Predicting the delay time of trains is an important task in intelligent transport systems, as an accurate prediction can provide a reliable reference for passengers and dispatchers of the railway system. However, due to the complexity of the railway system, interactions of various spatio-temporal variables make it difficult to find the rules of delay propagation. We introduce a Sequential Precoding Spatial-Temporal Network (SPSTN) model to predict the delay of trains. SPSTN consists of a Transformer encoder that captures long-term dependencies in time series, and spatio-temporal graph convolution blocks that model delay propagation at both temporal and spatial levels. Experiments on a subset of the British railway network show that SPSTN performs favorably against the state-of-the-art, which verifies that the combination of sequential precoding and spatio-temporal convolution can effectively model delay propagation on railway networks.
CITATION STYLE
Fu, J., Zhong, L., Li, C., Li, H., Kong, C., & Shao, J. (2023). SPSTN: Sequential Precoding Spatial-Temporal Networks for Railway Delay Prediction. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 13421 LNCS, pp. 451–458). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-25158-0_37
Mendeley helps you to discover research relevant for your work.