.png){kind=icon}
.png){kind=icon}
Return
Regional Bus Travel Time Prediction Using Graph Neural Networks
.png){kind=icon}
PRE
AI
DOI：10.1061/JTEPBS.TEENG-9006.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
An integral component of intelligent public transportation systems is providing accurate travel time information. Predicting bus arrival times in advance reduces passengers' waiting times and encourages public transport usage. However, current methods face two major challenges: the complexity of real-time traffic factors in urban environments and the absence of comprehensive forecasting models for regional bus networks. In this study, we modeled bus arrival time prediction in regional networks as a graph time series forecasting problem. We developed a graph structure that includes both station and interstation road nodes, unlike traditional single-node graphs used in traffic prediction tasks. Additionally, we proposed an algorithm that combines convolutional layers, attention mechanisms, and graph convolutional neural networks to capture dynamic spatiotemporal dependencies and forecast bus arrival times. By leveraging these techniques, the algorithm adjusts the input feature matrix and effectively learns spatiotemporal patterns. In experiments on a bus network with eight routes in Chongqing's Caijia area, the model achieved a mean absolute percentage error of 13.62% and a mean absolute error of 9.53 s, demonstrating its effectiveness and competitive advantage over existing models.
Keywords:
REAL-TIME
Journal
J
IF:
2.1
Papers:
123
Citations:
1.9K