Autonomous Topology Identification in Distribution Networks Using Smart Meter Data

This paper proposes a data-driven and autonomous topology identification framework for distribution networks when topology information is unknown. Using only smart-meter measurements and requiring no prior network information or phasor data, the proposed method identifies the network topology via admittance matrix (Ybus) estimation and a two-stage grouping process. To mitigate the accuracy degradation of the conventional Ybusestimation under partial observability, we develop an advanced Ybusestimation method based on singular value decomposition (SVD). For topology identification, Stage 1 performs preliminary grouping in a fully data-driven manner using the proposed optimal k-means++-based clustering on the estimated admittance matrix. Stage 2 autonomously corrects misgrouping by enforcing physical consistency with the measurements and finalizes the resulting topology. The framework is validated on multiple distribution networks. Furthermore, to verify its real-world applicability, unit tests and a hardware-in-the-loop (HIL) implementation are conducted.