A method, system, device and medium for scheduling a pilot station vehicle fleet pick-up and drop-off task

Abstract

The application discloses a kind of pilot station car team pick-up and drop-off task scheduling method, system, equipment and medium, applied to pilot information technology field.Therein, scheduling method includes: online entry pick-up and drop-off task information;Estimate the time consumption of journey to task, give the traffic cost weight lower than conventional road section to the link road section;In clustering, adopt double association basis, so that the task combination with connection relationship is preferentially included in the same task cluster;Generate dispatching scheme based on estimation and clustering result, and plan coherent operation sequence based on current task destination and historical to-be-executed task starting point, and trigger recalculation and update dispatching scheme synchronously when task changes.The application makes "end-point-starting point coherence" constraint run through journey estimation and clustering phase, changes scheduling from passive record mode of each task isolated processing to active decision mode of connection priority and full-link through, realizes carpooling efficiency maximization and empty mileage minimization, significantly reduces operating cost.

Description

Technical Field

[0001] This application belongs to the field of pilotage information technology, specifically relating to a method, system, equipment and medium for scheduling pilotage station vehicle convoy pick-up and drop-off tasks. Background Technology

[0002] Port vessel pilotage is a crucial link connecting shipping and port operations, and fleet scheduling, as a pre- or post-pilot operation, plays an indispensable role in the entire pilotage process.

[0003] Because pilotage tasks are affected by objective conditions such as weather, ship condition, and dock conditions, the pick-up and drop-off times and locations for pilots are also complex and changeable. Fleet dispatchers need to make reasonable arrangements for drivers to complete the pick-up and drop-off tasks based on factors such as time or location, and at the same time, they need to monitor the task completion status in real time and make timely adjustments if the task is changed temporarily.

[0004] In the dispatching of pilot station fleets, the following challenges are faced: On the one hand, the starting point and destination of each pilot are not exactly the same, and the pick-up and drop-off times are also flexible and changeable. Whether the same vehicle can be used to arrange multiple pilots becomes a challenge in the dispatching work. On the other hand, the experience of dispatchers, the condition of vehicles, and the familiarity of drivers with the roads also vary, which indirectly affects the operating cost of picking up and dropping off pilots.

[0005] In summary, how to ensure that pilot station fleets can transport pilots reasonably, efficiently, and at low operating costs is an urgent problem that needs to be solved. Summary of the Invention

[0006] The technical problem this application aims to solve is to provide a method, system, equipment, and medium for scheduling pilot station fleet transportation tasks, addressing the issues of existing pilot station fleet scheduling relying entirely on manual experience, poor task coordination, and delayed information synchronization. Through improved clustering algorithms, dynamic programming algorithms, and a closed-loop scheduling mechanism throughout the entire process, it achieves dual optimization of maximizing carpooling efficiency and minimizing empty mileage.

[0007] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a method for scheduling convoy pick-up and drop-off tasks at pilot stations, including: The pick-up and drop-off task information is entered online on the dispatch management system. The pick-up and drop-off task information includes the starting point, the destination and the pick-up and drop-off time, and the pick-up and drop-off task information is synchronized to the driver terminal and the pilot terminal in real time. The recorded pick-up and drop-off tasks are estimated to take travel time. When calculating the route cost, the travel time estimation assigns a lower toll cost weight to the connecting road segments that form the route from the end of the preceding task to the start of the following task than to the regular road segments, so as to prioritize the construction of travel routes that include connecting road segments. Cluster analysis is performed on multiple concurrent tasks. When determining the correlation between tasks, the spatial distance and time difference between the end point of the preceding task and the starting point of the subsequent task are used as necessary judgment dimensions, so that task combinations with a connection relationship are preferentially included in the same task cluster. Based on the estimated trip time and the results of the cluster analysis, a vehicle dispatch plan including pick-up and drop-off order and driving route is generated. Based on the destination location and estimated arrival time of the current task, it is matched with the starting location and estimated departure time of the tasks to be executed in the historical task data warehouse to plan the connection between the current task and the next task to be executed, forming a coherent operation sequence. When mission information changes, a recalculation is triggered and the updated scheduling scheme is synchronized to the driver's terminal and the pilot's terminal.

[0008] Preferably, the trip time estimation step includes: Connect to the map API to obtain real-time traffic data and the basic travel time for each road segment; An improved path search algorithm is used to generate basic driving paths. When calculating path weights, for connecting road segments where the endpoint of the preceding task coincides with or is adjacent to the starting point of the following task, an empty driving optimization coefficient is introduced to reduce the toll cost weight of the connecting road segment. The base travel time is corrected by combining real-time traffic index, time period coefficient and weather impact coefficient. Based on the proportion of the length of connecting road segments in the total travel path, the travel time is adjusted again by the empty-run optimization coefficient, and the final estimated travel time and candidate path are output.

[0009] Preferably, a multi-factor correction model is constructed to correct the base time consumption based on the multi-factor correction model; wherein the multi-factor correction model adopts the following formula:

[0010] Wherein, T(f) is the final estimated time, T(b) is the base time returned by the map API, k(traffic) is the real-time traffic congestion index, k(time period) is the time period coefficient, and k(weather) is the weather impact coefficient. Z is the empty-run optimization coefficient, and Z is the proportion of the length of the connecting road segment to the total path length.

[0011] Preferably, the improved path search algorithm is an improved Dijkstra algorithm. When the improved Dijkstra algorithm is executed, nodes in the road network that are the endpoints of the preceding tasks and the starting points of the following tasks are marked as connecting nodes. After each selection of the current node, the starting nodes of the following tasks associated with the connecting node are traversed first.

[0012] Preferably, the improved path search algorithm is an improved A* algorithm, and the heuristic function of the A* algorithm is:

[0013] in, Basic adjustment coefficient, Let n be the straight-line distance from node n to the current task target point. To connect the guiding coefficient, Let n be the straight-line distance from node n to the nearest connected node; when there is no connected node, ... The value is 0.

[0014] Preferably, the cluster analysis employs an improved DBSCAN density clustering algorithm, including: Each complete pick-up and drop-off task is defined as a bound trajectory unit containing a start point and an end point; Calculate the Euclidean distance between trajectory points of different tasks, and set the neighborhood radius, minimum number of tasks threshold, task time window overlap threshold, and return task connection time difference threshold. The criteria for determining the correlation between tasks are: the distance between at least two sets of trajectory points is less than or equal to the neighborhood radius, the distance between the end point of the preceding task and the start point of the following task is less than or equal to the neighborhood radius, and the time difference meets the return task connection time difference threshold. The core tasks that satisfy the above correlation determination and their related tasks are iteratively included into the same initial task cluster.

[0015] Preferably, an improved TSP dynamic programming algorithm is used to generate a vehicle dispatch plan that includes pick-up and drop-off order and driving route, including: With the goal of minimizing total driving distance and vehicle empty driving distance, constraints are set including pick-up and drop-off priority, vehicle passenger capacity, time window, and route closed loop. In the algorithm state definition, a connection marker is introduced to record whether the current path contains a connection unit between the end point of the preceding task and the start point of the following task. In the state transition equation, when the path contains the connecting unit, an empty-run optimization coefficient is introduced to calculate the reduction in empty-run mileage brought about by the connecting unit, so as to prioritize the selection of paths containing connecting units during the search process.

[0016] Preferably, the state transition equation of the improved TSP dynamic programming algorithm is:

[0017] Where X is the currently accessed trajectory point, Y is the previously accessed trajectory point, S is the binary status code, dp is the shortest travel distance in the current state, and dist(Y,X) is the distance between the two points. For empty running optimization coefficients, Z is the connection marker of the previous state, and Z is the empty mileage corresponding to the connection unit. Z takes the corresponding empty mileage value if and only if the end point of the previous task and the start point of the subsequent task are the same or adjacent connection units; otherwise, Z takes the value of 0.

[0018] Preferably, the connection between the current task of the planned vehicle and the next task to be executed includes: Obtain the destination location and estimated arrival time of the task currently being performed by the vehicle; In the historical task data warehouse, all tasks to be executed that are less than a preset threshold in terms of the spatial distance between the origin and the destination and less than a preset time difference threshold in terms of the difference between the estimated departure time and the estimated arrival time are retrieved and used as candidate connection tasks. Based on collaborative filtering, the similarity between the current task and each of the candidate connection tasks is calculated. Combined with the time connection window, the comprehensive score of each candidate connection task is calculated through multi-objective decision-making, and the candidate connection task with the highest score is selected as the optimal connection task.

[0019] Preferably, the triggering of recalculation and synchronization of the updated scheduling scheme to the driver's terminal and the pilot's terminal includes: When a change in task information is detected, the trip time estimation and cluster analysis are recalculated based on the changed task parameters within a preset time threshold, and an adjusted dispatch plan is generated. The adjusted dispatch plan is pushed to the driver terminal and the pilot terminal in real time, and the changed data is synchronously updated to the historical task data warehouse.

[0020] Preferably, the online entry of pick-up and drop-off task information further includes: automatically recording and archiving the executor, operation content and operation time of each task change, forming a change log.

[0021] Secondly, this application provides a pilot station vehicle convoy dispatching system, including: The integrated information management module is used to realize the online input, real-time synchronization and closed-loop confirmation of task information between the dispatch terminal, driver terminal and pilot terminal; The intelligent algorithm module is used to estimate the travel time of received pick-up and drop-off tasks and perform cluster analysis on multiple concurrent tasks to generate a dispatch plan that includes the pick-up and drop-off order and driving route. Specifically, when calculating the route cost, the travel time estimation assigns a lower toll cost weight to connecting road segments from the endpoint of a preceding task to the starting point of a subsequent task, prioritizing the construction of driving routes that include these connecting road segments. When determining the correlation between tasks, the cluster analysis uses the spatial distance and time difference between the endpoint of a preceding task and the starting point of a subsequent task as necessary judgment dimensions, ensuring that task combinations with a correlation relationship are preferentially included in the same task cluster. The historical task data warehouse module is used to store historical and current task data. Based on the destination location and estimated arrival time of the current task, it matches the starting location and estimated departure time of the tasks to be executed in the historical task data warehouse to plan the connection between the current task and the next task to be executed, forming a coherent operation sequence. The coordination and dynamic adjustment module is used to trigger the intelligent algorithm module to recalculate when the task information changes, and synchronize the updated scheduling scheme to the driver terminal and the pilot terminal.

[0022] Preferably, the intelligent algorithm module includes a trip time estimation unit, which is used for: Connect to the map API to obtain real-time traffic data and the basic travel time for each road segment; An improved path search algorithm is used to generate basic driving paths. When calculating path weights, for connecting road segments where the endpoint of the preceding task coincides with or is adjacent to the starting point of the following task, an empty driving optimization coefficient is introduced to reduce the toll cost weight of the connecting road segment. A multi-factor correction model is constructed, which combines real-time traffic index, time period coefficient and weather impact coefficient to correct the basic travel time. Based on the proportion of the length of connecting road segments in the total travel path, the travel time is adjusted again by the empty driving optimization coefficient, and the final estimated travel time and candidate path are output. And / or, the intelligent algorithm module includes a carpooling matching and route planning unit, which is used for: An improved DBSCAN density clustering algorithm is used to define each complete pick-up and drop-off task as a bound trajectory unit containing a start point and an end point. The Euclidean distance between trajectory points of different tasks is calculated, and the core task and its associated tasks that meet the correlation determination are iteratively included in the same initial task cluster. The correlation determination is based on the following criteria: the distance between at least two sets of trajectory points is less than or equal to a preset neighborhood radius, and the distance between the end point of the preceding task and the start point of the following task is less than or equal to the preset neighborhood radius, while the time difference meets the preset return task connection time difference threshold. An improved TSP dynamic programming algorithm is adopted. A connection marker is introduced into the state definition to record whether the path already contains a connection unit. An empty-run optimization coefficient is introduced into the state transition equation to calculate the reduction of empty-run mileage brought by the connection unit, so as to generate the pick-up and drop-off sequence and driving path containing connection units.

[0023] Thirdly, this application provides an electronic device, including at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform: a pilot station fleet dispatching method as described in any one of this application.

[0024] Fourthly, this application provides a computer storage medium storing computer-executable instructions, which are executed by a processor: the pilot station vehicle convoy dispatching method as described in any one of this application.

[0025] Compared to existing technologies, the beneficial effects achieved by this application may include one or more of the following aspects: On the one hand, it fundamentally solves the problem of "blind spots in task connection" and realizes the overall transformation from "passive recording" to "proactive decision-making".

[0026] Even with existing technologies for pilot station management systems that use computer systems, they merely migrate task information from paper to online. Essentially, they are "record-based" systems where each task is handled in isolation. The natural connection between tasks, where "the end of the preceding task is the start of the following task," relies entirely on manual discovery, resulting in a high rate of omissions.

[0027] This application is the first to solidify the business characteristic of "end-start continuity" into the core constraint at the algorithm level, and injects it simultaneously in the two key stages of trip time estimation and cluster analysis, so that the system is upgraded from "passively recording task information" to "actively discovering and utilizing connection relationships", realizing the fundamental transformation of pilot station fleet scheduling from "recording type" to "decision type".

[0028] Secondly, based on the dual-constraint collaborative operation, the inherent empty-running problem in the dispatching of pick-up and drop-off tasks can be resolved at the source.

[0029] In this application, by introducing a weight discount for connecting road segments, the paths generated during the trip prediction stage are already biased towards including connecting road segments; and by introducing a dual association criterion in clustering, it is further ensured that the task clusters formed by clustering naturally possess a clear return trip connection structure. Therefore, the dispatching scheme generated based on these two results has embedded empty-run avoidance logic from the very beginning. For example, in the typical example of three people and three tasks, zero empty-run mileage was achieved, and fuel costs were reduced by approximately 20% compared to manual dispatching.

[0030] In three aspects, historical data-driven task coordination planning can expand single-trip optimization into continuous operation across the entire chain, significantly improving efficiency and reducing operating costs.

[0031] It should be noted that in the traditional manual mode of pilot station vehicle dispatching, dispatchers need to manually browse through historical records to try to match and connect tasks, which is extremely inefficient and often misses opportunities to connect. Vehicles are often forced to return empty after completing their current task.

[0032] This application establishes a historical task data warehouse and models task connection as a spatiotemporal matching and recommendation problem. The system proactively recommends the optimal next task for vehicles that are about to complete their tasks, transforming vehicles from "single isolated operations" into "continuous operation chains." Even in scenarios without carpooling, this mechanism can still enable vehicles to operate continuously, avoiding empty returns and significantly improving the operational efficiency of a single vehicle.

[0033] In four aspects, a closed-loop dynamic adjustment mechanism is used to ensure the robustness of pilotage scheduling in special scenarios.

[0034] Pilotage operations are subject to frequent changes due to weather, ship condition, and berth availability. Traditional methods rely on individual telephone notifications, which can take several minutes to tens of minutes to respond, and the changes cannot be tracked or recorded.

[0035] The closed-loop mechanism of this application automatically triggers a full-process recalculation within seconds of a task change, pushes the updated plan to all parties' terminals in real time, and synchronously archives the changed data to the historical data warehouse. This mechanism enables the entire scheduling system to maintain the optimality of the plan and the consistency of information even in a highly dynamic environment, significantly improving the system's robustness, timeliness, and traceability. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall process of a pilot station vehicle convoy dispatching method according to this application; Figure 2 This is a simplified flowchart illustrating a method for scheduling a pilot station vehicle convoy in this application. Figure 3 This is a schematic diagram illustrating the optimal path planning for three people and three tasks in this application; Figure 4a This is a schematic diagram of the clustering results in this application where traditional DBSCAN clustering relies solely on spatial distance, resulting in a lack of connectivity information within clusters. Figure 4b This diagram illustrates the improved DBSCAN clustering method in this application, which ensures that clusters themselves have connectivity information based on dual associations. Figure 5 This is a schematic diagram of the recommendation mechanism based on historical tasks in this application; Figure 6 This is a schematic diagram of the dynamic adjustment closed-loop mechanism when the execution task changes in this application; Figure 7 This is a schematic diagram of the overall architecture of a pilot station vehicle dispatching system according to this application; Figure 8 This is a schematic diagram of an electronic device used for scheduling pilot station vehicle convoy pick-up and drop-off tasks, as described in this application. Detailed Implementation

[0037] The specific embodiments of the technical solution of this application will be described in detail below with reference to the accompanying drawings. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0038] The meanings of some terms in this application are illustrated as follows: "Task" refers to a complete operation involving the transportation of a pilot, comprising a start point and an end point; "Preceding Task" refers to a task that is preceding in time or prioritized in the route planning; "Subsequent Task" refers to a task that is subsequent in time or prioritized in the route planning; "Connecting Segment" refers to a segment of road where the end point of a preceding task coincides with or is spatially very close to the start point of a subsequent task; "Empty Run" refers to the mileage traveled by the vehicle without any pilots on board; "Empty Run Optimization Coefficient" This is an adjustment parameter used to give discounts to connecting road segments in route cost and time calculations; its value can typically be between 0.1 and 0.2.

[0039] Port pilotage is the core link between shipping and port operations. Pilotage station fleets are responsible for picking up and dropping off pilots from designated locations to their target wharves or vessels on time. While the point-to-point transportation of pilots may seem simple, it faces highly complex constraints in practice. For example, pilotage tasks are affected by multiple objective factors such as weather, ship condition, and wharf berth availability, leading to frequent changes in pick-up and drop-off times and locations. Furthermore, each pilot has different starting and ending points, requiring flexible and staggered time windows.

[0040] Therefore, how to combine multiple pilots into the same vehicle for pick-up and drop-off, and in what order to do so, i.e. how to pick up and drop off pilots reasonably, efficiently, and accurately, is a complex problem subject to multiple constraints.

[0041] Through in-depth research and analysis, it was found that the current fleet dispatching mode at pilot stations is still mostly based on the traditional solution of manual operation. That is, the dispatcher completes task registration by handwriting records, estimates the travel time based on personal experience, judges whether carpooling is possible and plans the carpooling order based on memory, and manually checks the historical records to achieve the connection of vehicle tasks.

[0042] Further investigation revealed the following problems with the manual dispatching model for pilot station vehicle fleets transporting pilots: (1) Information recording and synchronization are inefficient. Handwritten registration is prone to errors and difficult to archive. Task changes are communicated only by telephone, which cannot form a record of changes and makes it difficult to ensure the consistency of information among multiple parties. (2) The task planning lacks scientific basis and relies on the experience of personnel to estimate the time consumption, which may result in waiting in vain if the vehicle is dispatched too early or delays if the vehicle is dispatched too late; matching carpooling tasks based on memory may result in missing carpooling scenarios or adding unnecessary trips due to unreasonable carpooling order. (3) The planning of vehicle task continuity is difficult. Manually reviewing historical records is time-consuming and laborious, and it is difficult to cover all aspects, resulting in low vehicle task connection rate, increased empty mileage, and increased operating costs.

[0043] Therefore, how to use computer technology to establish an intelligent personnel dispatching method for pilot station fleets, effectively improve vehicle and driver resource utilization, reduce fuel costs, and reduce reliance on personal experience while ensuring timely arrival of pilots, has become an urgent technical problem to be solved.

[0044] Existing technology CN103426330A discloses a comprehensive pilotage service system that integrates fleet dispatch management, pilot vessel operation management, and pilot scheduling into a unified information platform. Compared to purely paper-based handwritten records, this existing technology enables the migration of task information from offline to online, solving the problem of convenient information storage and retrieval.

[0045] However, after in-depth research and practical experience, the applicant discovered that the existing technology and similar information systems still have a fundamental technical flaw: they are essentially record-keeping systems for scheduling, rather than intelligent decision-making systems for scheduling. Specifically, this manifests as follows: Deficiency 1: The problem of accurately estimating "reasonable travel time" has not been solved. The existing system only provides an information input and display interface, and dispatchers still need to rely on personal experience to estimate the travel time from point A to point B. When there are multiple pick-up and drop-off points, and weather and road conditions change dynamically, manual estimation has a large error, which can easily lead to dispatching vehicles too early, resulting in vehicles waiting in vain, or dispatching vehicles too late, causing pilot delays, and thus affecting ship berthing and unberthing operations.

[0046] Deficiency 2: The problem of "discovering and optimizing carpooling combinations" remains unresolved. Dispatchers rely on memory to determine if there are any carpooling opportunities for tasks with adjacent origins or destinations during the same period. When the number of tasks is high, human memory cannot cover all possible combinations, leading to the omission of many carpooling scenarios and wasting vehicle resources. Even if experienced dispatchers discover carpooling combinations, planning the pick-up and drop-off order to ensure a logical "pick-up first, drop-off later" approach and the shortest total route still relies on subjective judgment, lacking the support of scientific calculation tools.

[0047] Deficiency 3: The core pain point of "vehicle task connection to avoid empty runs" remains unresolved—this is the most critical deficiency identified by the applicant. The pilot station fleet has a very limited number of vehicles, and whether each vehicle can pick up the next task near its destination after completing its current task directly determines the empty run rate. The existing system lacks a structured analysis function to retain historical task execution data; dispatchers can only manually browse records to attempt matches, which is extremely inefficient and often misses potential connection opportunities. Furthermore, it is common for vehicles to return empty to the dispatch center after dropping off pilots to await the next assignment in the traditional dispatching model, resulting in consistently high empty run mileage and persistently high fuel and labor costs.

[0048] Defect 4: The system's modules operate independently, lacking coordination and dynamic response capabilities. Even with the integration of multiple functional modules, the existing system still relies on manual intermediaries for information flow. When tasks change temporarily, dispatchers must manually modify records, notify drivers and pilots by phone, and then readjust the plan based on experience, resulting in prolonged response times and a high risk of errors.

[0049] In response to the aforementioned shortcomings, the applicant conducted in-depth reflection and made attempts: Regarding the first deficiency: accurate estimation of travel time is essentially a path planning and time prediction problem in a dynamic environment. Existing technologies, such as map APIs, can provide travel time estimates based on real-time traffic conditions, but can only handle scenarios "from a single starting point to a single destination." The key uniqueness of pilot station pick-up and drop-off tasks lies in the fact that when a vehicle completes a previous task, the starting point of the next task happens to be adjacent to or overlaps with the destination of the previous task, forming a special "connecting" task. This connection means that the vehicle does not need to drive empty or only needs to travel a very short distance to pick up the next task, which is superior to the conventional "return from destination to starting point to pick up passengers" model in terms of both time and cost efficiency. Therefore, time estimation must be able to identify and assign different calculation weights to such "connecting segments" compared to regular segments in order to ensure that the estimation results truly reflect operational efficiency.

[0050] Regarding the second deficiency: the discovery of carpooling combinations is essentially a multi-dimensional clustering problem. Traditional spatial clustering algorithms group tasks solely based on geographical coordinate distance, failing to meet the specific constraints of pilot pick-up and drop-off. The applicant discovered that a "carpoolable" task combination must simultaneously satisfy two levels of relevance: first, spatial relevance, meaning the start or end points of the tasks are spatially proximate, facilitating convenient pick-up and drop-off; second, "end-point-start coherent relevance," meaning the end point of one task is precisely the start point of another task or its nearest neighbor. The latter is crucial for achieving return trip connections and is a dimension completely ignored by traditional clustering. If this "coherence" is incorporated as one of the core criteria for clustering, the clustering results themselves inherently contain the potential for avoiding empty runs.

[0051] Regarding defect three: vehicle task coordination is hampered by incomplete information acquisition. The low efficiency of manual review stems from the dispatcher's inability to comprehensively and in real-time grasp the spatiotemporal matching relationships between all pending tasks and currently executing tasks. If a structured data warehouse of historical and pending tasks is established, the system can automatically retrieve tasks whose starting point is near the current task's destination and whose departure time is close to the estimated arrival time when a vehicle is about to complete its current task. By modeling "task coordination recommendation" as a similarity calculation and ranking problem, a shift from "passive search" to "active recommendation" can be achieved.

[0052] Regarding defect four: dynamic response requires the establishment of automated data communication links between various modules of the system. When the data in any link changes, it triggers a recalculation and data synchronization, rather than waiting for manual instructions.

[0053] Guided by the aforementioned preliminary solutions, the applicant conducted trials and verifications of technical feasibility: Attempt 1: Differentiate the weights of connecting road segments in the path search algorithm. Based on the classic Dijkstra's algorithm and A* algorithm, add node type labeling during the road network node initialization phase—marking nodes that simultaneously serve as both a task endpoint and a task start point as "connecting nodes." In path weight calculation, assign discount coefficients (empty-run optimization coefficients) to the outgoing edges associated with connecting nodes. In the experiment, a weight of 0.1 was initially used, which naturally biased the algorithm towards paths passing through connecting nodes during the search process. Experimental results show that even in randomly generated multi-task test scenarios, after adding the connecting segment discount, the proportion of paths output by the algorithm containing connecting segments increased from about 12% to about 78%, proving that this differentiated weight strategy can effectively guide the algorithm to discover and utilize connecting opportunities.

[0054] Attempt 2: Adding an "endpoint-startpoint" association constraint to DBSCAN clustering. In addition to the traditional DBSCAN spatial distance determination rules, a new criterion is added: two tasks are considered associated only if at least one pair of "preceding endpoint and subsequent startpoint" has a distance less than the neighborhood radius and a time difference meeting a preset threshold. After running clustering with this criterion, each generated task cluster contains at least one pair of reversible task pairs. Comparative experiments show that introducing this constraint reduces the average empty mileage of task execution within a cluster by approximately 65%, demonstrating that this criterion significantly improves the coherent execution of tasks within the cluster.

[0055] Attempt 3: Extending the state space of TSP dynamic programming. In addition to the traditional TSP state code that only records the "set of visited nodes," a new "connection flag M" is added to explicitly indicate whether a connecting unit has been connected in the current path. In the cost function of state transitions, an empty-run optimization coefficient is introduced if and only if the state transition involves a connecting unit. The product of the cost and the empty mileage Z is used as the amount to be reduced, so that the cumulative cost of routes including connecting routes is lower than that of non-connecting routes of the same distance. Preliminary calculations show that this change further improves the selection rate of connecting routes in the searched optimal routes and can select routes with less empty mileage near the same total mileage.

[0056] Attempt 4: Constructing a task connection recommendation model based on collaborative filtering. Using the current destination and estimated arrival time of the pilot station convoy as query vectors, candidate connection tasks are retrieved from a historical task data warehouse. Recommendations are output based on a weighted score calculated using task attribute similarity and spatiotemporal matching. Tests on a small-scale historical dataset show that the model can effectively identify connection opportunities that dispatchers may easily overlook, with a connection recommendation hit rate of approximately 85%.

[0057] Based on the successful attempts in the above-mentioned sub-items, the applicant further discovered a key pattern: the various technical means do not operate in isolation, but rather there is a significant functional relationship of mutual support.

[0058] Specifically, further new discoveries were made in the above attempts: The discount weights for connecting routes introduced in the trip time estimation provide a data foundation for identifying "connecting relationships" in cluster analysis; The newly added "end-start coherence" criterion in cluster analysis ensures that each generated task cluster contains at least one set of connectable tasks, providing a target direction for subsequent path planning; The newly added connection markers and empty-run optimization coefficients in route planning transform the connection potential identified in the clustering stage into actual route construction, producing dispatching schemes with zero or low empty runs. Task connection recommendation based on collaborative filtering further extends single scheduling into a continuous job sequence, allowing the results of a single optimization to continue over a longer time scale.

[0059] Based on the above findings, the core technical concept of this application is to transform the scheduling of pilot pick-up and drop-off tasks from a passive recording mode of "isolated processing of each task" to a proactive decision-making mode of "seamless connection and end-to-end connectivity." (Reference) Figure 1 and Figure 2 As illustrated, after the task information is entered online in step S1 (or at the same time as the entry), the task combination is injected with the "end-start continuity" constraint in step S2 (trip time estimation) and step S3 (cluster analysis). This allows step S4 to generate a dispatch plan that includes the pick-up and drop-off order and driving route based on the trip time estimation and cluster analysis results under the constraints. Then, step S5 completes the task connection planning to obtain a coherent operation sequence, and step S6 achieves dynamic adjustment of the dispatch plan through closed-loop synchronous feedback. This ensures that the final pilot station fleet scheduling plan (i.e., the dispatch plan) has the dual attributes of high efficiency in carpooling and avoidance of empty driving from the very beginning, and can dynamically adapt to task changes through the closed-loop feedback mechanism.

[0060] It should be noted that this application does not simply arrange the algorithm flow, but rather injects "endpoint-startpoint continuity" connection constraints into key steps such as dynamic programming and clustering (e.g., steps S2 and S3). This means that when calculating path costs, connecting road segments are identified and assigned a lower toll cost weight than regular road segments. Furthermore, the spatial distance and time difference between the preceding endpoint and the following startpoint are used as necessary judgment dimensions for clustering criteria. This ensures that the entire process from trip prediction to carpooling clustering is embedded with empty-run avoidance logic. The resulting pilot station fleet pick-up and drop-off dispatch scheme has dual attributes and achieves dynamic adaptation through closed-loop feedback.

[0061] The following implementation examples illustrate this application.

[0062] Example 1 This embodiment describes the basic process of the pilot station vehicle fleet dispatching method of this application.

[0063] Traditional pilot station fleet dispatching relies entirely on manual operation, estimating time based on experience, matching carpooling tasks based on memory, and manually checking historical records to plan task connections, resulting in numerous information errors, high vehicle empty-running rates, and low dispatching efficiency.

[0064] The inventive concept of this application is to replace manual experience-based decision-making by constructing a closed-loop scheduling mechanism that integrates real-time information synchronization, intelligent algorithm planning, historical data-driven connection, and dynamic recalculation and adjustment. At the algorithm level, cost discounts for connecting road segments are introduced into the trip time prediction, and destination-origin continuity constraints are introduced into the cluster analysis. This fundamentally transforms the scheduling mode that simply pursues the shortest path into a dual-objective scheduling mode that simultaneously pursues the minimum empty runs and the optimal carpooling efficiency.

[0065] refer to Figure 1 and Figure 2 This application illustrates a method for scheduling a pilot station convoy for pick-up and drop-off missions, comprising: Step S1: Dispatchers enter pick-up and drop-off task information online into the dispatch management system. Task information includes the pick-up / drop-off origin, destination, and time. The system automatically synchronizes the task information in real time to the corresponding driver and pilot terminals via a communication interface (such as a wireless interface). After receiving the task information, the driver and pilot terminals can provide confirmation, forming a closed-loop information transmission.

[0066] For example, there are 3 tasks to be performed: Mission T1 (Pilot A, starting point A (117.928, 39.087), destination L (117.932, 39.090), estimated pick-up time 9:00-9:15). Mission T2 (Pilot B, starting point C (117.930, 39.088), ending point D (117.935, 39.092), estimated pick-up time 9:05-9:25). Mission T3 (Pilot C, starting point L (117.932, 39.090), ending point F (117.936, 39.089), estimated pick-up time 9:25-9:40). The dispatcher enters task T1: Pilot A, origin A, destination L, pick-up / drop-off time 9:00-9:15; Task T2: Pilot B, origin C, destination D, pick-up / drop-off time 9:05-9:25; Task T3: Pilot C, origin L, destination F, pick-up / drop-off time 9:25-9:40. The system pushes the above task information to the mobile terminals of the corresponding drivers and pilots in real time.

[0067] Step S2: The system estimates the travel time for the entered pick-up and drop-off tasks. When calculating path costs and travel time, the system identifies whether there is a connection between tasks where the endpoint of a preceding task is the starting point of a subsequent task. When a connecting segment is identified—for example, the endpoint L of task 1 is exactly the starting point L of task 3—the system assigns a lower toll cost weight to this connecting segment (where the distance between L and L is zero, essentially no movement within the same location) than to regular road segments. By introducing a weight discount mechanism, the time estimation algorithm naturally tends to construct travel paths containing such connecting segments when searching for the optimal path, thus providing an algorithmic basis for avoiding empty return trips.

[0068] Step S3: The system performs cluster analysis on multiple tasks within the same time period. Unlike traditional methods that cluster solely based on spatial distance, this application uses the "spatial distance between the endpoint of the preceding task and the starting point of the subsequent task" and the "time difference between them" as necessary judgment dimensions when determining whether tasks can be grouped into the same cluster. When distinguishing between tasks T1 and T3, the system not only calculates the distance between each trajectory point of the two tasks, but also specifically examines the relationship between the endpoint L of task T1 and the starting point L of task T3. Since the endpoint L of task T1 and the starting point L of task T3 are at the same location (distance is 0), and the time difference of 10 minutes between the estimated arrival time of task T1 (9:15) and the estimated departure time of task T3 (9:25) meets the preset connection time difference threshold (e.g., 15 minutes), tasks T1 and T3 are determined to have a strong connection relationship and are preferentially included in the same task cluster. At the same time, task T2 is also included in this cluster because its starting point C is spatially close to the starting point A of task T1 (e.g., 0.0022 kilometers). Therefore, the three tasks form a composite task cluster that can be carpooled and has a return trip connection.

[0069] Step S4: Based on the trip time estimation results of Step S2 and the cluster analysis results of Step S103, the system uses a dynamic programming algorithm (such as the improved TSP dynamic programming algorithm) to generate a dispatch plan that includes specific pick-up and drop-off sequences and driving routes.

[0070] refer to Figure 3 As illustrated, the explicitly output path of this scheme is as follows: Dispatch Center O → Task T1 Start Point A (to pick up A) → Task T2 Start Point C (to pick up B) → Task T1 End Point L (to send A, and this point is also the start point of Task T3, the vehicle does not need to run empty) → Task T3 Start Point L (to pick up C, the vehicle picks up the person at the same location) → Task T2 End Point D (to send B) → Task T3 End Point F (to send C) → Dispatch Center O.

[0071] Step S5: After the vehicle dispatch plan is issued and executed, the system continuously monitors the execution status of each task. Based on the destination location and estimated arrival time of the task to be completed, the system searches the historical task data warehouse for a matching task with both a starting point and time. For example, when a vehicle is about to complete the task of transporting pilot G to destination Q at 14:20, the system finds that the starting point of task T5 is Q, and the time difference between the estimated departure time of 14:25 and the estimated arrival time of 14:20 is only 5 minutes, meeting the connection condition. The system assigns this connection task to the vehicle, allowing it to directly accept the next task on the spot after completing the current task without having to return empty or wait, forming a continuous operation sequence.

[0072] Step S6: If a temporary change occurs during the task execution process, such as the pilot changing the boarding point, the pick-up and drop-off time being adjusted due to weather reasons, or the destination being changed due to changes in the berth, the system will re-trigger the full process recalculation of steps S2 to S5 based on the changed task parameters within a preset time threshold (such as within a few seconds) after listening to the change information, generate an adjusted dispatch plan, and push it to the affected driver terminal and pilot terminal in real time.

[0073] Therefore, compared with manual dispatch, in a three-person, three-task scenario, zero empty driving mileage is achieved, with a total driving mileage of only about 4.2 kilometers, saving one vehicle, reducing fuel consumption costs by about 20%, ensuring no delays in task execution, and zero errors in information transmission.

[0074] This embodiment ensures from the source that the generated scheduling scheme has embedded empty-run avoidance logic by using the dual constraints of connecting road segment cost discount and end-point-start continuity. During the clustering stage, it locks in the task combination that can be connected back to the destination. On this basis, the route planning stage further optimizes the pick-up and drop-off sequence, so that a single vehicle can complete the pick-up and drop-off of multiple pilots at the same time. After completing the previous delivery, it can pick up the subsequent task at the same place or very close by, completely avoiding empty return trips.

[0075] Example 2 This embodiment is a further implementation plan for optimizing and improving the trip time estimation step, involving an improved path search algorithm and a multi-factor correction model.

[0076] In traditional dispatching, dispatchers rely solely on experience to estimate the simple travel time from point A to point B, failing to accurately consider real-time road congestion, traffic characteristics at different times, and the impact of weather on vehicle speed. More importantly, manual estimation cannot reflect the benefit of "saving empty-running costs on connecting road sections" in the time calculation, leading to inaccurate vehicle utilization assessment.

[0077] The inventive concept of this application is as follows: Basic travel time data is obtained by connecting to a map API; a modified Dijkstra algorithm or A* algorithm is used to generate a driving route; and an empty-run optimization coefficient is introduced into the route weight calculation to discount the cost of connecting road segments. Subsequently, a multi-factor correction model is constructed to comprehensively correct the basic travel time using road condition index, time period coefficient, weather influence coefficient, and the proportion of connecting road segments, ensuring that the output estimated travel time is both accurate and reflects the benefits of empty-run optimization.

[0078] The following example uses Dijkstra's algorithm and multi-factor correction as illustrations.

[0079] In some examples, real-time traffic data of the road network is first obtained by connecting to a third-party map API, including the length of each road segment, speed limit information, real-time vehicle speed, and free-flow vehicle speed. Then, a basic driving path is generated using an improved Dijkstra algorithm.

[0080] In implementation, the improvement to Dijkstra's algorithm lies in the following: During initialization, all task information is traversed to identify nodes that are both the endpoint of one task and the starting point of another, and these nodes are marked as "connecting nodes." For example, node L, which is both the endpoint of task T1 and the starting point of task T3, is marked as a connecting node. After the algorithm selects the current node from the candidate set for processing each time, if the current node has a connecting node association, it prioritizes traversing the starting node of the subsequent task corresponding to that connecting node and its adjacent road segments.

[0081] In path weight calculation, the weight of a regular road segment is calculated using the formula w(i,j)=l / v, where l is the segment length and v is the speed limit of the segment. For node pairs constituting connecting road segments (such as E1(L) to S3(L)), the weight is optimized as follows: ,in This is the empty-run optimization factor (which can be between 0.1 and 0.2). In this embodiment, it is taken as... The distance between connecting road segments L and L is 0 kilometers, and the weight is still 0 after discounting. However, if there is a small actual driving distance (such as 0.004 kilometers) in the connecting road segment, the weight will be discounted and the algorithm will prioritize this path.

[0082] In some examples, the constructed multi-factor calibration model has the following formula:

[0083] Where T(b) is the sum of the basic travel times for each road segment returned by the map API; k(road condition) is the real-time traffic congestion index (where congestion index = real-time vehicle speed / free-flow vehicle speed, which can be set to 1.0 for smooth traffic and 2.0 for congested traffic); k(time period) is the time period coefficient (where 1.3-1.8 is used for morning and evening peak hours 7:00-9:00 and 17:00-19:00, and 1.0-1.2 is used for off-peak hours, etc.); k(weather) is the weather impact coefficient (where 1.1-1.5 is used for severe weather such as rain, snow, and fog, and 1.0 is used for sunny weather). Z is the empty-run optimization coefficient (with the same value as the path weight optimization coefficient, such as 0.1-0.2); Z is the proportion of the length of the connecting road segment to the total path length (where Z = total length of connecting road segment / total path length), and Z = 0 when there is no connecting road segment.

[0084] The step-by-step calculation example is as follows: Assuming a three-person, three-task scenario continuing from Example 1, with clear weather, off-peak hours, and smooth traffic. The basic travel time for each road segment was obtained using the map API: OA is 2 km, speed limit 60 km / h, T(b1) = 2 minutes; AC is 0.0022 km, speed limit 40 km / h, T(b2) = 0.0033 minutes; CL is 0.0028 km, speed limit 40 km / h, T(b3) = 0.0042 minutes; LF is 0.004 km, speed limit 40 km / h, T(b4) = 0.006 minutes; the remaining road segments were calculated similarly. The total basic travel time T(b) = 2.0135 minutes. k(road condition) = 1.0, k(time period) = 1.0, k(weather) = 1.0, the length of the connecting road segment LF is 0.004 km, the total path length is approximately 2.015 km, Z = 0.004 / 2.015 ≈ 0.002. Substituting into the formula: T(f) = [2.0135 × 1.0 × 1.0 × 1.0] × (1 - 0.1 × 0.002) = 2.0135 × 0.9998 ≈ 2.013 minutes. If there are no connecting tasks, LF is the empty return route, Z = 0, then T(f) = 2.0135 × 1.0 = 2.0135 minutes. The comparison shows that the time consumption in the connecting scenario is slightly optimized by about 0.0005 minutes, and crucially, empty runs are avoided.

[0085] In some examples, the heuristic function is set as an alternative to the improved A* algorithm:

[0086] in, The basic adjustment factor (range 0.8-1.2) Let n be the straight-line distance from node n to the current task target point. This is the connection guidance coefficient (generally ranging from 0.3 to 0.5). Let be the straight-line distance from node n to the nearest connected node. When there are no connected nodes in the road network... When the value is 0, the heuristic function degenerates into the form of the traditional A* algorithm.

[0087] set up , When there is a connecting node L, the heuristic value obtained by the algorithm during the search process not only considers the distance from the target point, but also the distance from the connecting node L, so that the search direction expands towards the region that is both close to the target point and close to the connecting node, and prioritizes finding the path through the connecting node.

[0088] The improved path search algorithm in this embodiment incorporates empty-run avoidance guidance during the path planning stage by prioritizing the traversal of connecting nodes and applying weight discounts to connecting road segments. The multi-factor correction model comprehensively considers the influence of road conditions, time of day, and weather, and refines the final travel time by using the proportion Z of connecting road segments. This ensures that the estimated travel time is not only accurate but also quantitatively reflects the empty-run reduction benefits brought about by connecting road segments, providing reliable data support for subsequent carpooling planning and task connection decisions.

[0089] Example 3 This embodiment details the improved DBSCAN density clustering algorithm and the improved TSP dynamic programming algorithm, and how the two work together to generate a vehicle dispatch plan that includes pick-up and drop-off order and driving route.

[0090] Traditional DBSCAN clustering algorithms in carpooling scenarios only cluster based on the spatial distance of trajectory points, failing to identify the "endpoint-startpoint" connection between tasks. This leads to logical confusion in the execution order of tasks that are spatially close, such as "drop-off first, pick-up later," or requiring long empty runs to pick up subsequent tasks after completing the preceding task. Furthermore, traditional TSP dynamic programming algorithms only pursue the shortest total path, neglecting the empty run savings brought about by the "previous endpoint is the subsequent startpoint" connection scenario, resulting in a single optimization objective dimension.

[0091] The inventive concept of this application is to introduce a dual association logic of "task trajectory correlation" and "end-start coherence correlation" into the DBSCAN density clustering algorithm. Using "at least two sets of trajectory points meeting the distance standard, and the distance and time difference between the preceding end point and the subsequent start point meeting the standard" as the clustering criterion, tasks with connectable return journeys are preferentially clustered into the same cluster. Furthermore, a connection marker M is added to the state definition of the TSP dynamic programming algorithm, and an empty-run optimization coefficient is introduced into the state transition equation. The empty mileage Z term enforces the "pick-up first, drop-off later" logic while prioritizing the connection units. The collaboration of these two elements embeds an empty mileage avoidance mechanism throughout the entire process from clustering to route planning.

[0092] Improved DBSCAN clustering: Each complete pick-up and drop-off task is treated as a "start-end" bound trajectory unit. The specific improved DBSCAN clustering process is as follows: Each complete pick-up and drop-off task is defined as a bound trajectory unit containing a start and an end point; the Euclidean distance between trajectory points of different tasks is calculated, and a neighborhood radius, a minimum task number threshold, a task time window overlap threshold, and a return task connection time difference threshold are set; the criteria for determining the correlation between tasks are: the distance between at least two sets of trajectory points is less than or equal to the neighborhood radius, and the distance between the end point of the preceding task and the start point of the following task is less than or equal to the neighborhood radius, while the time difference meets the return task connection time difference threshold; the core tasks that meet the above correlation criteria and their related tasks are iteratively included in the same initial task cluster.

[0093] The following explanation uses a three-person, three-task scenario as an example: The starting point of task T1 is S1(A), and the ending point is E1(L); the starting point of task T2 is S2(C), and the ending point is E2(D); the starting point of task T3 is S3(L), ​​and the ending point is E3(F). Set the core parameter: neighborhood radius. (The core area of ​​the port can be adjusted to 0.5-1.5 km), minimum number of tasks threshold Min(Pts)=2, task time window overlap threshold Time=30 minutes, return task connection time difference threshold F-Time=15 minutes.

[0094] Calculate the Euclidean distance between each task trajectory point: dist(S1, S2) = 0.0022 km ≤ ; dist(S2, E1) = 0.0028 km ≤ ; dist(E1, S3) = 0 km (completely overlapping) ≤ .

[0095] refer to Figure 4a This illustrates that traditional DBSCAN clustering relies solely on spatial distance. Although E1(L) and S3(L) overlap, the clustering lacks connectivity. Figure 4a The dashed elliptical box in the middle indicates a cluster that is only spatially adjacent and does not contain any connection information.

[0096] refer to Figure 4bThis application employs an improved DBSCAN clustering method, using a dual association logic when determining correlation: it requires not only that "at least two sets of trajectory points meet the distance standard," but also that "the distance between the endpoint of the preceding task and the starting point of the subsequent task meets the distance standard, and the time difference satisfies F-Time." Since task T1 arrives at E1(L) at 9:15, and T3 departs from S3(L) at 9:25, the time difference is 10 minutes ≤ F-Time = 15 minutes. Furthermore, E1(L) and S3(L) overlap in the distance determination, thus tasks T1 and T3 satisfy the "endpoint-startpoint" coherent association condition. Task T2 is also included in the association scope because it has two sets of trajectory point distances that meet the standard (S1-S2, S2-E1). The number of associated tasks for the three tasks is ≥ Min(Pts), and they are all marked as core tasks, included in the same initial task cluster, which contains connection information. Therefore, the improved DBSCAN clustering result naturally possesses return journey connection potential, laying the foundation for eliminating empty runs in subsequent TSP route planning.

[0097] If only a single set of trajectory points meets the distance requirement but lacks a coherent "endpoint-startpoint" connection, the corresponding task will not be included in the same cluster to ensure that the clustering result itself has the feasibility of return connection.

[0098] Improved TSP dynamic programming: The state is defined as dp[X][S][M], where X is the currently visited trajectory point, S is a binary status code (e.g., 7 bits, where the k-th bit is 1 to indicate that trajectory point k has been visited), and M is a connection flag (M=1 indicates that the path contains a connection unit, M=0 indicates that it does not). The dp value represents the shortest travel distance in the current state.

[0099] The state transition equation of the improved TSP dynamic programming algorithm is as follows:

[0100] Where X is the currently accessed trajectory point, Y is the previously accessed trajectory point, S is the binary status code, S\{X} is the status after removing X, dp is the shortest travel distance in the current state, and dist(Y,X) is the distance between the two points. The empty-run optimization factor is set to 0.1. Z is the connection marker for the previous state, and Z is the empty mileage corresponding to the connection unit. Z takes the corresponding empty mileage value if and only if Y is the end point of the preceding task and X is the start point of the following task, and the two are the same or adjacent connection units; otherwise, Z=0.

[0101] The step-by-step calculation process is as follows (only key states are shown): Initialization: dp[O][0000001][0]=0; Visiting single point S: dp[S1][0000011][0]=dist(O,S1)=2 km; dp[S2][0000101][0] = 2.2 km; dp[S3][0001001][0] = 2.3 km.

[0102] Critical connection state (O→S1→E1→S3): Starting from the small vehicle depot (i.e., dispatch center O), it first picks up the starting point S1 of task T1, travels to the ending point E1 of task T1 (i.e., delivering vehicle A), and then proceeds to the starting point S3 of task T3. Since E1 and S3 are the same connection node, the connection unit condition is satisfied. At this time, Z=dist(E1,S3)=0 kilometers. dp[S3][0001011][1] = dp[E1][0000011][0] + dist(E1,S3) - 0.1 × 0, at this time dp[S3][0001011][1] = (2 + 0.005) + 0 = 2.005 kilometers. If S3 is accessed before E1 in the state (i.e. the order of receiving before sending is reversed), then it is judged as an invalid state due to the violation of the "receive before sending" constraint and is not retained.

[0103] After incorporating the carpooling logic for task T2, calculate the closed-loop state: dp[O][1111011][1]=Min{dp[E3][0111011][1]+dist(E3,O); And, dp[E2][0111011][1]+dist(E2,O)}≈2.012+2.2=4.212 km.

[0104] The optimal path was obtained: O→S1(A)→S2(C)→E1(L)→S3(L)→E2(D)→E3(F)→O, with a total distance of 4.212 kilometers and an empty driving distance of 0 kilometers.

[0105] In practice, the path logic is as follows: Starting from the dispatch center O, first pick up pilot A for task T1, then pick up pilot B for task T2, send pilot A for task T1 to the destination (which is exactly the starting point of task T3), pick up pilot C for task T3 at the original location, send pilot B for task T2, send pilot C for task T3, and finally return to the dispatch center.

[0106] In this embodiment, the improved DBSCAN algorithm, through dual association logic, ensures that tasks within a cluster have "endpoint-startpoint" continuity during the clustering stage. The selected task clusters are naturally suitable for carpooling and avoid empty runs. The improved TSP dynamic programming algorithm adds a connection marker M and incorporates it into the state transition equation. This allows the algorithm to optimize not only the distance dimension when searching for the optimal path globally, but also to prioritize path schemes that contain connecting units.

[0107] Therefore, by working together, the dual goals of maximizing carpooling efficiency and minimizing empty mileage are achieved while satisfying the "pick-up first, drop-off later" constraint. For example, in a scenario with three people and three tasks, the optimal route achieves 0 kilometers of empty mileage, a total distance of 4.212 kilometers, and a total time of approximately 40 minutes.

[0108] Example 4 This embodiment describes in detail the specific implementation method of planning vehicle task connection based on collaborative filtering and multi-objective decision-making framework.

[0109] In traditional dispatching, after a pilot station vehicle completes a pick-up or drop-off task, its ability to pick up the next task near the destination relies entirely on manual memory and historical records. This is not only time-consuming and labor-intensive, but also difficult to comprehensively cover all possibilities. Moreover, when manually judging connecting tasks, one can only focus on a single dimension (such as distance), lacking a comprehensive consideration of multi-dimensional factors such as time matching degree and task attribute similarity. This leads to unstable connection decision quality and long vehicle waiting times.

[0110] The inventive concept of this application is to model "vehicle task coordination" as a recommendation problem, drawing on the idea of ​​collaborative filtering in recommendation systems. This involves calculating a recommendation score based on the attribute similarity and spatiotemporal connectivity between the current task and historical tasks to be executed. In practice, the spatial distance to the destination and the estimated arrival / departure time difference are used as initial screening criteria. Within the candidate set, collaborative filtering is used to calculate the similarity between tasks. Then, combined with a time connectivity window, a multi-objective decision-making process is used to calculate a comprehensive score and rank the tasks, outputting the optimal coordination task. This upgrades vehicle task coordination decision-making from manual, single-dimensional judgment to a data-driven, multi-dimensional comprehensive evaluation.

[0111] refer to Figure 5 The following is an illustrative explanation of the process of executing task T4: First, obtain the destination location and estimated arrival time of the task currently being performed by the vehicle. For example, the vehicle is performing task T4, the destination is Q, and the estimated arrival time is 14:20.

[0112] Then, an initial screening search was performed in the historical task data warehouse. The preset thresholds for spatial distance and time difference were set to 0.5 kilometers and 15 minutes, respectively. All tasks to be executed were searched if the spatial distance between the origin and point Q was ≤0.5 kilometers and the estimated departure time was ≤15 minutes from 14:20. The search results yielded two candidate connection tasks: Task T5 (origin Q, estimated departure 14:25, spatial distance 0 kilometers, time difference 5 minutes) and Task T6 (origin approximately 0.3 kilometers from Q, estimated departure 14:22, spatial distance 0.3 kilometers, time difference 2 minutes).

[0113] Next, the similarity between the current task T4 and each candidate task is calculated based on collaborative filtering. The similarity calculation considers dimensions such as the geographical features of the task's starting point, the geographical features of its ending point, the execution time period, and historical carpooling frequency. In this embodiment, the similarity between tasks T4 and T5 is 0.81, and the similarity between tasks T4 and T6 is 0.65.

[0114] Furthermore, by incorporating the time connection window, a multi-objective decision-making framework is used to calculate the comprehensive score of each candidate connection task. The scoring dimensions of the multi-objective decision-making framework include similarity, time matching degree (the shorter the time difference, the higher the score), spatial connection degree (the shorter the distance, the higher the score), and the empty driving distance from the vehicle's current position to the starting point of the candidate task. After weighting each dimension, the comprehensive score is calculated as follows: Task T5 has a comprehensive score of 0.85, and Task T6 has a comprehensive score of 0.72.

[0115] Finally, the candidate task T5 with the highest overall score is selected as the optimal connection task, and the connection plan is output to the dispatch management system and pushed to the driver terminal. After the vehicle arrives at point Q at 14:20 to complete the current task, it waits in place for 5 minutes before taking over task T5, achieving continuous operation with no empty mileage.

[0116] This embodiment upgrades manual one-dimensional judgment to data-driven multi-dimensional comprehensive evaluation through collaborative filtering and a multi-objective decision-making framework. It comprehensively covers all possible connection candidates and selects the optimal solution based on quantitative scores, significantly improving the vehicle task connection rate and minimizing vehicle waiting time. Even in scenarios without carpooling, this intelligent connection mechanism still enables continuous vehicle operation, avoiding empty returns and improving single-vehicle operational efficiency.

[0117] Example 5 This embodiment describes in detail the specific implementation method of triggering recalculation and synchronously updating the scheduling scheme when task information changes.

[0118] Because pilotage duties are affected by weather, vessel condition, and dock conditions, pick-up and drop-off times and locations frequently change. Under the traditional manual dispatching model, once a change occurs, the dispatcher needs to notify the driver and pilot by phone one by one, readjust the plan based on experience, and the whole process is slow, error-prone, and difficult to trace.

[0119] Therefore, the technical problem that needs to be solved is how to complete the algorithm recalculation in a very short time and push the adjustment plan to all relevant parties without loss, while ensuring real-time synchronization of information among multiple parties.

[0120] The inventive concept of this application is to construct an instant response chain of "change monitoring - rapid recalculation - targeted push". When the system detects a change in task data, it automatically loads the changed parameters within a preset short time threshold, re-calls modules such as trip time estimation and cluster analysis to perform a full recalculation, and generates an adjusted dispatch plan. Subsequently, the new plan is pushed to each terminal in real time through the system's interface, and the historical task data warehouse is updated synchronously to ensure information consistency among all relevant parties and full traceability of changes.

[0121] refer to Figure 6 The following is an illustrative example using a scenario of two people carpooling and a temporary change in the task: Scenario setting: Morning rush hour 7:30-8:30 (k(time period)=1.5), light rain (k(weather)=1.2), road network congestion (k(road condition)=2.0). Assume there are existing tasks T7 (starting point G, ending point H, 7:30-7:50) and T8 (starting point L, ending point J, 7:35-8:00), and the system has completed the initial scheduling plan generation and execution.

[0122] First, when the pilot station convoy dispatched vehicles to point G to pick up pilot Ding, pilot Ding initiated a change request at time t0 due to a temporary adjustment of the ship's berthing position, requesting that the destination be changed from H to H' (coordinates 117.946, 39.099). At this time, the system detected the change in task T7 at time t0+0.5s. During implementation, the change information was captured by the scheduling management system through the data monitoring interface.

[0123] Then, within a preset time threshold (e.g., 2 seconds), the system pushes the changed task T7 parameters (end point H') to the intelligent algorithm module (where the intelligent algorithm module can integrate and implement one or more algorithms such as time estimation, clustering, vehicle dispatching plan generation, task connection, and dynamic adjustment as mentioned in the previous example), triggering a full-process recalculation.

[0124] Full recalculation coverage: trip time estimation (introducing a new destination H' to recalculate the time of each segment of all possible routes), cluster analysis (verifying whether tasks T7 and T8 still meet carpooling and connection constraints), and route planning (re-executing improved TSP dynamic programming with H' as the target).

[0125] After recalculation at time t0+1.5s, the adjusted dispatch plan is generated: the new route is O→G→I→H'→J→O, and the total route time is updated to 42 minutes. Verification shows that after task T7 is changed to H', point J can still be used as an alternative starting point for the next connecting task, and the connection feasibility of task T4 (starting point J, estimated departure 8:05) is not affected.

[0126] After generating the adjustment plan, the revised dispatch plan will be pushed in real time to the driver terminals of the affected drivers and the pilot terminal of pilot Ding via the system interface. The push content includes details such as the updated pick-up and drop-off order, the new driving route, and the estimated arrival time.

[0127] At the same time, information such as the executor (dispatcher or pilot), operation content (the destination changes from H to H'), operation time, and recalculation plan will be used as change data, synchronously updated and archived to the historical task data warehouse to form a complete change log.

[0128] This embodiment's closed-loop real-time response mechanism enables the entire process from listening to change information to recalculating and issuing the solution to be completed within seconds, greatly shortening the long response latency (usually measured in minutes or even tens of minutes) required by traditional methods such as telephone communication, manual re-estimation, and scheduling. Furthermore, even under the combined effects of peak congestion and inclement weather, the system can still accurately control the time consumption and output feasible solutions. Simultaneously, end-to-end change tracking and archiving ensure the traceability and auditability of the scheduling process, improving the robustness and management standardization of the entire scheduling system.

[0129] Example 6 This embodiment describes an example of how a change log is generated during the online entry of pick-up and drop-off task information.

[0130] In the traditional handwritten record and telephone communication mode, task changes lack effective traceability. Once scheduling disputes occur or scheduling efficiency is analyzed in post-event review, it is difficult to trace the initiator, content and time of each change, which leads to difficulties in determining responsibility and lack of data support for experience extraction.

[0131] The inventive concept of this application is to embed an automated operation logging function during the online entry and modification of task information in the scheduling and management system. Each time a new or modified operation is performed on task data, the system captures the identity of the currently logged-in operator, the modified data fields and their old and new values, as well as the timestamp of the operation, forming a structured change log record and automatically archiving it, thereby achieving complete tracking of the task lifecycle change history.

[0132] An example is given below: First, the dispatcher creates a new task T1 on the dispatch management system, entering the starting point G, destination H, and pick-up / drop-off time 7:30-7:50. When the system saves the task data, it automatically generates a change log record L1: the operator is "Dispatcher Zhang San", the operation content is "Create Task T1", the operation time is "20xx-xx-xx 07:00:00", and the operation details include complete task starting point, destination, and time information.

[0133] Secondly, during the mission execution, the dispatcher received a request from pilot Ding to change the endpoint, and modified the endpoint of mission T1 from H to H' in the system. When the system saved the modification, it automatically generated a change log record L2: the operator was "dispatcher Zhang San", the operation content was "modify the endpoint of mission T1", the operation time was "20xx-xx-xx 07:32:15", and the operation details were "endpoint H→H', new coordinates (117.946, 39.099)".

[0134] Finally, all change log records are automatically archived and stored in the log table of the historical task data warehouse in real time, and can be retrieved and exported by task number, time period, operator, and other dimensions.

[0135] The automatic change log mechanism in this embodiment achieves end-to-end traceability of the task lifecycle with zero additional manual cost. Schedulers do not need to manually record changes; the system automatically completes operation recording with time accuracy down to the second. Through traceability, not only is objective evidence provided for post-event dispute resolution, but also high-quality historical data is accumulated for analyzing scheduling efficiency and optimizing algorithm parameters (such as adjusting time period coefficients and weather coefficients), supporting continuous iterative optimization of the algorithm.

[0136] Example 7 This embodiment describes the overall architecture of the pilot station vehicle fleet dispatching system and the collaborative working method of each module.

[0137] In the traditional scheduling model, information recording, task planning, and historical data query are all operated in a fragmented manner, relying on manual relay and transmission, which makes it impossible to form an efficient data flow closed loop.

[0138] Therefore, it is necessary to build an integrated system architecture that organically integrates information management, intelligent algorithms, data storage and dynamic adjustment functions to achieve real-time data communication and collaborative operation between modules.

[0139] The inventive concept of this application: Reference Figure 7 The system employs a modular architecture, dividing it into four functional components: an "integrated information management module," an "intelligent algorithm module," a "historical task data warehouse module," and a "collaboration and dynamic adjustment module." These modules communicate in real-time through standardized data interfaces. The information management module is responsible for collecting and synchronizing data from multiple sources; the intelligent algorithm module handles core scheduling calculations; the historical data warehouse is responsible for data accumulation and seamless recommendation; and the collaborative adjustment module monitors changes and coordinates the responses of all modules. These four components form a closed-loop scheduling system of "data collection - algorithm decision-making - data accumulation - dynamic feedback."

[0140] The following is an illustrative example of the implementation: Integrated information management module: Provides operation interface on PC and / or mobile terminal, supports dispatchers to enter task information online; synchronizes task information to driver and pilot terminals in real time through message push mechanism; automatically records every operation log; receives confirmation status feedback from driver and pilot and updates it to the dispatch terminal in real time, forming a three-party information closed loop.

[0141] The intelligent algorithm module receives task data from the integrated information management module and includes a trip time estimation unit, a clustering unit, and a vehicle dispatching and route planning unit. The trip time estimation unit interfaces with the map API and uses an improved route search algorithm and a multi-factor correction model to calculate accurate travel time. The clustering unit performs dual-association clustering, such as using an improved DBSCAN clustering algorithm to form task clusters. The vehicle dispatching and route planning unit generates the optimal pick-up and drop-off order and route using dynamic programming algorithms (such as the improved TSP dynamic programming algorithm), and outputs the calculation results in the form of a vehicle dispatching plan.

[0142] Historical Task Data Warehouse Module: Stores all completed and pending pick-up and drop-off task data (including carpooling plan details, actual mileage, change logs, etc.), supporting multi-dimensional retrieval. Based on the current task destination and estimated arrival time, it retrieves and recommends the best connecting task from the warehouse.

[0143] Collaboration and Dynamic Adjustment Module: Continuously monitors task data change events. Once a change is detected, the intelligent algorithm module is triggered to recalculate the task plan within a preset time. The recalculated adjustment plan is pushed to relevant terminals (such as the driver terminal of the affected driver, the pilot terminal of the pilot, etc.) through the integrated information management module, and the changed data is synchronized to the historical task data warehouse for storage.

[0144] The system architecture in this embodiment deeply couples the various functional components, eliminating the need for manual data transfer. From task generation to solution issuance, from change monitoring to recalculation push, the entire process is automatically connected. At the same time, the historical data warehouse continuously provides a data foundation for algorithm model iteration, and the algorithm decision results continuously enrich the data warehouse, forming a continuous evolutionary closed loop of "algorithm decision → system execution → data accumulation → model optimization," which is significantly more efficient than the fragmented operation of each link in the traditional model.

[0145] This embodiment also describes the composition and working logic of units such as the trip time estimation unit, clustering unit, and path planning unit in the intelligent algorithm module.

[0146] The intelligent algorithm module needs to simultaneously undertake three major computational tasks: trip time estimation, carpooling cluster analysis, and route planning. These tasks have data dependencies (the time estimation results provide segment time data for route planning, and the clustering results provide task grouping for route planning). How to design the internal unit structure of the module so that each algorithm component can run independently and collaborate seamlessly is a key issue at the system implementation level.

[0147] Preferably, the intelligent algorithm module is divided into sub-units such as a "trip time estimation unit," a "clustering unit," and a "vehicle dispatching and route planning unit." These sub-units share a data structure to transmit information about connecting road segments and task clusters. The trip time estimation unit integrates two optional engines in parallel: an improved Dijkstra algorithm and an A* algorithm. The clustering unit and the vehicle dispatching and route planning unit integrate two algorithms in series: an improved DBSCAN algorithm and an improved TSP algorithm. The units employ a "prediction first, planning later" sequential calling logic to ensure that the most accurate time data for each road segment is obtained during the route planning stage.

[0148] Trip Time Estimation Unit: The workflow includes three sub-steps. (1) Map API Interface Submodule obtains real-time traffic conditions and basic travel time for each road segment. (2) Algorithm Engine Selection Submodule automatically selects the improved Dijkstra algorithm or the improved A* algorithm to generate the basic path according to the task scale, and performs discount processing on connecting road segments in the path weight. (3) Correction Model Submodule runs the multi-factor correction formula, and outputs the final estimated travel time by combining traffic conditions, time period, weather and the proportion of connecting road segments.

[0149] Clustering Unit, Dispatch Scheme, and Route Planning Unit: The workflow is divided into two stages. In the first stage, the clustering unit runs an improved DBSCAN clustering algorithm, using dual-association criteria to cluster concurrent tasks. In the second stage, route planning, the clustering results from the first stage are read, and an improved TSP dynamic programming algorithm is run on each task cluster. In the state definition, connection markers are used to track connecting units, and in the state transition equation... The selection prioritizes routes that include connections and outputs the optimal pick-up / drop-off order and driving route.

[0150] This embodiment, through clear unit division and data interface design, enables the improvement of the trip time estimation algorithm to be independent of the carpooling and route planning algorithms, allowing both to be maintained and iteratively upgraded in parallel. Connecting road segment information, as a key transmission parameter between the two units, ensures that considerations for empty-run optimization in trip estimation can be carried over to the route planning stage, achieving full-link empty-run avoidance.

[0151] Based on the same inventive concept, this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform: a pilot station fleet pick-up and drop-off task scheduling method as described in any embodiment of this application.

[0152] like Figure 8 As shown in the figure, the present application also provides a schematic diagram of the structure of an electronic device 500. The electronic device 500 is merely an example and should not be construed as limiting the function and scope of use of the embodiments of the present application.

[0153] The electronic device 500 may include: at least one processor 510; and a memory 520 communicatively connected to the at least one processor; wherein the memory 520 stores instructions executable by the at least one processor 510, the instructions being executed by the at least one processor 510 to enable the at least one processor 510 to perform: the pilot station fleet pick-up and drop-off task scheduling method according to any embodiment of this application.

[0154] It should be noted that the electronic device 500 can be represented in the form of a general-purpose computing device, such as a server device.

[0155] In practice, the components of electronic device 500 may include, but are not limited to: at least one processor 510, at least one memory 520, and a bus 530 connecting different system components (including memory 520 and processor 510), wherein bus 530 may include a data bus, an address bus, and a control bus.

[0156] In implementation, memory 520 may include volatile memory, such as random access memory (RAM) 5201 and / or cache memory 5202, and may further include read-only memory (ROM) 5203.

[0157] The memory 520 may also include a program tool 5205 having a set (at least one) of program modules 5204, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0158] The processor 510 performs various functional applications and data processing by running computer programs stored in the memory 520.

[0159] Electronic device 500 can also communicate with one or more external devices 540 (e.g., keyboard, pointing device, etc.). This communication can be performed through input / output (I / O) interface 550. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 560, which communicates with other modules in electronic device 500 via bus 530. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 500, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0160] Based on the same inventive concept, embodiments of this specification provide a computer storage medium storing computer-executable instructions, which are configured as: a pilot station vehicle fleet dispatching method as described in any embodiment of this application.

[0161] It should be noted that the computer storage medium may include, but is not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0162] In possible implementations, this application may also provide a form of data processing implemented as a program product, which includes program code that, when the program product is run on a terminal device, causes the terminal device to perform several steps of the method described in any of the foregoing embodiments.

[0163] The program code for executing this application can be written in any combination of one or more programming languages, and the program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0164] While specific embodiments of this application have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this application is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this application, but all such changes and modifications fall within the scope of protection of this application.

Claims

1. A method for scheduling a pilot station vehicle fleet for pick-up and drop-off tasks, characterized by, include: The pick-up and drop-off task information is entered online on the dispatch management system. The pick-up and drop-off task information includes the starting point, the destination and the pick-up and drop-off time, and the pick-up and drop-off task information is synchronized to the driver terminal and the pilot terminal in real time. The recorded pick-up and drop-off tasks are estimated to take travel time. When calculating the route cost, the travel time estimation assigns a lower toll cost weight to the connecting road segments that form the route from the end of the preceding task to the start of the following task than to the regular road segments, so as to prioritize the construction of travel routes that include connecting road segments. Cluster analysis is performed on multiple concurrent tasks. When determining the correlation between tasks, the spatial distance and time difference between the end point of the preceding task and the starting point of the subsequent task are used as necessary judgment dimensions, so that task combinations with a connection relationship are preferentially included in the same task cluster. Based on the estimated trip time and the results of the cluster analysis, a vehicle dispatch plan including pick-up and drop-off order and driving route is generated. Based on the destination location and estimated arrival time of the current task, it is matched with the starting location and estimated departure time of the tasks to be executed in the historical task data warehouse to plan the connection between the current task and the next task to be executed, forming a coherent operation sequence. When mission information changes, a recalculation is triggered and the updated scheduling plan is synchronized to the driver's terminal and the pilot's terminal. An improved TSP dynamic programming algorithm is used to generate a vehicle dispatch plan that includes pick-up and drop-off order and driving route, including: With the goal of minimizing total driving distance and vehicle empty driving distance, constraints are set including pick-up and drop-off priority, vehicle passenger capacity, time window, and route closed loop. In the algorithm state definition, a connection marker is introduced to record whether the current path contains a connection unit between the end point of the preceding task and the start point of the following task. In the state transition equation, when the path contains the connecting unit, an empty-run optimization coefficient is introduced to calculate the reduction in empty-run mileage brought about by the connecting unit, so as to prioritize the selection of paths containing connecting units during the search process. The state transition equation of the improved TSP dynamic programming algorithm is as follows: wherein X is the current visiting track point, Y is the last visiting track point, S is a binary state code, dp is the shortest driving distance in the current state, dist(Y, X) is the distance between the two points, is the driving distance optimization coefficient, is the connection mark of the last state, and Z is the driving distance corresponding to the connection unit; Z takes the corresponding driving distance value only when the current sequence task end point and the subsequent sequence task start point are the same or adjacent connection units, otherwise Z takes the value of 0.

2. The pilot station vehicle convoy dispatching method according to claim 1, characterized in that, The trip time estimation steps include: Connect to the map API to obtain real-time traffic data and the basic travel time for each road segment; An improved path search algorithm is used to generate basic driving paths. When calculating path weights, for connecting road segments where the endpoint of the preceding task coincides with or is adjacent to the starting point of the following task, an empty driving optimization coefficient is introduced to reduce the toll cost weight of the connecting road segment. The base travel time is corrected by combining real-time traffic index, time period coefficient and weather impact coefficient. Based on the proportion of the length of connecting road segments in the total travel path, the travel time is adjusted again by the empty-run optimization coefficient, and the final estimated travel time and candidate path are output.

3. The pilot station vehicle convoy dispatching method according to claim 2, characterized in that, A multi-factor correction model is constructed to correct the baseline time consumption based on the multi-factor correction model; wherein the multi-factor correction model adopts the following formula: Wherein, T(f) is the final estimated time, T(b) is the base time returned by the map API, k(traffic) is the real-time traffic congestion index, k(time period) is the time period coefficient, and k(weather) is the weather impact coefficient. Z is the empty-run optimization coefficient, and Z is the proportion of the length of the connecting road segment to the total path length.

4. The pilot station vehicle convoy dispatching method according to claim 2, characterized in that, The improved path search algorithm is an improved Dijkstra algorithm. When the improved Dijkstra algorithm is executed, nodes in the road network that are the endpoints of the preceding tasks and the starting points of the following tasks are marked as connecting nodes. After each selection of the current node, the starting nodes of the following tasks associated with the connecting node are traversed first.

5. The pilot station vehicle convoy dispatching method according to claim 2, characterized in that, The improved pathfinding algorithm is an improved A* algorithm, and the heuristic function of the A* algorithm is: in, Basic adjustment coefficient, Let n be the straight-line distance from node n to the current task target point. To connect the guiding coefficient, Let n be the straight-line distance from node n to the nearest connected node; when there is no connected node, ... The value is 0.

6. The pilot station vehicle convoy dispatching method according to claim 1, characterized in that, The clustering analysis employs an improved DBSCAN density clustering algorithm, including: Each complete pick-up and drop-off task is defined as a bound trajectory unit containing a start point and an end point; Calculate the Euclidean distance between trajectory points of different tasks, and set the neighborhood radius, minimum number of tasks threshold, task time window overlap threshold, and return task connection time difference threshold. The criteria for determining the correlation between tasks are: the distance between at least two sets of trajectory points is less than or equal to the neighborhood radius, the distance between the end point of the preceding task and the start point of the following task is less than or equal to the neighborhood radius, and the time difference meets the return task connection time difference threshold. The core tasks that satisfy the above correlation determination and their related tasks are iteratively included into the same initial task cluster.

7. The pilot station vehicle convoy dispatching method according to claim 1, characterized in that, The coordination between the current task and the next task to be executed for the planned vehicle includes: Obtain the destination location and estimated arrival time of the task currently being performed by the vehicle; In the historical task data warehouse, all tasks to be executed that are less than a preset threshold in terms of the spatial distance between the origin and the destination and less than a preset time difference threshold in terms of the difference between the estimated departure time and the estimated arrival time are retrieved and used as candidate connection tasks. Based on collaborative filtering, the similarity between the current task and each of the candidate connection tasks is calculated. Combined with the time connection window, the comprehensive score of each candidate connection task is calculated through multi-objective decision-making, and the candidate connection task with the highest score is selected as the optimal connection task.

8. The pilot station vehicle convoy dispatching method according to claim 1, characterized in that, The process of triggering a recalculation and synchronizing the updated scheduling scheme to the driver's terminal and the pilot's terminal includes: When a change in task information is detected, the trip time estimation and cluster analysis are recalculated based on the changed task parameters within a preset time threshold, and an adjusted dispatch plan is generated. The adjusted dispatch plan is pushed to the driver terminal and the pilot terminal in real time, and the changed data is synchronously updated to the historical task data warehouse.

9. The pilot station vehicle convoy dispatching method according to claim 1, characterized in that, The online entry of pick-up and drop-off task information further includes: automatically recording and archiving the executor, operation content, and operation time of each task change, forming a change log.

10. A pilot station vehicle convoy dispatching system, characterized in that, include: The integrated information management module is used to realize the online input, real-time synchronization and closed-loop confirmation of task information between the dispatch terminal, driver terminal and pilot terminal; The intelligent algorithm module is used to estimate the travel time of received pick-up and drop-off tasks and perform cluster analysis on multiple concurrent tasks to generate a dispatch plan that includes the pick-up and drop-off order and driving route. Specifically, when calculating the route cost, the travel time estimation assigns a lower toll cost weight to connecting road segments from the endpoint of a preceding task to the starting point of a subsequent task, prioritizing the construction of driving routes that include these connecting road segments. When determining the correlation between tasks, the cluster analysis uses the spatial distance and time difference between the endpoint of a preceding task and the starting point of a subsequent task as necessary judgment dimensions, ensuring that task combinations with a correlation relationship are preferentially included in the same task cluster. The historical task data warehouse module is used to store historical and current task data. Based on the destination location and estimated arrival time of the current task, it matches the starting location and estimated departure time of the tasks to be executed in the historical task data warehouse to plan the connection between the current task and the next task to be executed, forming a coherent operation sequence. The coordination and dynamic adjustment module is used to trigger the intelligent algorithm module to recalculate when the task information changes, and synchronize the updated scheduling scheme to the driver terminal and the pilot terminal.

11. The system according to claim 10, characterized in that, The intelligent algorithm module includes a trip time estimation unit, which is used for: Connect to the map API to obtain real-time traffic data and the basic travel time for each road segment; An improved path search algorithm is used to generate basic driving paths. When calculating path weights, for connecting road segments where the endpoint of the preceding task coincides with or is adjacent to the starting point of the following task, an empty driving optimization coefficient is introduced to reduce the toll cost weight of the connecting road segment. A multi-factor correction model is constructed, which combines real-time traffic index, time period coefficient and weather impact coefficient to correct the basic travel time. Based on the proportion of the length of connecting road segments in the total travel path, the travel time is adjusted again by the empty driving optimization coefficient, and the final estimated travel time and candidate path are output. And / or, the intelligent algorithm module includes a carpooling matching and route planning unit, which is used for: An improved DBSCAN density clustering algorithm is used to define each complete pick-up and drop-off task as a bound trajectory unit containing a start point and an end point. The Euclidean distance between trajectory points of different tasks is calculated, and the core task and its associated tasks that meet the correlation determination are iteratively included in the same initial task cluster. The correlation determination is based on the following criteria: the distance between at least two sets of trajectory points is less than or equal to a preset neighborhood radius, and the distance between the end point of the preceding task and the start point of the following task is less than or equal to the preset neighborhood radius, while the time difference meets the preset return task connection time difference threshold. An improved TSP dynamic programming algorithm is adopted. A connection marker is introduced into the state definition to record whether the path already contains a connection unit. An empty-run optimization coefficient is introduced into the state transition equation to calculate the reduction of empty-run mileage brought by the connection unit, so as to generate the pick-up and drop-off sequence and driving path containing connection units.

12. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform: the pilot station fleet dispatching method as described in any one of claims 1 to 9.

13. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which are executed by a processor when executed: the pilot station vehicle convoy dispatching method as described in any one of claims 1 to 9.

Images