EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Hierarchical Multi-Agent Reinforcement Learning Control for Electric-Vehicle Virtual Power Plants in Balancing Markets

Koshin Hayashi, Sihui Xue, Daisuke Kodaira

IEEE Access · 2026 · 14 · 89505-89517

DOI: 10.1109/access.2026.3698457 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Koshin Hayashi und 2 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: JSPS KAKENHI Grant-in-Aid for Scientific Research.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Balancing markets require flexible resources that can follow dispatch signals within tight response-time budgets. Aggregated electric vehicles operated as a virtual power plant are promising for this purpose, but practical deployment requires control architectures that can accommodate station-level available power, user charging demand, and strict response-time requirements. This paper proposes a hierarchical controller based on multi-agent deep deterministic policy gradient, in which each charging station is modeled as an agent and trained under centralized training and decentralized execution. A case study motivated by Japan’s balancing-market Frequency Restoration Reserve (FRR) product, hereafter referred to as the Secondary 2 (S2) product, is conducted for a virtual power plant with five stations and 50 chargers. In a representative single-training-run simulation, the proposed controller achieves a 97% dispatch tracking rate within the allowable tracking-error band and an 80% target state-of-charge satisfaction rate when tested on held-out dispatch trajectories. Practical latency feasibility is also discussed using a computation-time analysis and measured communication/control-plane latency from a field demonstrator based on a distributed energy resource remote-control platform used in Japan. The results indicate that the proposed architecture is a promising alternative to fully centralized control for latency-constrained electric-vehicle virtual power plants.

Förderung

  • JSPS KAKENHI Grant-in-Aid for Scientific Research

Zitieren

APA

Hayashi, K., Xue, S., & Kodaira, D. (2026). Hierarchical Multi-Agent Reinforcement Learning Control for Electric-Vehicle Virtual Power Plants in Balancing Markets. IEEE Access, 14, 89505-89517. https://doi.org/10.1109/access.2026.3698457

Vancouver

Hayashi K, Xue S, Kodaira D. Hierarchical Multi-Agent Reinforcement Learning Control for Electric-Vehicle Virtual Power Plants in Balancing Markets. IEEE Access. 2026;14:89505-89517. doi:10.1109/access.2026.3698457

BibTeX
@article{Hayashi2026,
  author = {Hayashi, Koshin and Xue, Sihui and Kodaira, Daisuke},
  title = {Hierarchical Multi-Agent Reinforcement Learning Control for Electric-Vehicle Virtual Power Plants in Balancing Markets},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {89505--89517},
  doi = {10.1109/access.2026.3698457}
}
Kostenlos starten →