Low-Voltage Distribution Grid Topology Identification With Latent Tree Model

Distribution grid topology information is the basis of grid analysis functions such as power flow and state estimation. Due to lack of monitoring and measurement devices in low-voltage distribution grids (LVDGs), the intermediate nodes connecting LVDG transformers to end-users cannot upload the nodal operation status. The existence of these latent nodes poses a huge challenge to LVDG topology identification. This paper proposes a LVDG latent node and topology identification method based on end-user data from smart meters. Specifically, a special latent-node embedded Bayesian network, defined as latent tree model, is proposed to provide probabilistic representation for all possible LVDG topologies. A search-based algorithm is proposed to generate candidate topologies and then a Bayesian information criterion is proposed to describe the accuracy of the candidate topologies. Meanwhile, the Expectation-Maximization algorithm is introduced to complement measurement lacking latent nodes. Test results in different simulation scenarios and practical LVDGs demonstrate the effectiveness and robustness of the proposed method.