This paper proposes a probability topology identification framework by combining state estimation (SE) and data-driven methods. The proposed framework aims to obtain probabilistic information about the possible topologies from the real-time measurement data to exclude many low-probability topologies. It avoids the combinatorial explosion caused by too many topology errors in the topology search approach, and it solves the problem of SE-based topology identification when SE is non-observable or non-convergent. The proposed framework is mainly based on the Gaussian mixture model (GMM) to achieve clustering of simulated data with different topologies, so that probabilistic information about their possible topologies can be quickly obtained after collecting real-time measurements. Simulations based on the IEEE 14-bus system show that GMM-based topology clustering achieves better clustering results compared to K-means clustering and can be applied to the distribution network with only voltage measurements and a few phase angle measurements. The proposed probabilistic topology identification framework can provide a prior knowledge of the topology when the original SE is non-observable or provide additional topologies for identification when the SE does not converge. The proposed framework does not change the software architecture of the original SE, which is a beneficial complement to it.

Topology Identification, Gaussian Mixture Model, Power System, Cluster