EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Split Federated Learning With Behavioral Digital Twins and Multi-Head Temporal Networks for Internet of Vehicles Security

Farhan Ullah, Khalid Mahmood, Umar Raza, Jawad Ahmad, Jenhui Chen, Rubal Jeet, Yue Zhao

IEEE Open Journal of Vehicular Technology · 2026 · 7 · 1614-1626

DOI: 10.1109/ojvt.2026.3694907 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Open Journal of Vehicular Technology, 2026. Verfasst von Farhan Ullah und 6 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science and Technology Council, Taiwan.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid growth of the Internet of Vehicles (IoV) ecosystem has produced complicated security concerns that require a modern Intrusion Detection System (IDS) that can respond to emerging threats and maintain data protection. The dynamic behavior of clients and the heterogeneity of the system can affect the convergence and performance of models in federated learning for IoV security. Traditional federated learning may overlook data quality, computational resources, and client confidence. This issue is resolved by behavioral digital twin technology, which generates virtual replicas of real-world IoV clients that continuously monitor, analyze, and forecast client behavior. This paper proposes a novel approach, DT-ASFL, an adaptive Split Federated Learning (SFL) system that combines behavioral digital twins with multi-head temporal networks to improve IoV security. The lightweight multi-head Temporal Convolutional Network (TCN) extracts temporal features, and SFL stores sensitive vehicle data on edge devices. These attributes are analyzed by a more advanced classifier on a distant server, so raw data never leaves the device. For each client, behavioral digital twins dynamically modify aggregation weights based on trust and health measures, evaluate data quality, and monitor performance. This adaptive method provides protection against malicious users and system heterogeneity. TCN architecture captures diverse temporal patterns for optimal feature fusion using parallel processing heads with different dilation rates and an attention mechanism. Evaluation on several security datasets demonstrates considerable increases in attack detection accuracy and robustness. The approach ensures privacy and system flexibility while outperforming conventional federated learning techniques.

Förderung

  • National Science and Technology Council, Taiwan

Zitieren

APA

Ullah, F., Mahmood, K., Raza, U., Ahmad, J., Chen, J., Jeet, R., & Zhao, Y. (2026). Adaptive Split Federated Learning With Behavioral Digital Twins and Multi-Head Temporal Networks for Internet of Vehicles Security. IEEE Open Journal of Vehicular Technology, 7, 1614-1626. https://doi.org/10.1109/ojvt.2026.3694907

Vancouver

Ullah F, Mahmood K, Raza U, Ahmad J, Chen J, Jeet R, et al. Adaptive Split Federated Learning With Behavioral Digital Twins and Multi-Head Temporal Networks for Internet of Vehicles Security. IEEE Open Journal of Vehicular Technology. 2026;7:1614-1626. doi:10.1109/ojvt.2026.3694907

BibTeX
@article{Ullah2026,
  author = {Ullah, Farhan and Mahmood, Khalid and Raza, Umar and Ahmad, Jawad and Chen, Jenhui and Jeet, Rubal and Zhao, Yue},
  title = {Adaptive Split Federated Learning With Behavioral Digital Twins and Multi-Head Temporal Networks for Internet of Vehicles Security},
  journal = {IEEE Open Journal of Vehicular Technology},
  year = {2026},
  volume = {7},
  pages = {1614--1626},
  doi = {10.1109/ojvt.2026.3694907}
}
Kostenlos starten →