当前位置: X-MOL 学术Transportation › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Spatial–temporal multi-task learning for short-term passenger inflow and outflow prediction on holidays in urban rail transit systems
Transportation ( IF 3.5 ) Pub Date : 2025-01-18 , DOI: 10.1007/s11116-025-10583-z
Hao Qiu, Jinlei Zhang, Lixing Yang, Kuo Han, Xiaobao Yang, Ziyou Gao

The rapid growth of passengers has led to overcrowding in urban rail transit (URT) systems, especially during holidays, posing significant challenges to the safe management and operation of URT systems. Accurate and real-time short-term passenger inflow and outflow prediction on holidays is essential for operation management and resource allocation to alleviate such overcrowding. However, short-term passenger inflow and outflow prediction on holidays is a challenging task influenced by various factors, including temporal dependencies, spatial dependencies, the temporal evolution of spatial dependencies, the interaction between inflow and outflow, and the limited holiday samples. To address these challenges, we propose a Spatial–Temporal Multi-Task Learning (STMTL) for short-term passenger inflow and outflow prediction on holidays in URT systems. STMTL comprises three parts: (1) Multi-Graph Channel Attention Network (MGCA) extracts both static and dynamic spatial dependencies from inter-station interaction graphs and then adaptively integrates multi-graph features. (2) Time Encoding-Gated Recurrent Unit (TE-GRU), utilizes time encoding gates to capture long-term periodic variations and unique fluctuations caused by holidays. (3) Cross-attention block (CAB) captures complex interactions during holidays and facilitates the sharing of spatiotemporal features between passenger inflow and outflow. The effectiveness and robustness of STMTL are validated on two real-world datasets from the Nanning URT system in China during the New Year’s Day period. Experimental results demonstrate that STMTL consistently outperforms several classic and state-of-the-art models. STMTL achieves a 3.87% and 3.39% average improvement over the best-performing baseline models at 15-min and 30-min granularities, highlighting its potential for practical applications in URT systems during holidays.

更新日期:2025-01-18
down
wechat
bug