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

Optimal Power Flow Using Topology‐Aware Transformer Neural Networks

Arundhuti Haldar, Anderson Rodrigo de Queiroz

IET Generation, Transmission & Distribution · 2026-01 · 20 (1)

DOI: 10.1049/gtd2.70486 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Generation, Transmission & Distribution, online erschienen am 10.11.2026. Verfasst von Arundhuti Haldar und 1 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper presents a topology‐aware transformer neural network (TA‐TNN) for fast approximation of the AC optimal power flow (ACOPF) problem. Conventional optimisation methods provide high accuracy but can be computationally expensive for real‐time applications, while existing machine learning models often struggle to balance scalability, topological awareness and interpretability. The proposed TA‐TNN incorporates grid‐specific positional encodings derived from the network admittance matrix, embedding electrical topology directly into the transformer architecture while preserving global dependency modelling among buses. The framework is evaluated on four IEEE test systems and compared with an MLP, a graph convolutional network, a baseline transformer and a transformer using sinusoidal positional encoding. Results show that positional encoding substantially improves transformer‐based ACOPF prediction, with TA‐TNN providing further reductions in prediction error compared to the sinusoidal‐encoding transformer. Although ACOPF constraints are not explicitly enforced during inference, TA‐TNN predictions provide effective initialisations for Newton–Raphson (NR) power flow. Across all test cases, the NR solver achieves convergence and requires up to fewer iterations than a conventional flat start. Thus, TA‐TNN provides accurate ACOPF approximation and an efficient warm start for physics‐based correction, thereby reducing both data‐driven inference and subsequent physics‐based solution effort.

Zitieren

APA

Haldar, A., & Queiroz, A. R. D. (2026). Optimal Power Flow Using Topology‐Aware Transformer Neural Networks. IET Generation, Transmission & Distribution, 20(1). https://doi.org/10.1049/gtd2.70486

Vancouver

Haldar A, Queiroz ARD. Optimal Power Flow Using Topology‐Aware Transformer Neural Networks. IET Generation, Transmission & Distribution. 2026;20(1). doi:10.1049/gtd2.70486

BibTeX
@article{Haldar2026,
  author = {Haldar, Arundhuti and Queiroz, Anderson Rodrigo de},
  title = {Optimal Power Flow Using Topology‐Aware Transformer Neural Networks},
  journal = {IET Generation, Transmission & Distribution},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/gtd2.70486}
}
Kostenlos starten →