EISTNet: An Efficient Traffic Flow Prediction Method via Explicit–Implicit Spatiotemporal Collaborative Networks
IEEE Sensors Journal · 2026-08 · 26 (15) · 23087-23101
DOI: 10.1109/jsen.2026.3705396 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, 2026-08. Verfasst von Guiyuan Xie und 5 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: Guangdong Provincial Natural Science Foundation General Project, Special Project for Key Fields in Higher Education of Guangdong, Guangzhou Key Field Research and Development Plan Project.
Abstract (Originalsprache)
Traffic flow prediction is a core task in intelligent transportation systems (ITSs), yet it remains challenging due to the intricate coupling of local topological correlations and global macrodynamics. We propose the explicit–implicit spatiotemporal collaborative network (EISTNet), a novel framework that synergistically models these dependencies through a decoupled parallel architecture. The explicit branch utilizes temporal attention and dynamic graph modeling to capture local evolutions, while the implicit branch integrates the Mamba block to model long-range global trends. Extensive experiments demonstrate that the EISTNet delivers competitive accuracy and superior computational efficiency. Notably, analysis on PEMS03 reveals that its extreme spatiotemporal homogeneity—manifested through spatial redundancy and synchronized fluctuations—renders learnable embeddings (LEs) prone to overfitting stochastic noise. EISTNet maintains robustness in such scenarios by effectively decoupling global stability from local nuances, allowing each branch to fulfill its respective functional role where serial or individual architectures often fail. Furthermore, t-SNE visualizations on PEMS04 confirm that the fused features of EISTNet generate more discriminative and well-clustered representations compared to serial variants, demonstrating its superior capability in distinguishing heterogeneous traffic patterns. Overall, EISTNet provides a scalable and resilient solution for high-resolution traffic forecasting. The code is available at https://github.com/Kivly/EISTNet
Förderung
- Guangdong Provincial Natural Science Foundation General Project
- Special Project for Key Fields in Higher Education of Guangdong
- Guangzhou Key Field Research and Development Plan Project
- Guangdong Provincial Key Laboratory of Intellectual Property and Big Data
- Key Discipline Research Capacity Improvement Project of Guangdong Province
Zitieren
APA
Xie, G., Qiu, J., Yi, C., Zhan, J., Hu, J., & Liu, Y. (2026). EISTNet: An Efficient Traffic Flow Prediction Method via Explicit–Implicit Spatiotemporal Collaborative Networks. IEEE Sensors Journal, 26(15), 23087-23101. https://doi.org/10.1109/jsen.2026.3705396
Vancouver
Xie G, Qiu J, Yi C, Zhan J, Hu J, Liu Y. EISTNet: An Efficient Traffic Flow Prediction Method via Explicit–Implicit Spatiotemporal Collaborative Networks. IEEE Sensors Journal. 2026;26(15):23087-23101. doi:10.1109/jsen.2026.3705396
BibTeX
@article{Xie2026,
author = {Xie, Guiyuan and Qiu, Jiacheng and Yi, Chensheng and Zhan, Jin and Hu, Jiayu and Liu, Yu},
title = {EISTNet: An Efficient Traffic Flow Prediction Method via Explicit–Implicit Spatiotemporal Collaborative Networks},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {15},
pages = {23087--23101},
doi = {10.1109/jsen.2026.3705396}
}
EUVIMED