Predicting travel time of bus journeys with alternative bus services

• Main Authors: He, Peilan, Sun, Yidan, Jiang, Guiyuan, Lam, Siew-Kei
• Other Authors: School of Computer Science and Engineering
• Format: Conference or Workshop Item
• Language: English
• Published: 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Predictive Models; Public Transportation
• Online Access: https://hdl.handle.net/10356/147528
• Institution: Nanyang Technological University

Accurate travel time prediction of public transport services is essential for reliable journey planning. Existing methods for journey time prediction typically assume a fixed journey route with predefined bus services. However, there usually exist multiple alternative bus services that can serve the same journey route (or a segment of the route); thus the passengers could dynamically decide which bus service to take based on the dynamic bus arrivals. In this paper, we address the problem of travel time prediction of bus journeys with multiple alternative bus services (TP-BJMAS). We propose a novel framework to solve the TP-BJMAS problem by partitioning the journey route into several route segments based on the transfer points, such that each segment can be served by multiple bus services. The travel time of each segment is estimated using a segment prediction module based on neural network technique and the total journey time is obtained by aggregating the travel time of all segments. In the segment prediction module, the travel time using a specified bus service is obtained based on pre-trained riding time prediction model and waiting time prediction model. Since each route segment can be served by multiple alternative bus services, multiple estimations of segment travel time (ESTT) are calculated (each based on one bus service). The attention technique is utilized to fuse the ESTTs of all bus service considering the heterogeneous importance of different ESTTs. The effectiveness is evaluated using large scale real-world public transport networks and traffic data involving more than 30 bus services.