EUVIMEDEUVIMED entdecken
Crossref

Dynamic Prior Knowledge-Assisted Self-Supervised Soft Fault Diagnosis Under Limited Fault Samples in Multi-Span Elastic Optical Networks

Xue Xiao, Chunyu Zhang, Qian Guo, Min Zhang, Danshi Wang

Journal of Lightwave Technology · 2026-11 · 44 (21) · 9480-9493

DOI: 10.1109/jlt.2026.3734749 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Lightwave Technology, 2026-11. Verfasst von Xue Xiao und 4 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Gefördert von: National Natural Science Foundation of China, BUPT Excellent Ph.D. Students Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Timely diagnosis and management of soft faults are essential for maintaining reliable operation in elastic optical networks (EONs). Although data-driven machine learning (ML) methods have demonstrated considerable potential in optical network fault diagnosis, their effectiveness is limited by the scarcity of labeled fault samples, which stems from the subtle and progressive characteristics of soft faults, along with the high cost of expert annotation. While prior knowledge of fault-related spectral characteristics provides valuable diagnostic information, it is often difficult to incorporate directly into data-driven models. To address these challenges, this study introduces PKASS-DCNN, a prior knowledge-assisted self-supervised framework built upon a deep convolutional neural network. The proposed framework extracts interpretable spectral prior features from residual spectra and dynamically fuses them with data-driven general features through an attention mechanism, generating discriminative pseudo-labels for self-supervised pretraining. Extensive numerical results demonstrate that PKASS-DCNN consistently outperforms nine representative diagnostic schemes with limited training samples. With 50 training samples per class, PKASS-DCNN achieves an average macro-F1 score of 0.9937, with false negative and false positive rates of 0.0063 and 0.0013, respectively. Moreover, under extreme label scarcity, it maintains an F1 score of 0.9069 with only 10 labeled samples per class for fine-tuning. Ablation and generalization studies further confirm the contribution of prior knowledge embedding and adaptive feature fusion. This work provides a knowledge-guided and data-efficient solution for reliable soft fault diagnosis in EONs.

Förderung

  • National Natural Science Foundation of China
  • BUPT Excellent Ph.D. Students Foundation

Zitieren

APA

Xiao, X., Zhang, C., Guo, Q., Zhang, M., & Wang, D. (2026). Dynamic Prior Knowledge-Assisted Self-Supervised Soft Fault Diagnosis Under Limited Fault Samples in Multi-Span Elastic Optical Networks. Journal of Lightwave Technology, 44(21), 9480-9493. https://doi.org/10.1109/jlt.2026.3734749

Vancouver

Xiao X, Zhang C, Guo Q, Zhang M, Wang D. Dynamic Prior Knowledge-Assisted Self-Supervised Soft Fault Diagnosis Under Limited Fault Samples in Multi-Span Elastic Optical Networks. Journal of Lightwave Technology. 2026;44(21):9480-9493. doi:10.1109/jlt.2026.3734749

BibTeX
@article{Xiao2026,
  author = {Xiao, Xue and Zhang, Chunyu and Guo, Qian and Zhang, Min and Wang, Danshi},
  title = {Dynamic Prior Knowledge-Assisted Self-Supervised Soft Fault Diagnosis Under Limited Fault Samples in Multi-Span Elastic Optical Networks},
  journal = {Journal of Lightwave Technology},
  year = {2026},
  volume = {44},
  number = {21},
  pages = {9480--9493},
  doi = {10.1109/jlt.2026.3734749}
}
Kostenlos starten →