Search

CN-121998376-A - Customized bus route generation method and system based on passenger travel big data

CN121998376ACN 121998376 ACN121998376 ACN 121998376ACN-121998376-A

Abstract

The invention discloses a method and a system for generating a customized bus route based on large data of a passenger trip, wherein the method acquires an OD record of the passenger, a card swiping record of the bus, a platform monitoring video, road network data and bus stop data; the method comprises the steps of extracting the number of waiting vehicles, the number of getting on and off vehicles, the number of people not getting on the vehicles after leaving the stops and average stop time through target detection and multi-target tracking, constructing a space-time requirement tensor, inputting a space-time diagram attention network model to predict the getting on and off requirements and congestion risks, extracting a tide travel requirement corridor by combining historical OD distribution and a time impedance minimum path to generate a virtual stop point candidate set, inserting the virtual stop point under the constraint of vehicle capacity, adjacent stop distance and single-stop detour time to form a customized bus route and generate a departure shift and inter-stop schedule. The scheme improves the accuracy of demand prediction, the pertinence of route generation and the operation feasibility.

Inventors

  • LI SHUO
  • DENG YUKUN
  • ZENG XING

Assignees

  • 东莞市规划设计研究院有限公司

Dates

Publication Date
20260508
Application Date
20260331

Claims (10)

  1. 1. The method for generating the customized bus route based on the passenger trip big data is characterized by comprising the following steps: S1, acquiring an OD record of a passenger, a bus card swiping record, a platform monitoring video, road network data and bus station data; S2, carrying out target detection and multi-target tracking on the platform monitoring video, and extracting the number of waiting vehicles, the number of getting on and off the vehicle, the moment of getting off the vehicle, the number of people not getting on the vehicle after the vehicle leaves the station and average time consumption for stopping the station in each time window; s3, mapping OD recording, card swiping recording and video extraction features to a hexagonal space grid and a time window, and constructing a space-time requirement tensor representing the grid travel requirement and the site crowding state; s4, inputting the space-time demand tensor into a space-time diagram attention network model, predicting the on-off demand and congestion risk of each grid in the future period, constructing an OD intensity matrix by combining historical OD distribution, and extracting a tide travel demand corridor by combining a road network time impedance minimum path; S5, generating a line skeleton by taking existing bus stops at two ends of the demand corridor as starting and ending points, and generating a virtual stop point candidate set by combining a road side stoppable area, road width, stop forbidden information and queuing overflow risk; and S6, under the constraint of vehicle capacity, adjacent station distance and single station detour time, virtual stop points are inserted into a line skeleton one by one according to the maximum net gain increment of the line, a target customized bus route is generated, and a departure shift and inter-station timetable is generated according to the predicted passenger flow, the vehicle capacity, the historical road section transit time and the average stop time.
  2. 2. The method for generating a customized bus route based on passenger travel big data according to claim 1, wherein in the step S2, when target detection and multi-target tracking are performed on the platform monitoring video, a waiting area and a door area are first identified, then the number of passengers getting on and off is counted according to the passing event of the passenger track and the door area, and the number of passengers not getting on after the vehicle leaves the waiting area is determined according to the number of passengers still located in the waiting area after the vehicle leaves the waiting area.
  3. 3. The customized bus route generation method based on passenger travel big data according to claim 2, wherein the average stop time is determined according to a difference between a stop arrival time and a stop departure time of the same vehicle, the stop arrival time is a time when a front edge of the vehicle enters a stop zone, and the stop departure time is a time when a tail edge of the vehicle leaves the stop zone.
  4. 4. The method for generating a customized bus route based on passenger travel big data according to claim 1, wherein the space-time requirement tensor at least comprises a grid boarding requirement feature, a grid alighting requirement feature, a station waiting number feature, a number of people not boarding after leaving a station, an average station stopping time consuming feature and a station crowding state feature, wherein each feature is aligned and coded according to a hexagonal space grid and a time window.
  5. 5. The method for generating a customized bus route based on passenger travel big data according to claim 1, wherein the space-time diagram attention network model comprises an input embedded layer, a space diagram attention layer, a time convolution layer and a prediction output layer, and a corridor prior gating unit is inserted between the space diagram attention layer and the time convolution layer, and the corridor prior gating unit generates gating coefficients according to a time impedance minimum path overlap ratio, a road stop mark, a queuing overflow risk and virtual stop feasibility, and enhances or suppresses node characteristics and adjacent side weights.
  6. 6. The method for generating the customized bus route based on the passenger travel big data according to claim 5, wherein the space-time diagram attention network model adopts a double-adjacency matrix, the double-adjacency matrix comprises a space adjacency matrix constructed based on a road communication relation and a demand adjacency matrix constructed based on a historical OD flow intensity, and the corridor prior gating unit respectively performs weighted fusion on attention weights corresponding to the two adjacency matrices.
  7. 7. The method for generating the customized bus route based on the passenger travel big data according to claim 1, wherein when the tidal travel demand corridor is extracted in the step S4, an OD pair which is larger than a preset threshold in an OD intensity matrix is selected, superposition statistics is performed on a time impedance minimum path corresponding to each OD pair, and a continuous road section with the number of path superposition times being larger than a preset number of times and the direction consistency being larger than a preset proportion is determined as the tidal travel demand corridor.
  8. 8. The method for generating customized bus route based on passenger travel big data according to claim 1, wherein when generating the virtual stop point candidate set in S5, sampling road points according to a preset distance interval on a demand corridor, and screening sampling points which satisfy that the distance from an existing bus stop is larger than a preset rejection distance, the distance from a road intersection is larger than a preset safety distance, the road section is not a forbidden road section and the road width is larger than a preset threshold value as the virtual stop point candidate points.
  9. 9. The method for generating the customized bus route based on the passenger traveling big data according to claim 1, wherein the queuing overflow risk is determined according to the relation between the length of a waiting queue and the safety boundary of a platform, when the tail end of the waiting queue exceeds the boundary of the platform or invades a motor vehicle lane, the corresponding site is marked as a site with high overflow risk, the priority that the site and the adjacent sampling points thereof are selected as virtual stop points is reduced, the net return of the line in the S6 is obtained by subtracting the cost of newly increased bypass time, the cost of repeated service penalty and the cost of congestion penalty from the newly increased return of the covering passenger flow, wherein the cost of repeated service penalty is determined according to the overlapping proportion of the virtual stop points and the service radius of the existing bus site, and the cost of congestion penalty is determined according to the congestion risk of the corresponding site.
  10. 10. Customized bus route generation system based on passenger trip big data, which is characterized by comprising: The data acquisition module is used for acquiring the OD records of passengers, the bus card swiping records, the platform monitoring video, the road network data and the bus station data; The video analysis module is used for carrying out target detection and multi-target tracking on the platform monitoring video and extracting the number of waiting vehicles, the number of getting on and off the vehicle, the moment of getting off the vehicle, the number of people not getting on the vehicle after the vehicle leaves the station and average time consumption for stopping the station in each time window; The demand tensor construction module is used for mapping the OD record, the card swiping record and the video extraction characteristics to a hexagonal space grid and a time window to construct a space-time demand tensor representing the grid travel demand and the site crowding state; The demand prediction and corridor extraction module is used for inputting the space-time demand tensor into a space-time diagram attention network model, predicting the on-off demand and congestion risk of each grid in the future period, constructing an OD intensity matrix by combining historical OD distribution, and extracting a tidal travel demand corridor by combining a road network time impedance minimum path; The candidate point generation module is used for generating a line skeleton by taking existing bus stops at two ends of the demand corridor as starting and ending points, and generating a virtual stop point candidate set by combining a road side stoppable area, road width, stop prohibition information and queuing overflow risk; The route generation and scheduling module is used for inserting virtual stop points into a route framework one by one according to the maximum net gain increment of the route under the constraint of vehicle capacity, adjacent station distance and single station detour time to generate a target customized bus route, and generating a departure shift and inter-station timetable according to the predicted passenger flow, the vehicle capacity, the historical road section transit time and the average station stop time.

Description

Customized bus route generation method and system based on passenger travel big data Technical Field The invention relates to the technical field of path planning, in particular to a method and a system for generating a customized bus route based on passenger travel big data. Background Along with the continuous expansion of urban scale, continuous deepening of the separation degree of liveness and the increasingly diversified commuting demands, the demands of passengers on high efficiency, quasi-punctualization and differentiated traveling are difficult to meet in partial areas of the traditional fixed bus line. Particularly, in industrial parks, large living areas, transportation hubs, schools, hospitals and business offices, passenger travel often exhibits significant tidal, aggregate and repetitive nature, forming relatively stable travel aisles during peak morning and evening hours. The conventional public transportation line is generally arranged according to long-term statistical experience, the line adjustment period is long, and the regional passenger flow change is difficult to respond timely, so that on one hand, part of the line is full for a long time and is serious in congestion, and on the other hand, part of the line has the problems of idle operation capacity and high idle running rate. Therefore, how to dynamically generate the customized bus route more in line with the travel rule of the passengers based on the actual travel demands becomes an important research direction in the intelligent traffic field. Currently, aiming at route planning of customized buses or demand response buses, the prior art generally carries out statistics on passenger demands through reservation information, card swiping records, historical OD data or manual questionnaire results, and carries out route design or station selection according to demand hot spots. The scheme can improve the matching degree between the line and the demand to a certain extent, but still has more limitations. Firstly, the prior art lacks effective perception on the actual waiting behavior of passengers, the crowded degree of stations, the stopping time of vehicles and the situations of people who cannot get on the vehicles after full loading, and only depends on card swiping records, reservation records or historical OD data to carry out demand analysis, the card swiping data only can reflect the finished riding behavior and hardly reflect the potential unmet demands, reservation data is easily limited by active filling will of users, and all the people who go out are difficult to be covered comprehensively. This results in a deviation between the generated line plan and the actual site operating state. Secondly, in the prior art, single time sequence statistics, cluster analysis or a simple graph model is mostly adopted for demand prediction in the passenger flow modeling process, and only the history demand scale can be generally described, so that the spatial association relationship between different areas, the dynamic change relationship between different time windows and the influence of the external operation environment on the passenger flow are difficult to simultaneously describe. Especially under the condition of road traffic condition change, bus station crowding degree change or local area demand sudden increase, the existing method has insufficient representation capability on the passenger flow space-time evolution trend, and further the accuracy of subsequent line generation is affected. Thirdly, the existing customized bus route generation technology mostly regards stations as fixed facilities, mainly screens and combines the existing bus stations, and lacks a mechanism for automatically finding virtual stop points along a demand corridor and performing fine constraint screening. For the scene that the existing stations are unreasonable in layout and deviate greatly from the actual boarding and disembarking positions, only the existing stations are relied on for line organization, so that the problems of overlong walking distance of passengers, low boarding and disembarking distribution efficiency and large line detouring are easily caused. Meanwhile, some existing schemes are insufficient in consideration of parking feasibility factors such as road width, forbidden parking conditions, queuing overflow risks and the like, so that safety or implementation problems can exist when the generated route is actually landed. In the existing space-time modeling method facing bus passenger flow prediction, a fixed adjacency-based graph neural network, a time sequence convolution network or an attention network is generally adopted to predict the station or grid passenger flow, but most of the space-time modeling method only utilizes road topology adjacency relations or historical flow association relations to conduct characteristic propagation, and prior constraints directly related to custom bus formation such as tidal travel c