EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Dynamic reactive power optimization and cooperative control for distribution and transmission power grids based on multi-agent deep reinforcement learning

Weiyi Li, Liping Shen, Yang Cheng, Xu Yang

Frontiers in Physics · 17.06.2026 · 14

DOI: 10.3389/fphy.2026.1833425 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Frontiers in Physics, online erschienen am 17.06.2026. Verfasst von Weiyi Li und 3 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The high time-variability and multi-dimensional coupling characteristics of distribution and transmission power grids impose severe challenges on conventional reactive power optimization methods, which generally fail to balance real-time response performance and global optimality, thereby easily causing voltage limit violations and excessive network power losses. To address this issue, this paper proposes a dynamic reactive power optimization and collaborative control framework based on multi-agent deep reinforcement learning (MADRL). A constrained collaborative optimization model is established with the objectives of minimizing network loss and regulation cost, while satisfying power flow balance, voltage magnitude and line transmission security constraints. The multi-agent interactive learning process is formulated as a Markov game, where an attention communication module is embedded to reduce information interaction overhead, and an adversarial perturbation framework is introduced to enhance operational adaptability under stochastic and extreme grid disturbances. All hyperparameters are explicitly defined with reasonable selection principles and validated via sensitivity analysis to guarantee model reproducibility. Based on the Deep Deterministic Policy Gradient (DDPG) paradigm, the training efficiency and convergence stability are further improved. Comprehensive simulations are carried out on IEEE 33-bus and 69-bus medium-voltage distribution network benchmarks, as well as the IEEE 118-bus high-voltage transmission network benchmark, under diversified scenarios with varying distributed power penetration levels and random load fluctuations. Quantitative statistical comparisons in terms of voltage deviation, convergence performance and operational economy demonstrate that the proposed method outperforms traditional scheduling strategies and existing intelligent optimization algorithms in minimizing voltage deviation, maximizing cumulative rewards, and improving voltage control rates. Quantitative comparison results show that the proposed method reduces the mean voltage deviation by 85.0% compared with the state-of-the-art MADDPG baseline, achieving significantly higher voltage control accuracy than existing mainstream algorithms. The results verify the excellent collaborative coordination capability, anti-disturbance robustness and promising real-time application potential of the developed framework for medium- and large-scale distribution and transmission power grid operation scenarios.

Zitieren

APA

Li, W., Shen, L., Cheng, Y., & Yang, X. (2026). Dynamic reactive power optimization and cooperative control for distribution and transmission power grids based on multi-agent deep reinforcement learning. Frontiers in Physics, 14. https://doi.org/10.3389/fphy.2026.1833425

Vancouver

Li W, Shen L, Cheng Y, Yang X. Dynamic reactive power optimization and cooperative control for distribution and transmission power grids based on multi-agent deep reinforcement learning. Frontiers in Physics. 2026;14. doi:10.3389/fphy.2026.1833425

BibTeX
@article{Li2026,
  author = {Li, Weiyi and Shen, Liping and Cheng, Yang and Yang, Xu},
  title = {Dynamic reactive power optimization and cooperative control for distribution and transmission power grids based on multi-agent deep reinforcement learning},
  journal = {Frontiers in Physics},
  year = {2026},
  volume = {14},
  doi = {10.3389/fphy.2026.1833425}
}
Kostenlos starten →