Authors
Xiaojuan Chen, Yuanming Liu, Chang Qu, Xue Li
Published in
PloS one. Volume 21. Issue 9. Pages e0354026. Epub Sep 03, 2026.
Abstract
As the nerve center of power systems, power communication networks require risk assessment methods with lightweight architecture and high prediction accuracy to support reliable operation and maintenance decision-making. To overcome the limitations of conventional models-including weak temporal feature extraction, static weight assignment, and poor generalization in small-sample scenarios-this paper proposes a risk assessment method for power communication networks based on the fusion of Lightweight Temporal Attention (LTA) and Elastic Net (EN). First, a risk indicator system is constructed, and redundant features are eliminated through variance-based screening to reduce data dimensionality. The LTA module discards complex multi-head structures and computes dynamic weights solely by combining indicator-risk correlations and normalized interaction terms, enabling adaptive focusing on core time-series indicators. The EN regression is adopted to construct the prediction model, in which bi-regularization balances fitting performance and generalization ability to further improve prediction accuracy. Tests on 12-month small-sample datasets show that the MAE of the LTA-EN model is reduced by 23.4% compared with the conventional fixed-weight linear regression scheme and by 12.1% compared with the complex attention-ridge regression approach. The small-sample generalization error is decreased by 10% on average, and the core indicator recognition efficiency is improved by 40%. The model achieves an optimal tradeoff among small-sample adaptability, lightweight deployment, and high-precision prediction, and can provide efficient quantitative support for monthly risk early warning of power communication networks.
PMID:
42691136
Bibliographic data and abstract were imported from PubMed on 04 Sep 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 2
- Comments 0