.png){kind=icon}
.png){kind=icon}
Return
Adaptive Physics-Informed Data-Driven Topology Identification for Distribution Networks
.png){kind=icon}
PRE
AI
M
Z
J
W
Y
X
Y
Z
Y
DOI：10.1109/tii.2026.3684153.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate and real-time topology identification is essential for the reliable operation of power distribution networks (PDNs). The increasing complexity and dynamism of modern PDNs pose significant challenges for real-time topology identification, especially with limited synchronized measurement devices. While deep learning methods show promise for topology inference, the lack of physical awareness would lead to unreliable predictions. In this article, we propose an adaptive physics-informed learning model for topology identification in PDNs, where a physics-informed loss function derived from linear coupled power flow equations is introduced to improve identification accuracy and ensure physical consistency. An uncertainty-based weighting strategy is used to dynamically balance the contributions of supervised learning and physical constraints. To support this learning objective, the model employs a multiscale convolutional neural network enhanced with a channel attention mechanism to extract diverse features under limited deployment of synchronized measurement devices. The proposed method jointly considers both branch-level and topology-level inference accuracy, supports real-time application, and does not rely on radial topology assumptions. To validate the proposed method, experiments are conducted on the 33-node, 69-node, and 118-node distribution systems, and the simulation results demonstrate its effectiveness at both the branch and topology levels.
Keywords:
Data-driven
distribution networks
neural network
physics-informed
topology identification
Journal
IF:
9.9
Papers:
8.3K
Citations:
6.0W
Organization
Cited Papers