A Novel Trajectory Based Prediction Method for Urban Subway Design

Zhi Cai, Jiawei Wang, Tong Li, Bowen Yang, Xing Su, Limin Guo, Zhiming DingISPRS Int. J. Geo Inf 2022 · DOI

TL;DR

This paper introduces a novel method for predicting optimal urban subway station locations by integrating taxi trajectory data with a weighted LeaderRank algorithm and Gaussian Mixture Models. By analyzing passenger flow patterns and land use characteristics, the approach aims to improve subway planning and better meet urban residents' travel needs.

What Problem Does This Paper Address?

Planning new subway stations is a complex and costly endeavor, with traditional methods often relying on static factors like population density and land use, neglecting the dynamic travel behaviors of urban residents. This can lead to a mismatch between station locations and actual demand, hindering the effectiveness of public transportation systems. The paper addresses this gap by proposing a data-driven approach that leverages real-world travel patterns from taxi data to predict future subway station locations more scientifically and effectively.

Methods

The authors propose a multi-stage methodology:

  • Weighted Subway Network Construction: They build a weighted subway network using the "Space P" method, where nodes represent stations and directed edge weights represent passenger flow between them.
  • Weighted LeaderRank: To rank the importance of existing subway stations, they employ a weighted LeaderRank algorithm. This algorithm considers not only passenger flow but also the direction of travel and the dynamic network relationships, providing a more nuanced importance score than traditional methods.
  • Gaussian Mixture Model (GMM) for Prediction: Taxi Origin-Destination (OD) data is analyzed. Assuming a two-dimensional Gaussian distribution for taxi OD data (closer to the city center, higher density), GMM clustering is used to identify potential areas for new subway stations, particularly in regions outside the current subway network coverage.
  • Land Use Analysis: After predicting potential station areas, the paper analyzes the land use characteristics within these regions to understand their suitability for new stations.

Datasets

  • Taxi GPS Trajectory Data: Collected from 21,590 taxis in Beijing during June 2013. This data includes terminal ID, license plate, timestamp, latitude, longitude, speed, and vehicle status.
  • Beijing Subway Station Data: Information on existing subway stations (built up to 2013), new stations built from 2013 to 2020 (96 stations), and planned subway stations (84 stations). This data was obtained from OpenStreetMap.

Key Insights and Ideas

  • Beyond Static Factors: The research highlights the critical need to move beyond static factors like population density and land use for subway planning. Incorporating dynamic travel behavior, as captured by taxi trajectories, is crucial for more accurate predictions.
  • LeaderRank for Station Importance: The weighted LeaderRank algorithm proves effective in identifying key subway stations that are hubs of passenger activity, suggesting that new stations might be beneficial in their vicinity to alleviate congestion.
  • GMM for Spatial Prediction: The application of GMM to taxi OD data effectively identifies areas with high travel demand that are underserved by the current subway network, making it a powerful tool for spatial prediction.
  • Land Use as a Complementary Factor: While GMM identifies high-demand areas, analyzing land use characteristics provides further context, suggesting that areas with a mix of business, education, and transportation hubs are more likely candidates for new stations than purely residential areas.

Key Results

  • The proposed method predicted 200 potential station areas, of which 115 areas contained new subway stations built after 2013. This indicates a prediction accuracy of 57.5% for stations built after 2013.
  • For specific busy stations (Jinsong, Dongzhimen, Dawanglu, Sanyuanqiao, and Chongwenmen), the method predicted five adjacent station areas, correctly covering 4, 4, 2, and 5 new stations respectively.
  • The analysis of land use characteristics in predicted regions revealed that areas with a combination of business, education, and transportation hubs are strong indicators for future subway station locations.

Limitations and Future Work

  • User Behavioral Hypotheses: The current method is data-driven but does not explicitly incorporate user behavioral hypotheses about travel choices.
  • Integration with Transport System Models (TSM): The authors suggest future work should integrate their method with TSM to better estimate travel demand and understand how user choices might be modified.
  • Intermodal Relationships: Further research is needed to consider the relationships between subway stations and other travel modes, such as buses and shared bicycles.
  • Land Use Characteristics: A deeper study of land use characteristics in predicted areas is recommended.

Summary updated

BibTeX
@article{cai2022novel,
  author = {Zhi Cai and Jiawei Wang and Tong Li and Bowen Yang and Xing Su and Limin Guo and Zhiming Ding},
  title = {{A Novel Trajectory Based Prediction Method for Urban Subway Design}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2022},
  doi = {10.3390/ijgi11020126},
}
Figure