set. .

[0018] Based on the shuttle bus status, time window feasibility and path reachability checks are performed on tasks in the task pool to form each task. Corresponding driver A set of.

[0019] The time window feasibility and path reachability verification are specifically as follows: When the current time is detected The task has been exceeded. Latest start time, i.e. At that time, the latest start time of the task Updated to ,in, Preset grace threshold; Combined with driver Calculate arrival time from the current location of the shuttle bus Remove those that cannot be served within the specified service time window. Drivers who arrive at the task starting point within the time frame generate a candidate task-driver set; Specifically: combining the driver Calculate arrival time from the current location of the shuttle bus Further calculation of the driver Estimated arrival time: If the conditions are met Then it is believed that the driver If feasible within the time window, retain it as a candidate; otherwise, consider it unable to reach the task origin within the service time window, and discard the "driver-task" candidate pair. , Removed from the candidate set, the remaining drivers are those relevant to the task. Available drivers; The arrival time calculation model is as follows: driver When the corresponding shuttle bus is idle: ; in Based on the airport flight area vehicle road network, the current position of the vehicle is obtained. To the starting point of the mission The distance of the feasible path. The average speed of the shuttle bus corresponding to the driver's work area; driver When the corresponding shuttle bus is in passenger mode: ; in This represents the remaining time for the shuttle bus to complete its current task. This is the endpoint of the currently executing task; The feasible path distance This was calculated using a path planning model.

[0020] The path planning model (Dijkstra's algorithm or A* algorithm) takes the topology of the airport flight area vehicle road network (nodes, edges and their weights / costs, where edge weights can be obtained by calculating the edge length from the node's latitude and longitude or by a preset toll cost) as input, and uses Dijkstra's algorithm or A* algorithm to output a feasible path connecting the starting point and the ending point, and obtains the corresponding feasible path distance. .

[0021] All the "drivers" -Task This mapping relationship ( , The set of candidate tasks and drivers, i.e., the candidate task-driver set. .

[0022] Candidate Tasks - Driver Collection It can be represented as ( , The set of ) for each driver The set of acceptable tasks, denoted as .

[0023] Mapping relationship ( , ), that is, candidate pairs.

[0024] Task Pool and Candidate Tasks - Driver Collection The system updates based on flight status, shuttle bus status, driver information, and time progress.

[0025] Step 3: Solve for the driver uncertainty index. This includes: 1. Solve for the task acceptance probability .

[0026] It represents the driver's preference for different tasks.

[0027] Used for sorting and display on subsequent driver mobile terminals.

[0028] Based on the shuttle bus's current location, task origin, driver's continuous working time per trip, and task service time window collected in step 1, the task acceptance probability is calculated. .

[0029] Task acceptance probability Based on weighted utility function The calculation yielded: ; Linear weighted model: ; or Log-linear model: ; in, -Estimated travel time from the current location of the shuttle bus to the starting point of the mission; -Task priority; - Driver fatigue; Index for drivers, For task indexing, - Drivers can accept a collection of tasks, i.e., task entries; - The task index in the task entry; This is the utility weighting coefficient. They are not both 0.

[0030] This probability value will be used for the interface sorting on the driver's mobile terminal. The task with the highest value is displayed at the top, thus intelligently guiding drivers to make choices that are more likely to be accepted by the system, improving the efficiency of the order-grabbing mode and the overall coordination of the system.

[0031] Among them, driver fatigue The calculation model is as follows: Based on the driver's single continuous working time, it is determined using a piecewise linear function: ; in, -driver The duration of continuous operation in a single session; and and For the slope parameter, , and For two inflection point thresholds, .

[0032] If a rest period exceeds the preset effective rest threshold during the course. (For example, 15 minutes), then Reset to zero and start accumulating again.

[0033] That is, based on utility function By combining factors such as arrival time, task priority, and driver fatigue, the candidate tasks visible to the same driver are scored and normalized to obtain a personalized task acceptance probability sequence.

[0034] 2. Calculate the driver Short window response reliability index .

[0035] Used to determine if the driver is in a position with a length of [length missing]. Whether the order can be confirmed and accepted in a timely manner within a short time window.

[0036] Characterizes the driver In length The probability of completing order confirmation within a short time window is used to measure the reliability of driver response timeliness during the unified automatic allocation in step 4. Based on the driver's historical order acceptance response time collected in step 1, calculate the short-window response reliability index. .

[0037] ; ; - Decision-making time window, i.e., short time window; - The cumulative distribution function, whose numerical representation is 'in The probability that the driver has completed the order acceptance confirmation.

[0038] It is the cumulative distribution function of the standard normal distribution; It is the Gaussian error function; and They are drivers The logarithmic mean and logarithmic standard deviation of historical order response times are obtained by fitting historical data.

[0039] Figure 2 In Chinese, "short window response reliability" refers to... .

[0040] First, a log-normal distribution is fitted to the driver's historical order acceptance response time (the driver's historical response time for order confirmation). and Then, the short-window response reliability index was calculated. .

[0041] For distribution parameters ( and The calculation method for online calibration using the sliding window approach is as follows: Assuming the driver recent The sample of drivers' historical order acceptance response times is as follows: The sliding window size is ; Calculate each sample logarithmic response time Then, the parameters of the log-normal distribution for response delay estimation... and You can update online as follows: ; ; Step 4: Generate scheduling instructions based on the hybrid scheduling mechanism.

[0042] The system employs a combined mechanism of "normal autonomous order grabbing + automatic allocation due to conflicts and timeouts + smooth order matching + optimized scheduling for emergencies + manual intervention interface": (1) Regularly accepting orders independently: The tasks are displayed to each driver's mobile terminal and sorted in a personalized manner according to the task acceptance probability obtained in step 3. Drivers can choose independently provided that the task requirements are met.

[0043] Specifically, a candidate task-driver set is generated from the task pool. In the middle, the task items are displayed to each driver's mobile terminal. And based on the task acceptance probability Personalized sorting allows drivers to choose tasks independently, provided the task requirements are met.

[0044] For the required number of vehicles Task During the decision-making window There is no limit to the number of concurrent driver confirmations to form a candidate pool, but the final number of drivers actually assigned is the maximum. name; The number of concurrent driver confirmations refers to the number of times a driver confirms a decision within a pre-defined decision time window. Internally, for the same task The total number of drivers who bid for the order; any number of drivers are allowed to bid for the same task, but the number of drivers ultimately assigned will not exceed the number of vehicles required for the task.

[0045] Decision-making window Within the system, a single driver is only allowed to confirm one task. If a driver initiates confirmation for multiple tasks, the first task confirmed will be considered the valid confirmation, and the remaining confirmations will be automatically revoked and regarded as unconfirmed.

[0046] (2) Automatic allocation of conflicts and timeouts: Construct a set of decision-making tasks With decision candidate set Decision-making time window After the mission is completed, By comparing the number of concurrent confirmations Number of vehicles required for the mission Tasks that meet any of the following conditions Included in decision-making task set : when In other words, if the number of concurrent confirmations for the same task exceeds the required number of vehicles, it is considered a dispute over order submission, and the task will be reassigned. Included To perform this task With the clicked task confirmed Candidate pairs consisting of drivers were included in the decision candidate set. Call the unified automatic allocation method from Optimize driver assignments among top drivers; when In other words, if the number of concurrent confirmations for the same task is less than the required number of vehicles, it is considered a timeout for order grabbing, and the tasks that have already been confirmed will be prioritized. A driver is selected and a dispatch instruction is issued directly. The selected driver is removed from the set of candidate drivers automatically assigned in this round. Then, for the remaining ( There are 10 vacancy slots available, and the task will be completed. Included , will the task With sets The candidate pairs consisting of drivers who did not participate in the order-grabbing process were selected and included. The unified automatic allocation method is invoked for optimized assignment. Specifically: during the decision-making time window After that, the task was detected. When the above situations exist, construct a decision task set. With decision candidate set The unified automatic allocation method is called to optimize allocation. For each "driver-task" candidate pair, the allocation decision is made by comprehensively considering arrival time, task priority, driver fatigue, and short window response reliability indicators, and driver dispatch instructions are generated. The dispatch instruction is to send the driver's mobile terminal the task start point, task end point and service time window requirements for the shuttle service. Unified automatic allocation method: ; The objective function maximizes the total expected benefit, combining the probability of a successful assignment with the expected benefit after a successful assignment. For binary assignment variables; Characterizes the driver In length The probability of completing order confirmation within a short time window is used to measure the reliability of the driver's timely response; It represents the overall utility that a successful assignment can bring about; Multiplying the two together gives the expected benefit of this assignment. By maximizing the total expected benefit, a balance between scheduling stability and efficiency can be achieved. The constraints are: (Task capacity constraints) (Driver capacity constraints) ; in Index for drivers, For task indexing; The decision task set refers to the set of tasks that require a unified automatic allocation method to schedule drivers. For decision candidate set; This is a binary assignment variable used to represent the driver. With the task The matching state between them; its value rules are as follows: when When, it indicates that the system decision will be based on the task. Assigned to driver Execute; when At that time, it indicates the driver Not assigned to perform tasks ; For the task The upper limit of the required vehicle capacity; for tasks with order-grabbing conflicts or other sudden scheduling issues. For tasks that exceed the timeout period, ; For the driver The number of tasks that can be accepted simultaneously; ; The utility function has already taken into account factors such as arrival time, task priority, and driver fatigue. For decision-making time window; This is a reliability indicator for short-window response.

[0047] The unified automatic allocation method only sets The above proceeds and satisfies Candidate set The feasibility and accessibility of the time window have been verified, thus ensuring that the assigned driver meets the start and end points and time window requirements of the task.

[0048] (3) Successful order matching: Decision-making time window After the mission is completed, ,when If the number of concurrent confirmations for the same task equals the number of vehicles required, the order is considered successfully matched, and the task is directly assigned to the previously confirmed users. Dispatch instructions are issued to the corresponding drivers; (4) Sudden Optimization Scheduling: When a significant change is detected in the task-related input data, the affected tasks and the set of available drivers (the affected tasks constitute the decision task set) are considered. From the set Extract the matching pairs between the affected tasks and their available drivers to form a decision candidate set. The system calls a unified automatic allocation method for scheduling. Situations where the input data related to the task has been significantly updated include: Task-related input data includes the task's service time window, task start point, and task end point.

[0049] (5) Manual intervention interface: Dispatchers can modify, cancel, or replace assignments in real time.

[0050] It provides a dispatcher interface to modify, cancel, or replace dispatch instructions generated by the unified automatic allocation method in real time, and records the intervention log.

[0051] like Figure 4 As shown, step 4 specifically includes the following steps: Step 4A: For each driver , and the corresponding task entries Task acceptance probability The message is sent to the driver's mobile device. according to The data is displayed in a sorted manner, allowing drivers to make decisions within their designated time window. Internally selects tasks; Decision-making window After completion, statistics for each task will be compiled. j Concurrent confirmations Number of vehicles required for the mission If a significant change is detected in the task-related input data, proceed to step 4B or step 4C; if a significant change is detected in the task-related input data, proceed to step 4D. Step 4B, Unified automatic allocation of tasks triggered by conflicts and timeouts: If for tasks j The task will be completed if any of the following conditions are met. Included in decision-making task set And construct a decision candidate set : Order-grabbing conflict: ; Order timeout: ; Among the decision candidate set The construction rules are as follows: when hour, Includes the "task" Candidate pairs consisting of "drivers who have already been confirmed for the task"; when At that time, first contact the party that has already confirmed the task. The dispatch instructions are sent from one driver to the corresponding driver's mobile terminal. Includes "task" With sets The candidate pairs, consisting of "drivers who did not participate in the order-grabbing process," are used to fill the gaps. There are 10 vacant positions. Subsequently, in the assembly The system calls a unified automatic allocation method, which comprehensively considers arrival time, task priority, driver fatigue, and short-window response reliability indicators to make allocation decisions, generate scheduling instructions, and send them to the corresponding driver's mobile terminal. The scheduling instruction includes at least the task start point, the task end point, and the service time window; Step 4C: Successful order matching leads to direct issuance of scheduling instructions: If the task... satisfy This is considered a successful order matching; a dispatch instruction is directly generated and sent to the corresponding driver's mobile terminal; Step 4D: When a significant change is detected in task-related input data or constraints (including task time window / start / end point updates), the candidate set consisting of the affected task and available drivers (from the set...) is... Extraction and reconstruction and And call the unified automatic allocation method for scheduling; Step 4E: The dispatcher can modify, cancel, or replace the dispatch instructions generated by the unified automatic allocation method through the manual intervention interface, and record the intervention log.

[0052] Step 5: Execution and Feedback.

[0053] Dispatch instructions are sent to the corresponding driver's mobile terminal and order confirmation is obtained. The task execution trajectory and status are tracked in real time through the driver's mobile terminal application.

[0054] When the driver's mobile terminal detects that the driver has been stuck for a long time / drifting or is expected to exceed the service time window of the task, it will trigger the unified automatic allocation described in step 4B. The driver's mobile terminal records data such as the actual response time of the task, arrival / start / completion time, and reassignment events, which are used to update uncertainty indicators online.

[0055] (1) Short window response reliability update: Based on the driver’s actual order acceptance response time and confirmation / rejection records, the response delay distribution parameters are updated using a sliding window to obtain the updated short-window response reliability index. (2) Online fatigue level update: Based on the actual continuous working time and rest time, the driver's fatigue level is corrected online and used for subsequent task sorting and automatic allocation; (3) Risk of being late is triggered: When a vehicle is detected to be stuck or deviating from its course for an extended period, or when a task has been assigned but is not expected to be completed within the service time window, the unified automatic assignment described in step 4B is automatically triggered.

[0056] Consistency and logging: The timestamps and parameters of the above updates and reassignments are recorded in the execution log.

[0057] The timestamps include the moment the task instruction is officially issued, the moment the driver accepts the order, the moment the vehicle actually arrives at the task start point, the moment the vehicle arrives at the destination, and the moment when the risk of lateness is detected and the unified automatic allocation is automatically triggered. The parameters include the driver's real-time fatigue level when the task occurs, continuous working time, short-window response reliability index when the task is assigned, task acceptance probability, and calculated comprehensive utility value. Compared with the prior art, the advantages of the present invention are: 1. A hybrid scheduling method and system for airport shuttle vehicles that considers the uncertainty of driver behavior (driver order-grabbing behavior) is provided. The system revolves around a closed-loop process of "task generation - candidate screening - order-grabbing participation - unified decision-making - execution feedback". During the task generation and screening stages, constraints such as service time windows and multi-vehicle capacity are comprehensively considered to ensure feasibility and compliance.

[0058] 2. In the proposed driver behavior uncertainty modeling method (step 3), by modeling the driver's historical order acceptance response time data, the response delay estimate and task acceptance probability are given, and personalized sorting display is realized on the driver's side, forming decision inputs such as short window response reliability on the system side.

[0059] 3. The hybrid scheduling mechanism adopts a hybrid strategy of "autonomous order grabbing + unified automatic allocation": the same task can receive any number of concurrent confirmations within the decision time window to improve participation and fairness, but the number of people assigned in the end will not exceed the number of vehicles required for the task; when task conflicts, task timeouts or significant updates to task information occur, the unified automatic allocation model will make centralized optimization decisions on the candidate set, thereby improving the overall completion rate and stability while ensuring high-efficiency tasks.

[0060] In summary, this invention, under the premise of satisfying multiple constraints, quantifies the uncertainty of driver behavior and adopts a hybrid scheduling mechanism of "normal autonomous order grabbing + automatic allocation of conflict and timeout + smooth order matching + emergency optimized scheduling + manual intervention interface" for shuttle bus scheduling. This helps to improve the on-time completion rate and resource utilization efficiency of shuttle tasks, reduce the risk of empty runs and delays, and take into account on-site acceptance and scheduling fairness. It is suitable for the operation support of large and medium-sized airports in complex airside environments. Attached Figure Description

[0061] Figure 1 A flowchart illustrating a hybrid dispatching method for airport shuttle vehicles that takes into account the uncertainty of driver behavior. Figure 2 Flowchart of input data and feature calculation for modeling uncertainty; Figure 3 This is a schematic diagram showing the task sorting and display on the driver's mobile terminal based on the task acceptance probability; Figure 4 This is a schematic diagram illustrating the operation mechanism of the hybrid scheduling mechanism; Figure 5 A schematic diagram of the framework of a hybrid dispatching system for airport shuttle vehicles that takes into account the uncertainty of driver behavior. Figure 6 A schematic diagram of the hardware deployment architecture of a hybrid dispatching system for airport shuttle vehicles that takes into account the uncertainty of driver behavior. Detailed Implementation

[0062] The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to the present invention will be described in more detail below with reference to the schematic diagrams, which illustrate preferred embodiments of the invention. It should be understood that those skilled in the art can modify the invention described herein while still achieving its advantageous effects. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the invention.

[0063] like Figures 1-6 A hybrid scheduling method for airport shuttle vehicles that considers driver behavior uncertainty includes the following steps: Step 1: Information collection and preprocessing.

[0064] Data collected at a certain moment: Flight status information: Flight number CA1234, aircraft type A330 (standard passenger capacity is 293), estimated departure time is 2026-01-08 10:35:00, gate / terminal is 203 / T1-G01, gate coordinates are (121.812, 31.155); this is a departing domestic flight. Shuttle bus and driver information: The shuttle bus has a rated passenger capacity of 150 people; driver D001 is located in Area A of Terminal 1 and is currently idle. The current location of the shuttle bus he is driving is (121.805, 31.148), and the average speed of his operating area is... The speed is 20km / h, the continuous working time is 180 minutes, and the rest time is 5 minutes; the historical order response time sequence is {2s, 8s, 5s, 4s, ...}.

[0065] Airport flight area vehicle road network data: including the topological structure of road nodes and edges, as well as the latitude and longitude coordinates of the nodes; The system performs the following standardization processing on the collected raw stream data: Example of missing field completion: If the "aircraft type" field of flight CA1234 is lost in the sensor transmission, the preprocessing unit matches the normal operating aircraft type of this flight number (such as A330) through the historical database, or infers that it is a wide-body A330 based on its parking position (large gate 203). Coordinate and time unification: The latitude and longitude coordinates collected by all sensors are uniformly converted to the WGS-84 coordinate system; the time is uniformly converted to Beijing time (UTC+8). Anomaly Correction (Cleansing): If GPS positioning shows that driver D001 moved 200 meters within 1 second (anomaly in instantaneous speed), the preprocessing unit will treat this record as noise and remove it, and will use the previous position and average speed. Perform linear interpolation correction.

[0066] Step 2: Generate a candidate task-driver set based on flight dynamics. .

[0067] Step 2A: Generate a task pool.

[0068] Tasks generated from flight dynamic information provided by external data sources.

[0069] For example, mission F001 provides a shuttle service from gate G01 to parking stand 203 for departing passengers of flight CA1234. The starting point of the departing flight mission is the gate location in the terminal building, and the destination of the mission is the parking stand location.

[0070] The required number of vehicles is calculated by dividing the standard passenger capacity of 293 people by the shuttle bus's rated passenger capacity of 150 people, which gives 1.95. Rounding up, we get 2 vehicles.

[0071] The service time window is generated as follows: This is a departing flight, and the estimated departure time is... Set a preset advance time for 2026-01-08 10:35:00. and window width Its service time window is set to = [10:00, 10:15].

[0072] Assuming the current time t If it is 09:55, then its remaining time =20 minutes Set to 30; Set to 10, It is 1.5. and and If we set them to 1, 0.1, and 1 respectively, then: ; That is, the priority of task F001 is 2.5.

[0073] Step 2B: Filter drivers to obtain a candidate task-driver set. .

[0074] The system combines drivers Calculate arrival time from the current location of the shuttle bus Further calculation of the driver Estimated arrival time: If the conditions are met Then it is believed that the driver If feasible within the time window, retain it as a candidate; otherwise, consider it unable to reach the task origin within the service time window, and discard the "driver-task" candidate pair. , Removed from the candidate set, the remaining drivers are those relevant to the task. Available drivers; Driver D001: Currently idle.

[0075] The average speed of the shuttle bus corresponding to the driver's work area At a speed of 20 km / h, based on the shortest feasible path distance of the road network. ,according to The estimated travel time from D001 to gate G01 is approximately 8 minutes, meaning the driver can arrive at gate G01 before 10:15. Verification passed; add to the candidate set.

[0076] Driver D002: Currently carrying passengers. The estimated time for completion of their current task and arrival at gate G01 is 10:10, earlier than the latest start time of task F001, 10:15. Verification passed; added to the candidate set.

[0077] Driver D003: Currently carrying passengers. The estimated time for completion of their current task and arrival at gate G01 is 10:30, which is later than the latest start time of task F001, 10:15. Verification failed; removed from the candidate pool.

[0078] All available drivers for task F001 are sequentially selected, ultimately forming a candidate set for task F001, such as the candidate driver set {D001, D002, D006, D007}. The corresponding candidate pair sets are then included. .

[0079] When the current time is detected It has exceeded its latest start time for the task, that is At that time, the latest start time of the task Updated to ,in Preset grace threshold; For example, if the original time window for task F005 was [09:35, 09:50], and the current time is 09:55, then the time window for task F005 will be updated to [09:35, 10:10], which is the preset grace threshold. Set to 15 minutes.

[0080] In addition, task pool and candidate set The task pool can be refreshed based on flight dynamics and changes in vehicle / driver status. For example, if flight CA1345 is delayed until 11:30 due to air traffic control, this event will immediately trigger a task pool refresh, and the service time window of the corresponding task F034 will be adjusted from the original [10:10, 10:25] to [10:55, 11:10]. For task F034, as the remaining time of the task decreases naturally, its priority will also be recalculated according to the piecewise function model.

[0081] Step 3: Solve for the driver uncertainty index.

[0082] 1. Solve for the task acceptance probability .

[0083] This embodiment uses a linear weighted model: ; For example, the task priority of task F001 calculated in step 2. The arrival time of driver D001 service task F001 is 2.5. It lasts for 8 minutes; Query driver D001's work record; their current single continuous working time is... =180 minutes, set the threshold for the driver fatigue calculation model. =240 minutes =360 minutes; slope =0.02, =0.05, =0.08; Calculate the fatigue level of driver D001: ; Calculate utility to : ; The system simultaneously calculates the utility value of driver D001 for its task entries (such as F002, F003). Let's assume they are 0.8 and 0.4 respectively.

[0084] Subsequently, normalization is performed using the Softmax function to obtain the normalized personalized probability of D001 accepting task F001: ; This probability value will be used to sort the driver's mobile terminal interface, placing the task with the highest acceptance probability, F001, at the top, thereby intelligently guiding the driver to make a choice that is more likely to be accepted by the system, improving the efficiency of the order-grabbing mode and the overall coordination of the system.

[0085] 2. Calculate the driver Short window response reliability index .

[0086] Based on the historical order-acceptance time series of drivers collected in step 1, the logarithmic mean and logarithmic standard deviation were fitted. =2, standard deviation =0.8, and the decision time window ∆t is 60 seconds; Therefore, the short-window response reliability index of driver D001 within the decision-making time window, i.e., the probability that he can confirm the task in a timely manner, is calculated as follows: ; This indicator is close to 1, indicating that when the system performs unified automatic allocation, the success rate of assigning driver D001 can be considered extremely high, and he is a reliable scheduling target.

[0087] In online calibration methods: The sliding window size is N=100.

[0088] Step 4: Hybrid scheduling.

[0089] According to the decision time window After the order-grabbing process ends, the following five scenarios will be handled: Scenario 1: Order-grabbing conflict ( ) Obtain the candidate driver set (including D001, D002, D006, and D007) for task F001 (serving flight CA1234) from the task pool, and push the task information to the mobile terminals of these drivers.

[0090] In driver D001's terminal interface, task F001 is displayed at the top and highlighted with "Recommended" because it has the highest acceptance probability.

[0091] Within the preset decision time window (∆t=60 seconds), drivers D001, D002, and D007 saw the task and pressed the "accept order" button, while driver D006 did not accept the order.

[0092] The system received their confirmation request.

[0093] For the required number of vehicles For task F001 with a value of 2, the system allows any number of drivers to concurrently bid for the order to encourage participation.

[0094] Within 60 seconds, drivers D001, D002, and D007 simultaneously accepted orders, and the number of concurrent confirmations (3 people) exceeded the required number of vehicles (2), triggering a conflict. Task F001 was then added to the order list. The task will be added to the decision candidate set along with the candidate pairs consisting of the three drivers who have already clicked to confirm the task. ,Right now Candidate pairs included: (D001, F001), (D002, F001), (D007, F001).

[0095] The system automatically invokes the unified automatic allocation method.

[0096] Calculate the three candidate drivers value: ; ; ; The system selects two drivers (D001 and D007) to win the bid using a unified automatic allocation method. At this point, the status of task F001 changes from "pending allocation" to "allocated," and a dispatch instruction is sent to their terminals. Driver D002 receives feedback that the bid was unsuccessful.

[0097] Scenario 2: Order grabbing timeout ( ) For example, task F010 requires 3 vehicles, and only driver D003 can accept the order within 60 seconds; at the same time, task F011 requires 2 vehicles, and only driver D004 can accept the order within 60 seconds; both tasks end their decision windows at the same time, and both are cases of order-accepting timeout. For task F010, this task is included The system locks driver D003 and issues a dispatch instruction directly to D003. This driver is removed from the candidate driver set for this round of unified automatic allocation and will no longer participate in the unified automatic allocation calculation for this round. Next, it addresses the remaining two gaps ( ), from the candidate set The candidate pairs of drivers who are eligible to perform the task but did not participate in the order-grabbing process (including D010, D011, and D012) and task F010 are included in the decision candidate set. ; For mission F011, this mission is also included. The system locks driver D004 and issues a dispatch instruction directly to D004. This driver is removed from the candidate driver set automatically allocated in this round. Then, it addresses the remaining 1 gap (…). ), from the candidate set The candidate pairs of drivers who did not participate in the order-grabbing process and who can perform the task (including D010, D011, D013, and D014) are selected and included in the decision candidate set for task F011. ; In summary, this round of unified automatic allocation , Decision candidate set

[0098] ; Based on a unified automatic allocation method, the system ultimately selects two drivers (D010 and D011) for task F010 and sends scheduling instructions to their terminals; for task F011, it selects one driver (D013) and sends scheduling instructions to its terminal.

[0099] Scenario 3: Successful order matching ( ) For example, if task F012 requires 3 vehicles, and exactly 3 drivers bid for the order within 60 seconds, then dispatch instructions will be issued to these 3 drivers immediately after the decision window ends.

[0100] Scenario 4: Sudden Optimization Scheduling.

[0101] For example, after task F001 has been assigned to D001 and D007, the system detects that the destination of task F001 has changed (from remote gate 203 to remote gate 121). This change causes the original route planning and time window verification to fail. This event causes a change in the existing set of driver candidates for task F001 (including D001, D002, D006, and D007), and the change in destination affects the path distance and arrival time. After recalculation, the original candidate drivers were re-selected into (D020, D021, D022), which meets the preset "sudden update" condition. Therefore, task F001 constitutes the decision task set. At the same time from the set Extract the matching pairs between task F001 and its available driver set (D020, D021, D022) to form a decision candidate set. : ; The system immediately suspends the normal order-grabbing mode for all affected tasks and invokes a unified automatic allocation method globally.

[0102] Using the available drivers and task information at the current moment as input, the optimal match is recalculated to schedule drivers D020 and D021 to execute task F001, while the original drivers D001 and D007 become idle.

[0103] During this process, the system may perform cross-regional resource reallocation to ensure that task F001 can be completed.

[0104] Scenario 5: Human intervention.

[0105] For example, if a dispatcher sees on the monitoring screen that the vehicle icon for driver D001 remains stationary at a certain point for an extended period, the dispatcher can confirm by speaking with the driver that the vehicle has a minor malfunction that cannot be repaired in the short term.

[0106] The dispatcher directly revoked the assignment relationship between D001 and task F001 through the dispatcher console, and manually replaced the assigned driver with a nearby available driver, D016, from the list of available drivers.

[0107] All details of this intervention (operator, time, original assignment, new assignment, reason) were recorded by the system and formed an intervention log for post-event analysis and system optimization.

[0108] Step 5: Execution and Feedback.

[0109] Short window response reliability update: After receiving the dispatch instruction from task F001, driver D001 completed the order acceptance confirmation within 5 seconds.

[0110] The system records this successful order acceptance event and its actual response time (5 seconds).

[0111] The system maintains a sliding window of the most recent 100 response records for driver D001.

[0112] The new 5-second record has been added to the window, while the oldest record has been removed.

[0113] Based on this updated window data, the system refits the log-normal distribution parameters of driver D001's response delay.

[0114] Fatigue level updates online: After driver D021 completed task F001, the system recorded the actual service time for this task as 25 minutes.

[0115] He then took a break, parking the car for 20 minutes.

[0116] His continuous working time is then reset to 0. For a period of time afterward, the system will consider him to be in a state of high energy. Therefore, in task sorting and utility calculation, he may be assigned more or more important tasks, thus achieving dynamic and fair allocation of workload.

[0117] Risk of being late is triggered: For example, task F002 has been assigned to driver D005. The system detects through real-time monitoring of the vehicle's GPS that vehicle D005 lingered at a certain point for more than 5 minutes (long-term lingering) on ​​its way to the task's starting point.

[0118] Based on the estimated arrival time calculation, the system determined that it could not arrive within the task time window [10:35, 10:50].

[0119] The system will mark driver D005 as "delayed en route" and use a unified automatic assignment method to find the best match for task F002, and reassign the task to a new driver (such as driver D006).

[0120] This mechanism ensures that an unexpected situation involving a single driver will not cause the entire mission to fail, thus improving the system's robustness in dealing with emergencies and the overall on-time completion rate of the mission.

[0121] A hybrid dispatching system for airport shuttle vehicles that considers the uncertainty of driver behavior, such as Figure 5 and Figure 6As shown in the system framework diagram and hardware deployment architecture diagram, the hybrid scheduling method for airport shuttle vehicles that considers driver uncertainty, as described above, includes the following modules: (1) Data access and preprocessing unit: This unit is responsible for accessing and cleaning real-time data from multiple sources at the airport, including flight dynamics information, airport flight area vehicle road network data, shuttle vehicle positioning and status, and driver information, to provide a high-quality and consistent data foundation for subsequent decision-making; (2) Task Pool Management Unit: This unit generates a set of shuttle tasks to be executed based on real-time updates of flight information, records the service time window, priority, number of vehicles required, origin and destination points and task status of each task, constructs a candidate "driver-task" set, and dynamically updates the task pool and candidate task-driver set. ; (3) Uncertainty Modeling Unit: This unit establishes a driver behavior uncertainty model based on driver history and real-time data, including calculating the utility function and task acceptance probability of each "driver-task" pair, the response delay estimate of each driver, and the short window response reliability index. (4) Driver mobile terminal: As the “interaction interface” between the system and the driver, it is used to display the visible task items to the driver and perform personalized sorting and confirmation operations according to the task acceptance probability results. At the same time, the terminal automatically reports the driver’s location, status and task execution milestone data (confirmation / arrival / start / completion). (5) Unified automatic allocation engine: When a task conflict is detected (the number of concurrent confirmations for the same task exceeds the number of vehicles required), a task timeout (the number of concurrent confirmations is less than the number of vehicles required), or a sudden update (significant changes in the input related to the task), the unified automatic allocation method is called, the allocation result is output and the scheduling instruction is issued. The candidate set is provided by the task pool management unit; when the order is successfully matched, the scheduling instruction is issued directly. (6) Dispatcher console: Provides a graphical monitoring interface for dispatchers. Dispatchers can view the status of all vehicles and tasks in real time and exercise manual intervention authority, including modifying, canceling or replacing assignments, and generating intervention logs; (7) Execution monitoring and feedback unit: This unit is used to track the location / progress of the driver and the task in real time, update the estimated arrival time and task status, detect long-term delays / deviations or expected service time windows exceeding the task and trigger the unified automatic allocation engine for scheduling to avoid risks. (8) Parameter update unit: Based on the feedback data collected by the execution monitoring unit, this unit performs online sliding window updates on the response delay distribution parameters and driver fatigue, and updates the short window response reliability index and task sorting accordingly. The update results are used for subsequent driver terminal display and unified automatic allocation. The timestamps and parameters of the relevant updates are recorded in the execution log.

[0122] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.

Claims

1. A hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty, characterized in that, Includes the following steps: Step 1: Information collection and preprocessing, including flight status, driver information, and shuttle bus information; Step 2: Based on flight dynamics and driver information, generate a candidate task-driver set for the service flight number. ; Candidate Tasks - Driver Collection Including each driver Task entries ; Step 3: Solve for driver uncertainty indices, including: task acceptance probability and short-window response reliability indices; Among them, the task acceptance probability is used to display the sorting in step 4; the short window response reliability index is used for unified automatic allocation in step 4; Step 4: Generate scheduling instructions based on the hybrid scheduling mechanism.

2. The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to claim 1, characterized in that, Step 2 specifically includes: Step 2A: Generate a task pool, which contains several tasks to be executed; Step 2B: Filter drivers based on service time windows, arrival time calculation models, and shuttle bus information to obtain information for each driver. Task entries .

3. The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to claim 2, characterized in that, Task acceptance probability : ； ； ； ； ； Index for drivers, For task indexing, - Drivers can accept a collection of tasks, i.e., task entries; - The task index in the task entry; - Arrival time calculation model, calculates the estimated arrival time from the current position of the shuttle bus to the starting point of the task; -Task priority; - Driver fatigue; These are utility weighting coefficients, and they are not all 0 at the same time; , , These are weight parameters; - Weights set according to the mission security level; - The upper bound of the service time window; - Current moment; -driver The duration of continuous operation in a single session; , , This is the slope parameter.

4. The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to claim 3, characterized in that, Short window response reliability index : ； ； - Decision-making time window; -Gaussian error function; and They are drivers The logarithmic mean and logarithmic standard deviation of historical order response times.

5. The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to claim 4, characterized in that, Step 4 specifically includes the following steps: Step 4A: For each driver , and the corresponding task entries Task acceptance probability The message is sent to the driver's mobile device. according to The data is displayed in a sorted manner, allowing drivers to make decisions within their designated time window. Internally selects tasks; Decision-making window After completion, statistics for each task will be compiled. j Concurrent confirmations Number of vehicles required for the mission And perform step 4B or step 4C; Step 4B, Unified automatic allocation of tasks triggered by conflicts and timeouts: If for tasks j The task will be completed if any of the following conditions are met. Included in decision-making task set And construct a decision candidate set : Order-grabbing conflict: ; Order timeout: ; - The number of concurrent confirmations for the same task; - Number of vehicles required; Among them, the decision candidate set The construction rules are as follows: when hour, Includes this task Candidate pairs formed with drivers who have already been confirmed for the task; when At that time, first contact the party that has already confirmed the task. The dispatch instructions are sent from one driver to the corresponding driver's mobile terminal. Includes "task" With sets The candidate pairs, consisting of drivers who did not participate in the order-grabbing process, are used to fill the gaps. One vacancy remains; Subsequently, in the assembly The system calls a unified automatic allocation method to generate scheduling instructions and sends them to the corresponding driver's mobile terminal. The scheduling instruction includes at least the task start point, the task end point, and the service time window; Step 4C: Successful order matching leads to direct issuance of scheduling instructions: If the task... satisfy If the order is successfully matched, a dispatch instruction is generated and sent to the corresponding driver's mobile terminal.

6. The hybrid scheduling method for airport shuttle vehicles considering driver behavior uncertainty according to claim 5, characterized in that, The unified automatic allocation method in step 4B: ； ； ； ； in, - Binary assignment variables; - Decision-making time window; For the task The upper bound of the required number of vehicles in this round of unified automatic allocation; -driver The number of tasks that can be accepted simultaneously.

Images