Imitation‐Initialized Soft‐Masked DRL for Power Network Reconfiguration Against Stochastic Complex Faults
IET Generation, Transmission & Distribution · 2026-01 · 20 (1)
DOI: 10.1049/gtd2.70382 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Generation, Transmission & Distribution, online erschienen am 25.07.2026. Verfasst von Chenhao Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
As a crucial step in power system restoration, rapid power network reconfiguration (PNR) can improve system resilience against extreme events. However, with the increasing penetration of renewable energy and the rising frequency of extreme events, post‐blackout fault scenarios have become stochastic and complex. Consequently, conventional optimization methods fail to provide real‐time PNR decisions. Although deep reinforcement learning (DRL) possesses online decision‐making capabilities, the massive number of stochastic fault scenarios forces DRL into extensive ineffective exploration, thereby degrading training efficiency and solution quality. To address these issues, this paper proposes an imitation‐initialized soft‐masked DRL (ISM‐DRL) method for PNR. First, during the pre‐training stage, imitation learning is employed to initialize the agent using expert strategies derived from typical fault scenarios, effectively improving both training efficiency and solution quality. Second, a soft mask mechanism is introduced during the reinforcement learning stage to guide the agent in achieving an effective balance between exploration and exploitation. Simulation results show that the proposed method achieves average improvements of 2.54% over I‐DRL and 24.78% over SM‐DRL. Furthermore, it maintains an optimality gap of only 3.81% to mixed‐integer‐programming‐based methods in all 1‐line fault scenarios while reducing solution time by 99.88%, thereby enabling flexible and real‐time PNR decision‐making.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Zhang, C., Gu, X., Li, M., & Zhang, J. (2026). Imitation‐Initialized Soft‐Masked DRL for Power Network Reconfiguration Against Stochastic Complex Faults. IET Generation, Transmission & Distribution, 20(1). https://doi.org/10.1049/gtd2.70382
Vancouver
Zhang C, Gu X, Li M, Zhang J. Imitation‐Initialized Soft‐Masked DRL for Power Network Reconfiguration Against Stochastic Complex Faults. IET Generation, Transmission & Distribution. 2026;20(1). doi:10.1049/gtd2.70382
BibTeX
@article{Zhang2026,
author = {Zhang, Chenhao and Gu, Xueping and Li, Meiyi and Zhang, Jiaqi},
title = {Imitation‐Initialized Soft‐Masked DRL for Power Network Reconfiguration Against Stochastic Complex Faults},
journal = {IET Generation, Transmission & Distribution},
year = {2026},
volume = {20},
number = {1},
doi = {10.1049/gtd2.70382}
}
EUVIMED