EUVIMEDEUVIMED entdecken
Crossref

Frequency-Aware Multi-Task Forecasting for Integrated Energy Systems via Variational Mode Decomposition and Convolution-Attention Encoding

Yuchao Gao, Jinran Wu, Yang Yang, Zijin Wang, Zhe Ding

IEEE Transactions on Smart Grid · 2026-01 · 17 (1) · 818-831

DOI: 10.1109/tsg.2025.3621486 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Smart Grid, 2026-01. Verfasst von Yuchao Gao und 4 weiteren Autor:innen. Die Arbeit wurde bisher 13-mal zitiert (Crossref) und nennt 39 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Nanjing University of Posts and Telecommunications, Huali Program for Excellent Talents in Nanjing University of Posts and Telecommunications.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Integrated Energy Systems (IESs) integrate electricity, heating, cooling, and gas through coupled infrastructures, posing forecasting challenges due to heterogeneous temporal patterns and cross-energy interference. These heterogeneous temporal patterns often reflect distinct frequency characteristics, which existing multi-task approaches struggle to disentangle due to the lack of frequency-aware task decoupling. To address this, this paper proposes VATLNet, a unified multi-task forecasting framework that integrates frequency-aware task decoupling with hybrid convolution-attention feature extraction. Specifically, multivariate energy inputs are decomposed via variational mode decomposition, and the resulting components are clustered into frequency-similar groups based on sample entropy. A shared encoder combining temporal convolutional networks and multi-head attention extracts frequency features across all clusters, which are then independently decoded by parallel long short-term memory network branches. Extensive experiments on real-world IES load datasets demonstrate that VATLNet outperforms state-of-the-art baselines, achieving 2.44% and 1.03% MAPE in single- and multi-task forecasting, respectively. The framework improves cross-frequency-aware feature extraction, reduces inter-task interference, and enhances forecasting robustness and interpretability across diverse energy modalities.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Nanjing University of Posts and Telecommunications
  • Huali Program for Excellent Talents in Nanjing University of Posts and Telecommunications
  • Postgraduate Research and Practice Innovation Program of Jiangsu Province

Zitieren

APA

Gao, Y., Wu, J., Yang, Y., Wang, Z., & Ding, Z. (2026). Frequency-Aware Multi-Task Forecasting for Integrated Energy Systems via Variational Mode Decomposition and Convolution-Attention Encoding. IEEE Transactions on Smart Grid, 17(1), 818-831. https://doi.org/10.1109/tsg.2025.3621486

Vancouver

Gao Y, Wu J, Yang Y, Wang Z, Ding Z. Frequency-Aware Multi-Task Forecasting for Integrated Energy Systems via Variational Mode Decomposition and Convolution-Attention Encoding. IEEE Transactions on Smart Grid. 2026;17(1):818-831. doi:10.1109/tsg.2025.3621486

BibTeX
@article{Gao2026,
  author = {Gao, Yuchao and Wu, Jinran and Yang, Yang and Wang, Zijin and Ding, Zhe},
  title = {Frequency-Aware Multi-Task Forecasting for Integrated Energy Systems via Variational Mode Decomposition and Convolution-Attention Encoding},
  journal = {IEEE Transactions on Smart Grid},
  year = {2026},
  volume = {17},
  number = {1},
  pages = {818--831},
  doi = {10.1109/tsg.2025.3621486}
}
Kostenlos starten →