EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Decentralized Unsupervised Learning Approach for AC-OPF Based on Multi-Branch Deep Neural Network

Wanjun Huang, Kangyu Gong, Zhufeng Jiang, Xinran Zhang, Le Zheng

IEEE Access · 2025 · 13 · 206932-206947

DOI: 10.1109/access.2025.3637542 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2025. Verfasst von Wanjun Huang und 4 weiteren Autor:innen. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Guangdong Basic and Applied Basic Research Foundation, State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The increasing integration of renewable energy sources necessitates efficient and scalable solutions for the AC optimal power flow (AC-OPF) problem. Existing mathematical programming and centralized machine learning approaches often suffer from high computational complexity, limited scalability, and difficulty in maintaining feasibility under complex operating conditions. To address these drawbacks, this study proposes DeepOPF-MB, a novel decentralized AC-OPF solver based on a multi-branch deep neural network (MB-DNN) architecture. To enable decentralized learning, the framework employs the Louvain community detection algorithm to partition the power network into loosely coupled subnetworks. For each subnetwork, a dedicated branch of the MB-DNN is trained to learn the mapping from bus loads to voltages at non-zero injection buses, leveraging a physics-informed unsupervised learning paradigm that incorporates domain-specific constraints. Voltages at zero-injection buses are subsequently reconstructed via Kron reduction. Comprehensive simulations on IEEE 30/118/300-bus systems demonstrate that DeepOPF-MB achieves superior performance to graph neural network-based approach, and centralized supervised and unsupervised learning approaches with comparable neural network size in optimality, feasibility, and computational complexity, and it is robust to the community detection algorithm employed for network partitioning. These results underscore the potential of DeepOPF-MB as a viable and efficient solution for real-time decentralized AC-OPF in large-scale power systems.

Förderung

  • Guangdong Basic and Applied Basic Research Foundation
  • State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources
  • National Natural Science Foundation of China

Zitieren

APA

Huang, W., Gong, K., Jiang, Z., Zhang, X., & Zheng, L. (2025). A Decentralized Unsupervised Learning Approach for AC-OPF Based on Multi-Branch Deep Neural Network. IEEE Access, 13, 206932-206947. https://doi.org/10.1109/access.2025.3637542

Vancouver

Huang W, Gong K, Jiang Z, Zhang X, Zheng L. A Decentralized Unsupervised Learning Approach for AC-OPF Based on Multi-Branch Deep Neural Network. IEEE Access. 2025;13:206932-206947. doi:10.1109/access.2025.3637542

BibTeX
@article{Huang2025,
  author = {Huang, Wanjun and Gong, Kangyu and Jiang, Zhufeng and Zhang, Xinran and Zheng, Le},
  title = {A Decentralized Unsupervised Learning Approach for AC-OPF Based on Multi-Branch Deep Neural Network},
  journal = {IEEE Access},
  year = {2025},
  volume = {13},
  pages = {206932--206947},
  doi = {10.1109/access.2025.3637542}
}
Kostenlos starten →