EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Large language model‐enhanced deep reinforcement learning for voltage control in active distribution networks

Jing Ma, Xian Zhang, Guibin Wang, Yujian Ye, Fushuan Wen, Jing Qiu, Abdelmonem Elsawy Khalil

Energy Conversion and Economics · 2026-10 · 7 (5) · 361-373

DOI: 10.1049/enc2.70051 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Energy Conversion and Economics, online erschienen am 23.10.2026. Verfasst von Jing Ma und 6 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Key Research and Development Program of China, National Natural Science Foundation of China, Basic and Applied Basic Research Foundation of Guangdong Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As active distribution networks (ADNs) become increasingly saturated with intermittent renewable energy sources, maintaining voltage within safe operational limits has become a critical and complex task, requiring advanced control methods beyond traditional rule‐based approaches. This paper proposes a novel voltage control framework that integrates a large language model (LLM) with the soft actor‐critic (SAC) algorithm to enhance the performance of deep reinforcement learning (DRL) in ADNs. The core idea is to leverage the LLM as an external knowledge source during the early stage of training to generate highly interpretable and physically meaningful control action suggestions for key voltage regulation devices. A dynamic action fusion mechanism is designed to combine the LLM‐guided actions with SAC policy outputs. As training progresses, the influence of the LLM gradually diminishes, ultimately yielding a fully autonomous SAC policy. Tests on modified IEEE 33‐ and 123‐bus distribution systems show that the proposed method outperforms the standard SAC in terms of convergence efficiency, voltage deviation reduction and control stability. It effectively mitigates policy oscillations during the early exploration phase and significantly reduces the switching frequency of mechanical devices under the learned policy, thereby contributing to longer equipment service life and higher operational reliability.

Förderung

  • National Key Research and Development Program of China
  • National Natural Science Foundation of China
  • Basic and Applied Basic Research Foundation of Guangdong Province

Zitieren

APA

Ma, J., Zhang, X., Wang, G., Ye, Y., Wen, F., Qiu, J., & Khalil, A. E. (2026). Large language model‐enhanced deep reinforcement learning for voltage control in active distribution networks. Energy Conversion and Economics, 7(5), 361-373. https://doi.org/10.1049/enc2.70051

Vancouver

Ma J, Zhang X, Wang G, Ye Y, Wen F, Qiu J, et al. Large language model‐enhanced deep reinforcement learning for voltage control in active distribution networks. Energy Conversion and Economics. 2026;7(5):361-373. doi:10.1049/enc2.70051

BibTeX
@article{Ma2026,
  author = {Ma, Jing and Zhang, Xian and Wang, Guibin and Ye, Yujian and Wen, Fushuan and Qiu, Jing and Khalil, Abdelmonem Elsawy},
  title = {Large language model‐enhanced deep reinforcement learning for voltage control in active distribution networks},
  journal = {Energy Conversion and Economics},
  year = {2026},
  volume = {7},
  number = {5},
  pages = {361--373},
  doi = {10.1049/enc2.70051}
}
Kostenlos starten →