EUVIMEDEUVIMED entdecken
Crossref

An Urban Traffic Prediction Method With Environmental Anomaly Awareness and Multi-Level Context Learning

Guodong Zhu, Bo Du, Yunyun Niu

IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 24137-24151

DOI: 10.1109/tmc.2026.3731506 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Guodong Zhu und 2 weiteren Autor:innen. Die Arbeit nennt 55 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Urban traffic prediction is an important topic in mobile computing, and traffic data collected by mobile sensors are often affected by environmental factors. However, existing forecasting paradigms usually ignore environmental factors or treat them only as coarse-grained auxiliary information, lacking dedicated modeling of their heterogeneous impacts. Based on stability, environmental factors are categorized into time-sensitive and space-specific factors. Anomalies in time-sensitive factors are prone to causing traffic anomalies, but their impacts are difficult to capture due to sparse samples and the spatiotemporal heterogeneity of their effects on traffic. Space-specific factors are relatively stable, yet their impacts are multi-level, a characteristic often overlooked by existing methods. To address these issues, this paper proposes an Environment-Guided Spatio-Temporal Network (EGSTN). It introduces a Variational Anomaly-Enhanced Module to enhance the learning of evolution patterns of sparse environment-associated traffic anomalies. Meanwhile, a Multi-Level Context Learning Layer is designed to capture the multi-level impacts of space-specific factors while preserving graph topology. Finally, environmental representations are integrated with traffic features to model environment-associated spatiotemporal dependencies. Extensive experiments on six real-world traffic datasets demonstrate that EGSTN outperforms state-of-the-art baselines and exhibits strong robustness under extreme environmental conditions and noise interference.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Zhu, G., Du, B., & Niu, Y. (2026). An Urban Traffic Prediction Method With Environmental Anomaly Awareness and Multi-Level Context Learning. IEEE Transactions on Mobile Computing, 25(12), 24137-24151. https://doi.org/10.1109/tmc.2026.3731506

Vancouver

Zhu G, Du B, Niu Y. An Urban Traffic Prediction Method With Environmental Anomaly Awareness and Multi-Level Context Learning. IEEE Transactions on Mobile Computing. 2026;25(12):24137-24151. doi:10.1109/tmc.2026.3731506

BibTeX
@article{Zhu2026,
  author = {Zhu, Guodong and Du, Bo and Niu, Yunyun},
  title = {An Urban Traffic Prediction Method With Environmental Anomaly Awareness and Multi-Level Context Learning},
  journal = {IEEE Transactions on Mobile Computing},
  year = {2026},
  volume = {25},
  number = {12},
  pages = {24137--24151},
  doi = {10.1109/tmc.2026.3731506}
}
Kostenlos starten →