EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Source-grid-load-storage multi-terminal collaborative control technology for photovoltaic distribution network based on DQN

Weiru Niu, Suyuan Jiao, Zhihao Wang, Lingzhi Li, Tie Cao

International Journal of Renewable Energy Development · 01.11.2026 · 15 (6) · 1429-1441

DOI: 10.61435/ijred.2026.62121 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Renewable Energy Development, online erschienen am 01.11.2026. Verfasst von Weiru Niu und 4 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As the photovoltaic penetration rate continues to increase, the dynamic characteristics of the multi-dimensional coupling of source, grid, load and storage in the distribution network have become increasingly significant. Traditional control methods often face response lag and regulation instability in this multi-time-varying scenario. Therefore, the study proposes a source-grid-load-storage multi-terminal collaborative control strategy for photovoltaic distribution network. A centralized and distributed two-level collaborative architecture is established to enable deep reinforcement learning agents to learn the optimal power regulation strategy. The multi-time scale control of voltage regulator taps, photovoltaic, energy storage systems, and static reactive power compensation devices is coordinated to achieve voltage optimization and efficient energy allocation. A comparative test was conducted using the IEEE 33-node system as the simulation platform. The results showed that the proposed method could increase the peak photovoltaic output by about 20%-25% on the power generation side, and the average power generation utilization rate reached 93%. The load transfer rate increased to 26.2%, and the load balance dropped to 0.21. In terms of voltage stability, the system fluctuation was controlled within 0.15 p.u., and the steady-state error was less than 0.01 p.u. In typical operating scenarios, deep reinforcement learning control increased the energy storage utilization rate to 86.3%-88.9% and reduced the network loss rate to 3.1%-3.5%. The results show that this method is significantly better than the comparison strategy in terms of multi-source collaboration and dynamic stability, and provides an efficient and intelligent optimization solution for the safe and economical operation of high-penetration photovoltaic distribution networks.

Zitieren

APA

Niu, W., Jiao, S., Wang, Z., Li, L., & Cao, T. (2026). Source-grid-load-storage multi-terminal collaborative control technology for photovoltaic distribution network based on DQN. International Journal of Renewable Energy Development, 15(6), 1429-1441. https://doi.org/10.61435/ijred.2026.62121

Vancouver

Niu W, Jiao S, Wang Z, Li L, Cao T. Source-grid-load-storage multi-terminal collaborative control technology for photovoltaic distribution network based on DQN. International Journal of Renewable Energy Development. 2026;15(6):1429-1441. doi:10.61435/ijred.2026.62121

BibTeX
@article{Niu2026,
  author = {Niu, Weiru and Jiao, Suyuan and Wang, Zhihao and Li, Lingzhi and Cao, Tie},
  title = {Source-grid-load-storage multi-terminal collaborative control technology for photovoltaic distribution network based on DQN},
  journal = {International Journal of Renewable Energy Development},
  year = {2026},
  volume = {15},
  number = {6},
  pages = {1429--1441},
  doi = {10.61435/ijred.2026.62121}
}
Kostenlos starten →